Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Application of the novel fractional grey model FAGMO(1,1,k) to predict China's nuclear energy consumption

Application of the novel fractional grey model FAGMO(1,1,k) to predict China's nuclear energy... At present, the energy structure of China is shifting towards cleaner and lower amounts of carbon fuel, driven by environmental needs and technological ad- vances. Nuclear energy, which is one of the major low-carbon resources, plays a key role in China’s clean energy development. To formulate appropriate energy policies, it is necessary to conduct reliable forecasts. This paper discusses the nuclear energy consumption of China by means of a novel fractional grey model FAGMO(1,1,k). The fractional accumulated generating matrix is introduced to analyse the fractional grey model properties. Thereafter, the modelling proce- dures of the FAGMO(1,1,k) are presented in detail, along with the transforms of its optimal parameters. A stochastic testing scheme is provided to validate the accuracy and properties of the optimal parameters of the FAGMO(1,1,k). Finally, this model is used to forecast China’s nuclear energy consumption and the results demonstrate that the FAGMO(1,1,k) model provides accurate pre- diction, outperforming other grey models. Keywords: Nuclear energy consumption, Grey system, Fractional order https://doi.org/10.1016/j.energy.2018.09.155 Corresponding author: Xin Ma Email address: cauchy7203@gmail.com (Xin Ma ) arXiv:1809.10882v1 [stat.AP] 28 Sep 2018 accumulation, Optimised parameter, Energy forecasting 1. Introduction Energy is the most important strategic resource and provides a key material basis for economic development and social progress. Energy consumption pre- diction constitutes an important aspect of energy policies for countries globally, particularly developing countries such as China, where the energy consumption structure is changing at a rapid speed. Numerous models have been intro- duced for forecasting energy consumption, such as dynamic causality analysis [1], nonlinear and asymmetric analysis [2], time-series analysis [3, 4], machine learning models [5], the coupling mathematical model [6, 7, 8], autoregressive distributed lag model [9], hybrid forecasting system [10, 11], machining system [12], fuzzy systems [13], LEAP model [14, 15], TIMES model [16, 17], NEMS model [18, 19] and grey model [20, 21, 22, 23, 24, 25, 26, 27, 28]. Among these prevalent methods, simple linear regression, multivariate linear regression, and time-series analysis are often significant in accurately demonstrating the phe- nomena of long-term trends. However, these exhibit the limitations of requiring a large amount of observed data, at least 50 or more sets, to construct mod- els. The computational intelligence method requires a substantial amount of training data to derive the optimised parameters. However, in many practical situations, it is very difficult and sometimes even impossible to obtain com- plete information. Therefore, it is important to identify a favourable method for forecasting the trend of an analysed system using scarce information with less errors. The grey forecasting theory, proposed by Professor Deng [29], offers a fea- sible and efficient method for dealing with uncertain problems containing poor information. The main advantage of this theory is that only four or more sam- ples are required to describe the behaviour and evolution of the analysed system. In Deng’s pioneering work, the first-order one variable grey model GM(1,1) was discussed in detail. Over three decades of development, the classical contin- 2 uous GM(1,1) model has been studied extensively; for example, by Xie et al. [30, 31, 32], Wang et al. [33, 34, 35], Ma et al. [36, 37, 38, 39] and others. However, we note that these generalised grey models all include integer-order accumulation, which results in less flexibility in time-series forecasting. Thus, the fractional-order accumulation grey model is considered in this paper. By extending the integer accumulated generating operation into the frac- tional accumulated generating operation, Wu et al. [40] first proposed the frac- tional accumulation GM(1,1) model known as the FAGM(1,1) model. The com- putational results demonstrated that the novel model outperformed the conven- tional GM(1,1) model. Later, Wu and his peers successfully applied fractional accumulation to the fuel production of China [41], tourism demand [42] and electricity consumption [43]. Subsequently, Xiao et al. [44] studied the GM(1,1) model, in which they regarded the fractional accumulated generator matrix as a type of generalised accumulated generating operation. Gao et al. [45] presented a new discrete fractional accumulation GM(1,1) model known as FAGM(1,1,D) and applied it to China’s CO emissions. Mao et al. [46] investigated a novel fractional grey model FGM(q,1). Interested readers can refer to [47, 48, 49, 50] for further details on fractional accumulation grey models. A further significant issue in grey system theory is that the solution applied for prediction does not match the grey difference equation. In 2009, Kong and Wei [51] proposed a parameter optimisation technique to study the DGM(2,1) model. Later, Chen et al. used a similar technique to improve the GM(1,1) [52] and ONGM(1,1) models [53], in which the basic structure of the original models remain in the optimised ones. Recently, Ma and Liu [54] studied the exact non- homogeneous grey prediction model (ENGM) with an exact basic equation and background value. Thereafter, Ma and Liu [55] considered the GMC(1,n) model with optimised parameters and applied it to forecasting the urban consumption per capita and industrial power consumption of China. Following the concept of fractional accumulation and the parameter optimisation method, we propose a novel FAGMO(1,1,k) model. In this paper, we study the nuclear energy consumption of China by means 3 of the FAGMO(1,1,k) model. The computational results indicate that the proposed grey model outperforms the existing ENGM model, optimised non- homogeneous grey model abbreviated as the ONGM(1,1,k) model, FAGM(1,1) model and FAGM(1,1,k) model. The main contributions of our paper are listed below. 1) A fractional accumulation grey model with optimised parameters is developed. 2) Detailed properties of optimised parameters are studied ac- cording to two theorems. These indicate that the first parameter is the most important factor affecting the accuracy of the FAGM(1,1,k) model. 3) Simula- tion results and two practical cases are considered to assess the effectiveness of the FAGMO(1,1,k) model compared to other models. 4) The FAGMO(1,1,k) grey forecasting model is implemented to forecast the nuclear energy consump- tion of China. It is demonstrated in the results that the newly proposed model offers higher precision than other grey models. The remainder of this paper is organised as follows. Section 2 provides a compendium of China’s energy consumption. Section 3 discusses several prelim- inaries. A detailed discussion of the FAGM(1,1,k) model is provided in section 4. Section 5 discusses the optimised parameters. Modelling evaluation criteria and detailed steps are provided in section 6. Section 7 discusses the valida- tion of the FAGMO(1,1,k) model. Applications are explained in section 8 and conclusions are drawn in the final section. 2. Brief overview of China’s energy consumption This section presents a systematic and comprehensive investigation of China’s energy consumption using five fuels, namely coal, oil, natural gas, nuclear energy and renewables. In China, renewables include hydroelectricity, wind, solar, geothermal, biomass and others. According to the statistical data of British Petroleum (BP) Statistical Review of World Energy 2018 (www.bp.com/ statisticalreview), the International Energy Agency (IEA) World Energy Out- look 2017 (www.iea.org/weo2017), Asia-Pacific Economic Cooperation (APEC) Energy Overview 2017 (www.apec.org/Publications), and National Bureau of 4 Statistics of China (NBS) China Statistical Yearbook 2017 (www.stats.gov.cn/tjsj/ndsj), China’s primary energy consumption increased from 142.9 million tonnes oil equivalent (Mtoe) in the first year of the third Five-Year Plan (1966 to 1970) th to 3132.2 Mtoe in the second year of the 13 Five-Year Plan (2016 to 2020), and increased dramatically since the turn of the millennium owing to continu- ous economic growth. According to the statistical data of BP, China’s primary energy consumption from 1966 to 2017 is plotted in Fig. 1. 1000 the third Five-Year Plan (1966-1970) Figure 1: Total primary energy consumption of China from 1966 to 2017 It is well known that China is the world’s largest energy consumer, account- ing for 23% and 23.2% of the global energy consumption in 2016 and 2017, respectively. While coal remains the dominant fuel, its share of total energy th consumption was 62% in 2016 and 60.4% in 2017. China’s 13 Five-Year Plan set an ambitious target for adjusting the primary energy consumption structure. th The energy plan set by China for the 13 Five-Year Plan can met the adjust- ment target of the primary energy consumption structure. A brief overview of China’s primary energy consumption from the perspective of five fuel types is provided below. total primary energy consumption, Mtoe 2.1. Coal Since the foundation of the People’s Republic of China, coal has always been the primary energy fuel, owing to abundant domestic reserves and its low cost [15]. From Figs. 2 and 3, it can be observed that coal soared from 122.4 Mtoe in 1966 to 1892.6 Mtoe in 2017, although the percentage of coal in the total primary energy consumption decreased from 85.7% in 1966 to 60.4% in 2017. Specifically, despite a continuous increase in coal consumption during the third and fourth Five-Year Plan periods, the proportion of coal in the total primary energy consumption has decreased from 85.7% to 72.5%. Following China’s Reform and Opening-Up Policy in 1978, the coal consumption has expanded rapidly from 282.8 Mtoe in 1978 to 1685.8 Mtoe in 2010, while the share of coal in the total primary energy consumption is around at 73.8%. At the beginning th of the 12 Five-Year Plan (2011 to 2015), the Chinese government has been stepping up its efforts to reduce coal consumption to deal with air pollution and climate change. The “supply-side reform” removes unnecessary and out-dated production capacity to avoid supply overcapacity in the coal mining industry. During this period, the consumption of coal decreased from 1903.9 Mtoe in 2011 to 1892.6 Mtoe in 2017, while the share of coal decreased from 70.8% to 60.4%. 2.2. Oil Oil is a major component of primary energy resources globally and plays a strategic role in economic growth. It can be observed in Figs. 2 and 3 that the oil consumption in China increased from 14.3 Mtoe in 1966 to 608.4 Mtoe in 2017, with an average annual growth rate of 7.5%. Owing to China’s oil reserves accounting for only 2% of the global amount, China is highly dependent on overseas oil imports of more than 60%. China became a net importer of crude oil in 1993 and the world’s second largest oil consumer in 2002 [15]. In April 2015, China surpassed the US as the world’s largest oil importer, with imports of 7.4 million barrels per day (Mbbl/D), thereby exceeding the US imports of th 7.2 Mbbl/D. Following the 11 Five-Year Plan, the oil consumption increased rapidly owing to economic growth and the improved quality of life. 6 coal oil natural gas nuclear renewables 1000 the third Five-Year Plan (1966-1970) Figure 2: China’s primary energy consumption under fuel types 0.9 0.8 0.7 0.6 0.5 coal 0.4 oil natural gas 0.3 nuclear renewables 0.2 0.1 Figure 3: Percentage of China’s primary energy consumption under fuel types different fuel types of energy consumption,Mtoe percentage of different fuel types 2.3. Natural gas Natural gas is a fossil fuel for electricity generation, chemical feedstock, heating and cooking, among others. Chinese organisations have estimated that the technically and ultimately recoverable resources of natural gas are 6.1 trillion cubic meters (tcm) and 37 tcm [56], respectively. However, natural gas has not become a major energy resource in China because the domestic natural gas industry has developed slowly. In recent years, the Chinese government has set the stable natural gas supply as one of the country’s energy strategies and encourages gas transportation from areas with significant resources to East China. The National Development and Reform Commission constructed three west-east gas pipelines in 2004, 2007 and 2015, respectively. Furthermore, the “shifting from coal to gas” policy has a significant impact on the natural gas market. The natural gas consumption has increased from 116.2 Mtoe in 2011 to 206.7 Mtoe in 2017, with an average annual growth rate of 8.6%. 2.4. Nuclear energy Nuclear energy is almost always used to generate electricity. To reduce the air pollution from coal-fired power plants, nuclear energy is an inevitable strategic option for China. In fact, China began to develop nuclear energy in the 1980s and the Qinshan Nuclear Power Plant began operating in 1991. In 2012, the State Council set a goal of 58 GW nuclear capacity by 2020. At th the beginning of the 13 Five-Year Plan, 38 nuclear power reactors were in operation with a production of 213.3 TWh, while 19 nuclear power reactors were under construction. At present, the Chinese government focuses on fourth- th generation reactors with increased safety. From the 11 Five-Year Plan (2006 to2010), nuclear energy consumption has soared rapidly from 12.4 Mtoe in 2006 to 56.2 Mtoe in 2017, with an average annual growth rate of 13.4%. 2.5. Renewables China’s renewable energy has been expanding rapidly in recent decades, owing to the development of the modern renewable energy industry. In 2017, 8 China’s renewables consumption accounted for 21.9% of the total global amount, increasing by 31% and accounting for 36% of the global renewables consumption growth. Meanwhile, the renewables consumption increased from 101.1 Mtoe in 2006 to 368.3 Mtoe in 2017; the share has increased from 5.1% to 11.8% with an th average annual growth rate of 11.4%. The 13 Five-Year Plan set targets for an installed wind power generation capacity of 250 GW, solar power generation capacity of 110 GW, and hydropower generation capacity of 350 GW by 2020. In summary, China’s primary energy consumption using five fuels for the period of 1966 to 2017 can be provided below. The coal consumption has grad- ually declined, the oil consumption has gradually increased, and the natural gas, nuclear energy and renewables have rapidly increased. China’s primary energy consumption structure exhibits a diversified trend, and the clean energy has increased yearly. 3. Definitions and properties of fractional accumulation This section provides the fractional accumulated generating operation (AGO), which can reduce the randomness of raw data in grey theory. Correspondingly, the inverse operation of accumulated generation is known as the inverse accumu- th th lated generating operation (IAGO). The r AGO and r IAGO are provided below, which can be found in paper [40, 46]. (0) (0) (0) (0) Definition 1. Let X = x (1), x (2), . . . , x (n) be an original sequence (r) th and X (r > 0) be the r accumulated generating operation (r-AGO) sequence (0) (r) (r−1) r of X , where x (k) = x (i), k = 1, 2, . . . , n. Denote by A the r-AGO i=1 (r) (0) r matrix that satisfies X = X A , and   r r r r [ ] [ ] [ ] ··· 0 1 2 n−1     r r r  0 [ ] [ ] ···  0 1 n−2     r r r A =  0 0 [ ] ···  , 0 n−3    . . . .  . . . . .   . . . .   0 0 0 ··· [ ] n×n r(r+1)···(r+i−1) r+i−1 (r+i−1)! 0 r 0 0 with [ ] = = = , = 0, = = 1. i i 0 i! i i!(r−1)! 0 (1) (0) Obviously, the 1-AGO sequence x (k) = x (i), k = 1, 2, . . . , n, namely i=1   1 1 1 ··· 1     0 1 1 ··· 1       (1) (0) X = X A with A = 0 0 1 ··· 1 .     . . . .  .  . . . . .  .  . . . .   0 0 0 ··· 1 n×n (r−1) Definition 2. The inverse accumulated generation is defined as x (k) = (r) (r) r th x (k) − x (k − 1), k = 1, 2, . . . , n. Denote by D the r inverse accumulated (0) (r) r generating operation (r-IAGO) matrix, which satisfies X = X D , and   −r −r −r −r ··· 0 1 2 n−1     −r −r −r  0 ···  0 1 n−2     −r −r D =  0 0 ···  , 0 n−3    . . . .  . . . . .   . . . .   −r 0 0 0 ··· n×n −r −r(−r+1)···(−r+i−1) i r(r−1)···(r−i+1) i r −r with = =(−1) =(−1) , =0, i> r. i i i! i! i (0) (1) (1) Similarly, the 1-IAGO sequence x (k) = x (k)−x (k−1), k = 1, 2, . . . , n;   1 −1 0 ··· 0     0 1 −1 ··· 0       (0) (1) that is, X = X D, with D = 0 0 1 ··· 0   .    . . . .  . . . . .   . . . .   0 0 0 ··· 1 n×n r + + Theorem 1. The expression [ ] , r ∈ R , i ∈ N is a function of r and i; for any value i, r ∈ (0, 1) , [ ] is a monotonically decreasing function of i; r = 1, [ ] ≡ 1; and r ∈ (1, +∞) , [ ] is a monotonically increasing function of i. 10 Proof 1. We consider the difference (r + i − 1)! (r + i − 2)! r r r+i−1 r+i−2 [ ] − = − = − i i−1 i i−1 i! (r − 1)! (i − 1)! (r − 1)! (r + i − 2)! r + i − 1 = − 1 (i − 1)! (r − 1)! i (r + i − 2)! r − 1 (i − 1)! (r − 1)! i (r + i − 2)! = (r − 1). i! (r − 1)! From the difference results, we complete the proof. To gain an improved understanding of Theorem 1, two figures are displayed in the following Fig. 4 r=1.00 10 r=1.35 r=1.00 0.9 r=1.45 r=0.35 9 r=1.55 0.8 r=0.45 r=1.65 r=0.55 r=1.75 0.7 r=0.65 r=0.75 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 i i Figure 4: Function versus values r and i: left r ∈ (0, 1), right r ∈ (1, +∞) (r) (0) r It follows from X = X A that k k−1 X X (r) r (0) r (0) x (k) = x (i) = [ ]x (k − i), (1) k−i i i=1 i=0 (r) (0) which means that x (k) is the weight of x (i), i = 1, 2, . . . , k. From Theorem 1, when r ∈ (0, 1), the weight of the old data is smaller than that of the new data. When r = 1, the weights of the old and new data are all 1. When r ∈ (1, +∞), the weight of the old data is larger than that of the new data. the values of [ ] the values of [ ] i −1 r r r r Theorem 2. The values of r-AGO A and r-IAGO D satisfy (A ) = D . Proof 2. From the definition of A , it is easy to calculate the determinant r r det (A ) = 1, which means that A is reversible. Employing mathematical induction, when r = 1, we obtain    1 1 1 ··· 1 1 −1 0 ··· 0       0 1 1 ··· 1 0 1 −1 ··· 0          AD =  0 0 1 ··· 1  0 0 1 ··· 0  = I.     . . . .  . . . .  . . . . . . . . . . . .    . . . . . . . . . .    0 0 0 ··· 1 0 0 0 ··· 1 Assuming that the properties hold true when r = m, this means that    −m −m −m −m m m m m [ ] [ ] [ ] ··· ··· 0 1 2 n−1 0 1 2 n−1       −m −m −m m m m 0 [ ] [ ] ··· 0 ···    0 1 n−2 0 1 n−2       −m −m m m 0 0 [ ] ··· 0 0 ··· =I.    0 n−3 0 n−3     . . . .  . . . .  . . . . . . . . . . . .    . . . . . . . . . .    −m 0 0 0 ··· [ ] 0 0 0 ··· 0 0 Then, when r = m + 1, we obtain m+1 m+1 m m m m m m A D = A (AD)D = A ID = A D = I, −1 r r so the result (A ) = D is proven. 4. Fractional grey FAGM(1,1,k) model Definition 3. The first-order differential equation (r) dx (t) (r) + ax (t) = bt + c, r > 0 (2) dt is known as the whitening differential equation of the FAGM(1,1,k) model. The parameter a is a development coefficient, while bt+c is the grey action quantity. The discrete differential equation 2k − 1 (r−1) (r) x (k) + az (k) = b + c (3) 12 (r−1) (r) is referred to as the basic equation of the FAGM(1,1,k). x (k) = x (k) − (r) (r) (r) (r) x (k − 1), z (k) = 0.5 x (k − 1) + x (k) . The least-squares estimation for φ = (a, b, c) of the FAGM(1,1,k) model satisfies −1 T T φ = B B B Y, (4) where     (r) (r−1) −z (2) 1 x (2)         (r) (r−1) −z (3) 1 x (3)         B = , Y = ,  . . .   .  . . . .     . . . .     (r) 2ν−1 (r−1) −z (ν) 1 x (ν) in which ν is the number of samples used to construct the model. Theorem 3. The time response function of the FAGM(1,1,k) model is b b c b b c (r) (0) −a(k−1) xˆ (k) = x (1) − + − e + k − + , k = 2, 3, . . . , n, (5) 2 2 a a a a a a (0) and the restored value of xˆ (k) k = 2, 3, . . . , n can be expressed by (0) (r) r ˆ ˆ X = X D . (6) Proof 3. From Eq. (2), we have (r) dx (t) (r) = −ax (t) + bt + c. (7) dt (r) Let u(t) = −ax (t) + bt + c; then, Eq. (7) is transformed into (r) du(t) dx (t) = −a + b = −au(t) + b. (8) dt dt To perform the indefinite integral on Eq. (8) and reduce it, we obtain (r) −at κ −a −ax (t) + bt + c + b = e e , (9) where e is a constant to be determined. (r) (0) Substituting t = 1 and x (t) = x (1) into Eq. (9), we obtain t=1 κ a 2 (0) e = e a x (1) − ab − ac + b . (10) It follows from Eqs. (9) and (10) that b b c b b c (r) (0) −a(t−1) x (t) = x (1) − + − e + t − + . (11) 2 2 a a a a a a Thus, the time response function of the FAGM(1,1,k) model is b b c b b c (r) (0) −a(k−1) xˆ (k) = x (1) − + − e + k − + , k = 2, 3, . . . , n, 2 2 a a a a a a (0) and the restored value of xˆ (k) k = 2, 3, . . . , n can be expressed by (0) (0) (0) (r) (r) (r) r xˆ (1), xˆ (2), . . . , xˆ (n) = xˆ (1), xˆ (2), . . . , xˆ (n) D . Setting b = 0 in Eq. (2), the fractional FAGM(1,1,k) model is reduced to the fractional FAGM(1,1) model [40] with the form (r) dx (t) (r) + aX (t) = c. (12) dt Setting r = 1 in Eq. (2), the fractional FAGM(1,1,k) model is reduced to the GM(1,1,k,c) model [53] with the form (1) dx (t) (1) + ax (t) = bt + c. (13) dt Setting r = 1, c = 0 in Eq. (2), the fractional FAGM(1,1,k) model is reduced to the GM(1,1,k) model [23] with the form (1) dx (t) (1) + ax (t) = bt. (14) dt Setting r = 1, b = 0, c = 1 in Eq. (2), the fractional FAGM(1,1,k) model is reduced to the GM(1,1) model [29] with the form (1) dx (t) (1) + ax (t) = c. (15) dt Thereafter, the flaw of the FAGM(1,1,k) model is provided. Integrating both sides of Eq. (2) in the interval [k − 1, k], we obtain Z Z Z Z k k k k (r) (r) dx (t) + ax (t)dt = btdt + cdt. (16) k−1 k−1 k−1 k−1 14 R (r) (r) (r) (r−1) With the knowledge of dx (t) = x (k) − x (k − 1) = x (k), k−1 R R k k tdt = (2k − 1)/2 and dt = 1, the exact discrete differential equation k−1 k−1 is expressed by (2k − 1) b (r−1) (r) x (k) + a x (t)dt = + c. (17) k−1 A comparison between Eq. (17) and the basic Eq. (3) indicates that differ- (r) (r−1) (r) ences exist in the background value z (k) = 0.5 x (k − 1) + x (k) and (r) x (t)dt. It is highly inaccurate to compute the integration utilising the k−1 (r) trapezoid formula if x (t) is not a linear function. Thus, the basic form and whitenisation differential equation of the FAGM(1,1,k) model do not strictly match. 5. Parameter optimisation of FAGM(1,1,k) model It can easily be verified that the parameters φ = (a, b, c), derived by the least-squares estimation in Eq. (3) and the parameters of the time response (r) function xˆ (k), k = 2, 3, . . . , n derived by Eq. (2), have different meanings. When the response function dose not satisfy the basic equation, large errors may arise. To match the basic Eq. (3) and response function (5), the system parameters are optimised in this system. Setting the optimised parameters of the grey system as (α, β, γ) and replac- ing the parameters φ = (a, b, c) in Eq. (2), the whitening differential equation is rewritten as (r) dx (t) (r) + αx (t) = βt + γ, r > 0. (18) dt Similarly, the general solution of Eq. (18) is given by β β γ β β γ (r) (0) −α(t−1) x (t) = x (1) − + − e + t − + . (19) 2 2 α α α α α α Furthermore, we have β β γ β β γ (r) (0) −α(k−1) xˆ (k)= x (1) − + − e + k − + , k = 2, 3, . . . , n. (20) 2 2 α α α α α α 15 Substituting Eq. (20) into the left side of Eq. (3), we obtain (r) (r) (r−1) (r) L(t) = x (k) − x (k − 1) + x (k − 1) + x (k) a a (r) (r−1) = 1 + x (k) − 1 − x (k − 1) 2 2 a β β γ β β γ (0) −α(k−1) = 1 + x (1) − + − e + k − + 2 2 2 α α α α α α a β β γ β β γ (0) −α(k−2) − 1 − x (1) − + − e + (k − 1) − + 2 2 2 α α α α α α h    i a a β β γ α (0) −α(k−1) = 1 + − 1 − e x (1) − + − e 2 2 α α α β a β β γ + ak + 1 − − − a. (21) α 2 α α α Owing to the left side L(t) and right side R(t) equivalence, namely L(t) − R(t) = 0, it is implied that a a 1 + − 1 − e = 0, (22) 2 2 a = b, (23) a β β γ b 1 − − − a = c − . (24) 2 α α α 2 It follows from Eqs. (22) to (24) that 2 + a α = ln , (25) 2 − a b 2 + a β = ln , (26) a 2 − a αc αb β β β γ = − + + − . (27) a 2a α 2 a Thus, the optimised parameters (α, β, γ) are obtained by Eqs. (25) to (27), and they also indicate that the parameters (a, b, c) derived by the least- squares estimation satisfy the relationship in Eqs. (25) to (27). In this pa- per, the FAGM(1,1,k) model with optimised parameters is referred to as the FAGMO(1,1,k) model. (0) (0) (0) Theorem 4. Assuming that the original data x (1), x (2), . . . , x (n) sat- isfy Eq. (20) with the given parameters αˆ, β, γˆ , the parameters (α, β, γ) of 16 FAGMO(1,1,k) obtained by Eqs. (4) and (25) to (27) satisfy the relationship (0) (0) (0) αˆ = α, β = β, γˆ = γ, and the predicted values xˆ (1), xˆ (2), . . . , xˆ (n) are (0) (0) (0) equal to the given data x (1), x (2), . . . , x (n) . (0) Proof 4. Substituting the original data x (k), k = 1, 2, . . . , n into Eq. (4), the parameters αˆ, β, γˆ can be derived. The parameters (α, β, γ) of FAGMO(1,1,k) can be obtained from Eqs. (25) to (27). Obviously, αˆ = α, β = β, γˆ = γ. (0) Therefore, the predicted values xˆ (k), k = 1, 2, . . . , n of the FAGMO(1,1,k) are (0) equal to the given data x (k), k = 1, 2, . . . , n. Theorem 4 demonstrates that the FAGMO(1,1,k) model is accurate for predicting arbitrary sequences that can be modelled by Eq. (20), while the FAGM(1,1,k) model cannot describe the sequences accurately owing to there always being a non-zero difference between the real parameters and the param- eters (a, b, c). Theorem 5. The optimised parameters (α, β, γ) are approximately equivalent to the parameters (a, b, c) when the value of |a| is very small; that is, α ≈ a, β ≈ b, γ ≈ c. (28) Proof 5. We first consider the difference between parameter α and a, which is 2 + a ε (a) = α − a = ln − a. (29) 2 − a It is known that ε (a)| = 0 and the first-order derivative is a=0 2+a d ln − a 2 dε (a) 2−a 4 a = = − 1 = . (30) 2 2 da da 4 − a 4 − a When |a| < 2, the derivative of ε (a) is positive, which indicates that the function ε (a) is a monotonically increasing function in the interval [-2,2]. Thus, the value ε (a) approaches zero as |a| decreases. Therefore, α ≈ a when |a| is very small. Secondly, the difference between β and b is expressed as b 2 + a b ε (a) = β − b = ln − b = ε (a) . (31) 2 1 a 2 − a a 17 1 2+a 2−a 2−a+2+a 4 Owing to lim ln = lim = lim = 1, we know that 2 2 a 2−a 2+a (2−a) 4−a a→0 a→0 a→0 ε (a) → 0 when a → 0. The first-order derivative of ε (a) is dε (a) b 4a 2 + a = − ln , (32) 2 2 da a 4 − a 2 − a which is also positive when |a| < 2. Thus, ε (a) decreases when the value of |a| decreases and β ≈ b when |a| is very small. Thirdly, the difference between γ and c is expressed as αc αb β β β ε (a) b ε (a) = γ − c = − + + − − c = c − . (33) a 2a α 2 a a a It follows from α ≈ a and β ≈ b that γ ≈ c when |a| is very small. From Theorem 5, we know that the differences between the parameters (α, β, γ) and (a, b, c) are decrease along with smaller |a|. Table 1 provides the values of ε (a) and ε (a)/a under different values of |a|. 1 1 Table 1: Values of ε (a) and ε (a) /a under different values of |a| 1 1 |a| 0.1 0.2 0.3 0.5 0.7 1.0 1.3 1.6 1.9 ε (a) 0.0001 0.0007 0.0023 0.0108 0.0309 0.0986 0.2506 0.5972 1.7636 ε (a) /a 0.0008 0.0034 0.0076 0.0217 0.0441 0.0986 0.1928 0.3733 0.9282 6. Modelling evaluation criteria and detailed modelling steps To evaluate forecasting accuracy of the FAGMO(1,1,k) model, the root mean squared percentage error (RMSPE) is applied to the prior-sample period (RM- SPEPR) and post-sample period (RMSPEPO). In general, the RMSPEPR, RM- SPEPO and RMSPE are defined as ν (0) (0) u X 1 xˆ (k) − x (k) t 1 1 RMSPEPR = × 100%, (34) (0) x (k) k=1 1 (0) (0) u X 1 xˆ (k) − x (k) t 1 1 RMSPEPO = × 100%, (35) (0) n − ν x (k) k=ν+1 1 (0) (0) 1 xˆ (k) − x (k) t 1 1 RMSPE = × 100%, (36) (0) x (k) k=1 1 18 where ν is the number of samples used to construct the model and n is the total number of samples. The index of agreement of the forecasting results is defined as (0) (0) xˆ (k) − x (k) k=1 IA = 1 − , (37) (0) (0) xˆ (k) − x + x (k) − x k=1 which is also a useful performance measure for sensitivity to differences in the observed and predicted data, where x is the average sample value. The average forecasting error (AE) and the mean absolute forecasting error (MAE) are (0) (0) AE = xˆ (k) − x (k) , (38) k=1 (0) (0) MAE = xˆ (k) − x (k) , (39) k=1 where AE reflects the positive and negative errors between the predicted and ob- served values, while MAE is applied for estimating the change in the forecasting model. The detailed modelling steps of the fractional FAGMO(1,1,k) are provided below. (0) Step 1: Determine the original data series x (i), i = 1, 2, . . . , n, and r- (r) (0) r AGO series X = X A . Step 2: Calculate the matrices B and Y to determine (a, b, c) using Eq. (4). Step 3: Compute the parameters (α, β, γ) by employing Eqs. (25) to (27). (0) Step 4: Substitute the values of x (1) and (α, β, γ) into Eq. (20) to (r) compute the predicted values X . (0) Step 5: Apply the r-IAGO matrix to obtain the restored values X = (r) r X D . 7. Validation of FAGMO(1,1,k) model This section provides numerical examples to validate the accuracy of the FAGMO(1,1,k) model compared to the FAGM(1,1,k) model and others. 19 7.1. Validation of FAGMO(1,1,k) and FAGM(1,1,k) models This subsection presents a numerical example to validate the accuracy of the FAGMO(1,1,k) and FAGM(1,1,k) models. The values r and α are provided (r) in the interval [0.01, 2] and [-1.99, 1.99], respectively. The initial point x (1) is randomly generated in the interval [1, 2] by the uniform distribution, while the parameters β and γ are randomly generated in the intervals [0, 5] and [0, (r) 100], respectively, by the uniform distribution. The other x (i) (i > 1) are generated with the aid of Eq. (20). All data used for the example are explained in Fig. 5. given in an interval randomly generated ✁✕ ✁✕ ▼❇ ❑❆ ✁ ❇ (r) ❇ dx (t) (r) + αx (t) = βt + γ dt (r) x (1) randomly generated (r) x (i) calculated by Eq.(20) Figure 5: Diagram of data for validation We define the notation in the following analysis 2 2 2 ε = (p − α) + (l − β) + (q − γ) , (40) params where (α, β, γ) are the provided parameters of Eq. (20) and (p, l, q) are the estimated parameters of the FAGM(1,1,k) or FAGMO(1,1,k) model. When applying the above parameters, the graphs are displayed in Figs. 6 and 7. We observe from Fig. 6 that the maximum ε of FAGMO(1,1,k) and params −5 FAGM(1,1,k) are 5.4228×10 and 489.9434, respectively, where the magnitude is approximately 9034932. Furthermore, the ε of the FAGM(1,1,k) model params is very small when α is near zero, which is coincident with Theorem 5. From Fig. 20 7, the maximum RMSPEs of FAGMO(1,1,k) and FAGM(1,1,k) are 0.0103% and 814.3864%, respectively, where the magnitude is approximately 79100. (r) It is known that the parameters β, γ, and initial points x (1) are all ran- (r) domly generated, which implies that the values of parameters β, γ and x (1) have no influence on the output series. Here, the values r and α are the most important factors affecting the accuracy of the grey models. −5 x 10 6 500 0 0 2 2 1 2 1 2 1.5 1.5 0 0 1 1 −1 −1 0.5 0.5 −2 −2 0 0 α α r r Figure 6: Values of ε of FAGMO(1,1,k) (left) and FAGM(1,1,k) (right) models params 0.012 1000 0.01 0.008 0.006 0.004 0.002 0 0 2 2 1 2 1 2 1.5 1.5 0 0 1 1 −1 −1 0.5 0.5 −2 0 −2 0 α α r r Figure 7: Values of RMSPE of FAGMO(1,1,k) (left) and FAGM(1,1,k) (right) models 7.2. Validation of FAGMO(1,1,k) model and other grey models This subsection further demonstrates the advantage of the FAGMO(1,1,k) model using two real cases. RMSPE params RMSPE ε params Case 1: (Predicting cumulative oil field production). We consider an ex- ample from the paper [54] that provides sample data. The data from 1999 to 2009 are applied to construct the grey model, while the data from 2010 to 2012 are used for prediction. The values are listed in Table 2, indicating that the FAGMO(1,1,k) model outperforms the other models in this case. Table 2: Results of ENGM, FAGM(1,1), FAGM(1,1,k) and FAGMO(1,1,k) models Year Data ENGM FAGM(1,1) FAGM(1,1,k) FAGMO(1,1,k) r = 1 r = 0.1106 r = 0.4073 r = 0.4052 1999 73.8217 73.8217 73.8217 73.8217 73.8217 2000 136.8817 138.4900 138.1621 137.1758 136.4573 2001 195.0590 195.4541 195.5377 196.1598 195.7633 2002 247.8547 247.9776 247.7638 249.3183 249.1781 2003 297.0902 296.4067 295.7629 297.2895 297.2750 2004 342.6394 341.0604 340.1238 341.0008 341.0322 2005 382.4312 382.2332 381.2700 381.2882 381.3320 2006 420.0399 420.1964 419.5291 418.8204 418.8699 2007 454.0430 455.2001 455.1670 454.1099 454.1712 2008 485.1171 487.4752 488.4068 487.5452 487.6290 2009 519.8508 517.2342 519.4402 519.4217 519.5393 2010 552.6569 544.6734 548.4350 549.9665 550.1281 2011 581.6092 569.9736 575.5400 579.3572 579.5714 2012 608.1863 593.3015 600.8887 607.7346 608.0086 RMSPEPR 0.4521% 0.4582% 0.3539% 0.3259% RMSPEPO 2.0066% 1.0185% 0.3617% 0.3332% Case 2: (Predicting foundation settlement close neighbouring Yangtze River). We consider an example from the paper [53], which provides sample data to con- struct the grey model. The values are presented in Table 3, indicating that the FAGMO(1,1,k) model outperforms the other models in this case. 22 Table 3: Results of ONGM(1,1,k,c), FAGM(1,1), FAGM(1,1,k) and FAGMO(1,1,k) models Day Data ONGM(1,1,k,c) FAGM(1,1) FAGM(1,1,k) FAGMO(1,1,k) r = 1 r = 0.0065 r = 0.2266 r = 0.2295 10 23.36 23.3600 23.3600 23.3600 23.3600 20 43.19 42.1779 43.3517 43.0586 43.0644 30 58.73 59.2549 59.4403 58.8205 58.8124 40 70.87 72.8374 72.4009 71.9763 71.9545 50 83.71 83.6405 82.8451 83.0247 82.9932 60 92.91 92.2330 91.2620 92.2158 92.1789 70 99.73 99.0672 98.0442 99.6885 99.6491 80 105.08 104.5030 103.5079 105.5215 105.4805 90 109.73 108.8264 107.9079 109.7568 109.7127 100 112.19 112.2652 111.4497 112.4117 112.3598 110 113.45 115.0002 114.2991 113.4857 113.4181 RMSPE 1.2730% 1.3257% 0.6030% 0.6011% 8. Applications In this section, the FAGMO(1,1,k) model is applied to forecast the nuclear energy consumption of China. The computational results of the FAGMO(1,1,k) model are compared to the ENGM [54], ONGM(1,1,k) [53], FAGM(1,1) [40] and FAGM(1,1,k) models. 8.1. Raw data Raw data of the nuclear energy consumption of China were collected from the report of the BP Statistical Review of World Energy 2018. The first 10 samples th th belonging to the 11 and the 12 Five-Year Plans are applied to construct the th prediction model, while the remaining samples of the 13 Five-Year Plan are used to validate and compare the forecasting results (see Table 4). Table 4: Raw data of nuclear energy consumption of China, Mtoe Year Data Year Data Year Data 2006 12.4 2011 19.5 2016 48.2 2007 14.1 2012 22.0 2017 56.2 2008 15.5 2013 25.3 2009 15.9 2014 30.0 2010 16.7 2015 38.6 23 8.2. Simulation and prediction results The simulation and prediction results are listed in Table 5 and Fig. 8, while the errors are listed in Table 6 and Fig. 9. The nuclear energy consumption of China from 2016 to 2017 is predicted ac- cording to the established grey models. It can be observed in Table 5 and Fig. 8 that five grey models, namely ENGM, ONGM(1,1,k), FAGM(1,1), FAGM(1,1,k) and FAGMO(1,1,k), successfully identify the trend of China’s nuclear energy consumption. However, these grey models differ from one another in terms of the prediction values from 2016 to 2020. From Fig. 8, China’s nuclear energy consumption is overestimated by the ENGM, ONGM(1,1,k) and FAGM(1,1,k) models, and underestimated by the FAGM(1,1) model. The values predicted by FAGMO(1,1,k) are substantially closer to the raw data than those predicted by the other models. We can observe from Table 6 and Fig. 9 that the RMSPEPR, RMSPEPO and RMSPE of FAGMO(1,1,k) are 3.1409%, 4.1502% and 3.3304%, respectively. The RMSPEPR, RMSPEPO and RMSPE of ENGM are as high as 8.3788%, 30.3663% and 14.5667%, those of ONGM(1,1,k) are 2.0494%, 12.0510% and 5.2635%, those of FAGM(1,1) are 4.8680%, 11.7968% and 6.5529%, and those of FAGM(1,1,k) are 2.3299%, 6.3828% and 3.3636%, respectively. The IA, AE and MAE of FAGMO(1,1,k) are 0.9985, 0.2526 and 0.7513, those of ENGM are 0.9538, 4.0536 and 4.0536, those of ONGM(1,1,k) are 0.9911, 1.0225 and 1.1896, those of FAGM(1,1) are 0.9887, -1.1818 and 1.7105, and those of FAGM(1,1,k) are 0.9971, 0.2736 and 0.8043, respectively. The computational results indicate that the FAGMO(1,1,k) model outperforms ENGM, ONGM(1,1,k), FAGM(1,1) and FAGM(1,1,k), while ENGM exhibits the most inferior performance. 24 Table 5: Simulation and prediction results of nuclear energy consumption by grey models Year Data ENGM ONGM(1,1,k) FAGM(1,1) FAGM(1,1,k) FAGMO(1,1,k) r = 1 r = 1 r = 1.4127 r = 1.0593 r = 1.1595 2006 12.4 12.4000 12.4000 12.4000 12.4000 12.4000 2007 14.1 14.9788 14.4788 15.0242 14.7054 15.0891 2008 15.5 15.6744 15.1057 13.9808 15.0121 14.8608 2009 15.9 16.6846 16.0032 15.0566 15.8012 15.5886 2010 16.7 18.1520 17.2884 16.9219 17.0700 16.9760 2011 19.5 20.2831 19.1286 19.3953 18.9344 19.0534 2012 22.0 23.3785 21.7635 22.4687 21.5861 21.9432 2013 25.3 27.8741 25.5363 26.1951 25.3029 25.8633 2014 30.0 34.4036 30.9383 30.6625 30.4740 31.1013 2015 38.6 43.8871 38.6732 35.9872 37.6390 38.0473 2016 48.2 57.6610 49.7483 42.3129 47.5433 47.2178 2017 56.2 77.6662 65.6063 49.8133 61.2149 59.2933 2018 – 106.7219 88.3125 58.6959 80.0704 75.1679 2019 – 148.9226 120.8244 69.2074 106.0614 96.0147 2020 – 210.2150 167.3766 81.6403 141.8758 123.3723 raw data ENGM ONGM(1,1,k,c) FAGM(1,1) FAGM(1,1,k) FAGMO(1,1,k) data for building models Figure 8: Comparison among five grey models for nuclear energy consumption data of nuclear energy consumption Table 6: Relative error values of nuclear energy consumption by five grey models Year ENGM ONGM(1,1,k) FAGM(1,1) FAGM(1,1,k) FAGMO(1,1,k) 2006 0 0 0 0 0 2007 0.0623 0.0269 0.0655 0.0429 0.0701 2008 0.0112 0.0254 0.0980 0.0315 0.0412 2009 0.0493 0.0065 0.0530 0.0062 0.0196 2010 0.0869 0.0352 0.0133 0.0222 0.0165 2011 0.0402 0.0190 0.0054 0.0290 0.0231 2012 0.0627 0.0108 0.0213 0.0188 0.0026 2013 0.1017 0.0093 0.0354 0.0001 0.0222 2014 0.1468 0.0313 0.0221 0.0158 0.0367 2015 0.1370 0.0019 0.0677 0.0249 0.0143 2016 0.1963 0.0321 0.1221 0.0136 0.0204 2017 0.3820 0.1674 0.1136 0.0892 0.0550 RMSPEPR 8.3788% 2.0494% 4.8680% 2.3299% 3.1409% RMSPEPO 30.3663% 12.0510% 11.7968% 6.3828% 4.1502% RMSPE 14.5667% 5.2635% 6.5529% 3.3636% 3.3304% IA 0.9538 0.9911 0.9887 0.9971 0.9985 AE 4.0536 1.0225 -1.1818 0.2736 0.2526 MAE 4.0536 1.1896 1.7105 0.8043 0.7513 0.4 ENGM ENGM ENGM ONGM(1,1,k,c) ONGM(1,1,k,c) ONGM(1,1,k,c) 30.3663 0.35 FAGM(1,1) FAGM(1,1) FAGM(1,1) FAGM(1,1,k) FAGM(1,1,k) FAGM(1,1,k) FAGMO(1,1,k) FAGMO(1,1,k) FAGMO(1,1,k) 0.3 0.25 20 RMSPEPO 0.2 14.5667 12.051 11.7967 RMSPE 0.15 0.1 6.3828 6.5529 5.2635 4.1502 3.3636 3.3304 0.05 0 0 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 1 2 3 4 5 6 7 8 9 10 year Figure 9: Errors among five grey models for nuclear energy consumption relative error 8.3. Further discussions As demonstrated by the case study, the novel FAGMO(1,1,k) model outper- forms other grey models. Moreover, it should be noted that in this paper we only conduct short-term forecasting, while it is well known that several exist- ing energy models can perform long-term forecasting, such as LEAP, TIMES and NEMS. We will discuss the difference between our model and these models further, following a very brief introduction to such models. • LEAP (long-range energy alternatives planning system) [14, 15] is a scenario- based energy environment modelling tool for climate change mitigation and energy policy analysis. It can be applied to examine energy production and consumption, as well as resource extraction in all sectors. The model studies the effects of various factors on energy consumption under different scenarios given an objective. LEAP is generally used for forecasting studies of between 20 and 50 years. • TIMES (The Integrated MARKAL-EFOM System) [16, 17] is an evolution of MARKAL, which was developed by the Energy Technology Systems Anal- ysis Programme of the IEA. It combines technical engineering and economic approaches, and uses linear programming to produce a least-cost energy system under numerous user-specified constraints. The software is used to analyse en- ergy, economic and environmental issues at different levels over several decades. • NEMS (National Energy Modeling System) [18, 19] is a long-standing US government policy model, which computes equilibrium fuel prices and quantities for the US energy sector. NEMS is used to model the demand side explicitly; in particular, to determine consumer technology choices in the residential and commercial building sectors. These models can perform long-term energy consumption projections. How- ever, they may require a large amount of data, such as population growth, GDP, urbanisation, energy policies and energy strategies. In numerous prac- tical situations, it is very difficult to obtain complete information because of time and cost limitations. The grey prediction model is an efficient method for conducting accurate forecasting with at least four samples. Compared to the 27 major energy models, the grey model is an effective choice for predicting China’s nuclear energy consumption. This paper collected 12 samples of China’s nuclear energy consumption from the BP Statistical Review of World Energy 2018. Thus, the LEAP, TIMES and NEMS models are all inapplicable owing to poor information. By employing th th the grey system theory and actual data from the 11 and the 12 Five-Year Plans, the FAGMO(1,1,k) model was constructed. It can be observed in Table 5 that the prediction value of FAGMO(1,1,k) is 123.3723 Mtoe in 2020, which is larger than the 84.6318 Mtoe provided in the BP energy outlook 2018. The main reasons for this are as follows. i) It is infeasible to consider factors such as energy policies and China’s energy strategies, which affect the current situation of China’s nuclear en- ergy consumption, in our proposed model because the FAGMO(1,1,k) model is univariate. However, the forecasting models of institutions including BP, the IEA and APEC are based on widely collected data. Furthermore, grey models are mainly used for short-term forecasting in the calculation process, such as [22, 23, 24, 27, 57]. Therefore, the forecasting results are relatively acceptable, reflecting the growth trend of future nuclear energy consumption in China. ii) In China’s nuclear energy market, 38 nuclear power reactors are in op- eration, 19 nuclear power reactors are under construction and more are to be constructed by the end of 2016. This is the reason for the increase in nuclear en- ergy consumption in recent years. However, no new nuclear projects have been approved for construction in 2016. Moreover, the State Council approved new safety rules and a nuclear power development plan following Japan’s Fukushima Daiichi crisis in 2011. These factors have also resulted in a slight slowdown in China’s nuclear energy consumption. In the future, nuclear energy could provide an important alternative to fos- sil fuels such as coal and oil, and its proportion of the total primary energy consumption will increase yearly. Based on our forecasting results using the FAGMO(1,1,k) model, the future nuclear energy consumption of China will in- crease rapidly if no certain restrictions are placed thereon. This implies that 28 higher management and technical levels are necessary to meet the safety and quality requirements. Therefore, China’s government and policy makers should pay additional attention to the safety and quality issues of nuclear energy to achieve long-term, environmentally friendly and low-carbon energy goals and lay the foundation for the sustainable development of China’s energy and economy. 9. Conclusions By applying the grey modelling technique and parameter optimisation method, the fractional FAGMO(1,1,k) model was proposed to predict China’s nuclear en- th ergy consumption of the 13 Five-Year Plan, based on the updated data from 2006 to 2015. The forecasting results provide the growth trend of the future nu- clear energy consumption of China, and also offer a guideline for policymaking and project planning. It can be observed that FAGMO(1, 1, k) is quite easy to use, with satisfac- tory accuracy in short-term nuclear consumption forecasting. For long-term prediction, its error will be larger because only 10 samples are used for mod- elling. This study is expected to be able to forecast the energy consumption of other countries that share similar patterns of economic development and energy consumption structures, among others. Furthermore, the optimised method applied to improve the FAGM(1,1,k) model can be used to improve other first- order grey models, such as NGBM(1,1), GMC(1,n) and RDGM(1,n). These are possible extensions and suggested directions for our future research. Acknowledgments This research was supported by the National Natural Science Foundation of China (No. 71771033), Longshan academic talent research supporting program of SWUST (No. 17LZXY20), Doctoral Research Foundation of Southwest Uni- versity of Science and Technology (No. 15zx7141, 16zx7140) and the Open Fund (PLN201710) of the State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation (Southwest Petroleum University). 29 References [1] F. M. Mirza, A. Kanwal, Energy consumption, carbon emissions and eco- nomic growth in Pakistan: dynamic causality analysis, Renewable & Sus- tainable Energy Reviews 72 (5) (2017) 1233–1240. [2] M. Shahbaz, T. H. V. Hoang, M. K. Mahalik, D. Roubaud, Energy con- sumption, financial development and economic growth in India: new evi- dence from a nonlinear and asymmetric analysis, Energy Economics 66 (3) (2017) 199–212. [3] H. A. Bekhet, A. Matar, T. Yasmin, CO emissions, energy consumption, economic growth, and financial development in GCC countries: dynamic simultaneous equation models, Renewable & Sustainable Energy Reviews 70 (4) (2017) 117–132. [4] M. Minda, C. Wei, Do commercial building sector-derived carbon emissions decouple from the economic growth in Tertiary Industry? A case study of four municipalities in China, Science of The Total Environment 650 (Part 1) (2019) 822–834. [5] J. Fan, X. Wang, L. Wu, H. Zhou, F. Zhang, X. Yu, X. Lu, Y. Xiang, Comparison of support vector machine and extreme gradient boosting for predicting daily global solar radiation using temperature and precipitation in humid subtropical climates: A case study in china, Energy Conversion and Management 164 (2018) 102–111. [6] Y. Wang, X. Yi, Transient pressure behavior of a fractured vertical well with a finite-conductivity fracture in triple media carbonate reservoir, Journal of Porous Media 20 (8) (2017) 707–722. [7] Y. Wang, X. Yi, Flow modeling of well test analysis for a multiple-fractured horizontal well in triple media carbonate reservoir, International Journal of Nonlinear Sciences and Numerical Simulation 19 (5) (2018) 439–457. doi:https://doi.org/10.1515/ijnsns-2016-0075. 30 [8] Y. Hu, E. Mackay, O. Vazquez, I. O, Streamline simulation of barium sul- fate precipitation occurring within the reservoir coupled with analysis of observed produced water chemistry data to aid scale management, SPE Production & Operations 33 (1) (2018) 85–101. [9] R. Brini, M. Amara, H. Jemmali, Renewable energy consumption, inter- national trade, oil price and economic growth inter-linkages: the case of Tunisia, Renewable & Sustainable Energy Reviews 76 (9) (2017) 620–627. [10] P. Du, J. Wang, W. Yang, T. Niu, Multi-step ahead forecasting in electrical power system using a hybrid forecasting system, Renewable Energy 122 (7) (2018) 533–550. [11] W. Cai, C. Liu, C. Zhang, M. Ma, W. Rao, W. Li, K. He, M. Gao, Devel- oping the ecological compensation criterion of idustrial solid waste based on emergy for sustainable development, Energy 157 (15) (2018) 940–948. [12] W. Cai, F. Liu, O. Dinolov, J. Xie, P. Liu, J. Tuo, Energy benchmarking rules in machining systems, Energy 142 (1) (2018) 258–263. [13] X. Wu, X. Ding, T. Lu, J. Wang, Topological dynamics of zadeh’s extension on upper semi-continuous fuzzy sets, International Journal of Bifurcation and Chaos 27 (10) (2017) 1–13. [14] S. P. Schnaars, How to develop and use scenarios, Long Range Planning 20 (1) (1987) 105–114. [15] K. Dong, R. Sun, H. Li, H. Jiang, A review of China’s energy consumption structure and outlook based on a long-range energy alternatives modeling tool, Petroleum Science 14 (1) (2017) 214–227. [16] J. Shi, W. Chen, X. Yin, Modelling building’s decarbonization with appli- cation of China Times model, Applied Energy 162 (15) (2016) 1303–1312. [17] H. Zhang, W. Chen, W. Huang, Times modelling of transport sector in China and USA: comparisons from a decarbonization perspective, Applied Energy 162 (15) (2016) 1505–1514. 31 [18] S. A. Gabriel, A. S. Kydes, P. Whitman, The national energy modeling system: a large-scale energy-economic equilibrium model, Operations Re- search 49 (1) (2001) 14–25. [19] R. Soroush, A. Koochi, M. Keivani, M. Abadyan, A bilayer model for in- corporating the coupled effects of surface energy and microstructure on the electromechanical stability of NEMS, International Journal of Structural Stability and Dynamics 17 (4) (2017) 1771005. [20] B. Zeng, Y. Tan, H. Xu, J. Quan, L. Wang, X. Zhou, Forecasting the electricity consumption of commercial sector in Hong Kong using a novel grey dynamic prediction model., Journal of Grey System 30 (1) (2018) 157–172. [21] B. Zeng, S. Liu, A self-adaptive intelligence gray prediction model with the optimal fractional order accumulating operator and its application, Math- ematical Methods in the Applied Sciences 40 (18) (2017) 7843–7857. [22] S. J. Feng, Y. D. Ma, Z. L. Song, J. Ying, Forecasting the energy con- sumption of China by the grey prediction model, Energy Sources Part B-economics Planning and Policy 7 (4) (2012) 376–389. [23] J. Cui, S. Liu, B. Zeng, N. Xie, A novel grey forecasting model and its optimization, Applied Mathematical Modelling 37 (6) (2013) 4399–4406. [24] H. Duan, G. R. Lei, K. Shao, Forecasting crude oil consump- tion in china using a grey prediction model with an optimal fractional-order accumulating operator, Complexity 2018 (2018) 1–12. doi:https://doi.org/10.1155/2018/3869619. [25] B. Zeng, H. Duan, Y. Bai, W. Meng, Forecasting the output of shale gas in china using an unbiased grey model and weakening buffer operator, Energy 151 (15) (2018) 238–249. 32 [26] B. Zeng, L. Chuan, Improved multi-variable grey forecasting model with a dynamic background-value coefficient and its application, Computers & Industrial Engineering 118 (4) (2018) 278–290. [27] L. Wu, Y. Gao, Y. Yang, X. Chen, Using a novel multi-variable grey model to forecast the electricity consumption of Shangdong province in China, Energy 157 (15) (2018) 327–335. [28] J. Wang, P. Du, H. Lu, W. Yang, T. Niu, An improved grey model optimized by multi-objective ant lion optimization algorithm for an- nual electricity consumption forecasting, Applied Soft Computing, doi: https://doi.org/10.1016/j.asoc.2018.07.022. [29] J. Deng, Control problems of grey systems, Systems & Control Letters 1 (5) (1982) 288–294. [30] N. Xie, S. Liu, Discrete grey forecasting model and its optimization, Ap- plied Mathematical modelling 33 (2) (2009) 1173–1186. [31] N. Xie, S. Liu, Y. Yang, C. Yuan, On novel grey forecasting model based on non-homogeneous index sequence, Applied Mathematical Modelling 37 (7) (2013) 5059–5068. [32] N. Xie, R. Wang, A historic review of grey forecasting models, Journal of grey System 29 (4) (2017) 1–29. [33] Z. Wang, D. Ye, Forecasting Chinese carbon emissions from fossil energy consumption using non-linear grey multivariable models, Journal of Cleaner Production 142 (20) (2017) 600–612. [34] Z. Wang, Q. Li, L. Pei, Grey forecasting method of quarterly hydropower production in China based on a data grouping approach, Applied Mathe- matical Modelling 51 (11) (2017) 302–316. [35] Z. X. Wang, H. H. Zheng, L. L. Pei, T. Jin, Decomposition of the factors influencing export fluctuation in China’s new energy industry based on a constant market share model, Energy Policy 109 (10) (2017) 22–35. 33 [36] X. Ma, Z. Liu, Application of a novel time-delayed polynomial grey model to predict the natural gas consumption in China, Journal of Computational & Applied Mathematics 324 (11) (2017) 17–24. [37] X. Ma, Z. Liu, Y. Wang, Application of a novel nonlinear multivari- ate grey bernoulli model to predict the tourist income of China, Jour- nal of Computational and Applied Mathematics 347 (2018) 84 – 94. doi:https://doi.org/10.1016/j.cam.2018.07.044. [38] X. Ma, Z. Liu, The kernel-based nonlinear multivariate grey model, Applied Mathematical Modelling 56 (4) (2018) 217–238. [39] X. Ma, Z. Liu, Predicting the oil production using the novel multivariate nonlinear model based on Arps decline model and kernel method, Neural Computing & Applications 29 (2) (2018) 579–591. [40] L. Wu, S. Liu, L. Yao, S. Yan, D. Liu, Grey system model with the fractional order accumulation, Communications in Nonlinear Science and Numerical Simulation 18 (7) (2013) 1775–1785. [41] L. Wu, S. Liu, W. Cui, D. Liu, T. Yao, Non-homogenous discrete grey model with fractional-order accumulation, Neural Computing & Applica- tions 25 (5) (2014) 1215–1221. [42] L. Wu, S. Liu, L. Yao, R. Xu, X. Lei, Using fractional order accumulation to reduce errors from inverse accumulated generating operator of grey model, Soft Computing 19 (2) (2015) 483–488. [43] L. Wu, S. Liu, Z. Fang, H. Xu, Properties of the GM(1,1) with fractional order accumulation, Applied Mathematics and Computation 252 (1) (2015) 287–293. [44] X. Xiao, H. Guo, S. Mao, The modeling mechanism, extension and opti- mization of grey GM (1,1) model, Applied Mathematical Modelling 38 (5-6) (2014) 1896–1910. 34 [45] M. Gao, S. Mao, X. Yan, J. Wen, Estimation of Chinese CO emission based on a discrete fractional accumulation grey model, The Journal of Grey System 27 (4) (2015) 114–130. [46] S. Mao, M. Gao, X. Xiao, M. Zhu, A novel fractional grey system model and its application, Applied Mathematical Modelling 40 (7-8) (2016) 5063– [47] Y. Shen, P. Qin, Optimization of grey model with the fractional order accumulation, Journal of Grey System 17 (3) (2014) 127–132. [48] Y. Yang, D. Xue, Continuous fractional-order grey model and electricity prediction research based on the observation error feedback, Energy 115 (1) (2016) 722–733. [49] Y. Yang, D. Xue, Modified grey model predictor design using optimal fractional-order accumulation calculus, IEEE/CAA Journal of Automat- ica Sinica 4 (4) (2017) 1–10. [50] Q. Liu, D. Yu, Non-equidistance and nonhomogeneous grey model NN- FGM(1,1) with the fractional order accumulation and its application, Jour- nal of Interdisciplinary Mathematics 20 (6-7) (2017) 1423–1426. [51] X. Kong, Y. Wei, Optimization of DGM(2,1), Journal of Grey System 12 (1) (2009) 9–13. [52] C.-S. Yu, S.-H. Chen, Applying genetic algorithm and GM(h, N) in sales forecasting, Journal of Grey System 20 (2) (2017) 109–116. [53] P. Chen, H. Yu, Foundation settlement prediction based on a novel NGM model, Mathematical Problems in Engineering 2014 (1) (2014) 1–8. [54] X. Ma, Z. Liu, Predicting the cumulative oil field production using the novel grey ENGM model, Journal of Computational & Theoretical Nanoscience 13 (1) (2016) 89–95. 35 [55] X. Ma, Z. Liu, The GMC(1, n) model with optimized parameters and its applications, Journal of Grey System 29 (4) (2017) 121–137. [56] Z. Hou, H. Xie, H. Zhou, P. Were, O. Kolditz, unconventional gas resources in china, Environmental Earth Sciences 73 (10) (2015) 5785–5789. [57] C.-I. Chen, H. Chen, S. Chen, Forecasting of foreign exchange rates of Taiwan’s major trading partners by novel nonlinear grey Bernoulli model NGBM(1, 1), Communications in Nonlinear Science and Numerical Simu- lation 13 (6) (2008) 1194–1204. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Statistics arXiv (Cornell University)

Application of the novel fractional grey model FAGMO(1,1,k) to predict China's nuclear energy consumption

Statistics , Volume 2018 (1809) – Sep 28, 2018

Loading next page...
 
/lp/arxiv-cornell-university/application-of-the-novel-fractional-grey-model-fagmo-1-1-k-to-predict-0sbp0FAINC
ISSN
0360-5442
eISSN
ARCH-3347
DOI
10.1016/j.energy.2018.09.155
Publisher site
See Article on Publisher Site

Abstract

At present, the energy structure of China is shifting towards cleaner and lower amounts of carbon fuel, driven by environmental needs and technological ad- vances. Nuclear energy, which is one of the major low-carbon resources, plays a key role in China’s clean energy development. To formulate appropriate energy policies, it is necessary to conduct reliable forecasts. This paper discusses the nuclear energy consumption of China by means of a novel fractional grey model FAGMO(1,1,k). The fractional accumulated generating matrix is introduced to analyse the fractional grey model properties. Thereafter, the modelling proce- dures of the FAGMO(1,1,k) are presented in detail, along with the transforms of its optimal parameters. A stochastic testing scheme is provided to validate the accuracy and properties of the optimal parameters of the FAGMO(1,1,k). Finally, this model is used to forecast China’s nuclear energy consumption and the results demonstrate that the FAGMO(1,1,k) model provides accurate pre- diction, outperforming other grey models. Keywords: Nuclear energy consumption, Grey system, Fractional order https://doi.org/10.1016/j.energy.2018.09.155 Corresponding author: Xin Ma Email address: cauchy7203@gmail.com (Xin Ma ) arXiv:1809.10882v1 [stat.AP] 28 Sep 2018 accumulation, Optimised parameter, Energy forecasting 1. Introduction Energy is the most important strategic resource and provides a key material basis for economic development and social progress. Energy consumption pre- diction constitutes an important aspect of energy policies for countries globally, particularly developing countries such as China, where the energy consumption structure is changing at a rapid speed. Numerous models have been intro- duced for forecasting energy consumption, such as dynamic causality analysis [1], nonlinear and asymmetric analysis [2], time-series analysis [3, 4], machine learning models [5], the coupling mathematical model [6, 7, 8], autoregressive distributed lag model [9], hybrid forecasting system [10, 11], machining system [12], fuzzy systems [13], LEAP model [14, 15], TIMES model [16, 17], NEMS model [18, 19] and grey model [20, 21, 22, 23, 24, 25, 26, 27, 28]. Among these prevalent methods, simple linear regression, multivariate linear regression, and time-series analysis are often significant in accurately demonstrating the phe- nomena of long-term trends. However, these exhibit the limitations of requiring a large amount of observed data, at least 50 or more sets, to construct mod- els. The computational intelligence method requires a substantial amount of training data to derive the optimised parameters. However, in many practical situations, it is very difficult and sometimes even impossible to obtain com- plete information. Therefore, it is important to identify a favourable method for forecasting the trend of an analysed system using scarce information with less errors. The grey forecasting theory, proposed by Professor Deng [29], offers a fea- sible and efficient method for dealing with uncertain problems containing poor information. The main advantage of this theory is that only four or more sam- ples are required to describe the behaviour and evolution of the analysed system. In Deng’s pioneering work, the first-order one variable grey model GM(1,1) was discussed in detail. Over three decades of development, the classical contin- 2 uous GM(1,1) model has been studied extensively; for example, by Xie et al. [30, 31, 32], Wang et al. [33, 34, 35], Ma et al. [36, 37, 38, 39] and others. However, we note that these generalised grey models all include integer-order accumulation, which results in less flexibility in time-series forecasting. Thus, the fractional-order accumulation grey model is considered in this paper. By extending the integer accumulated generating operation into the frac- tional accumulated generating operation, Wu et al. [40] first proposed the frac- tional accumulation GM(1,1) model known as the FAGM(1,1) model. The com- putational results demonstrated that the novel model outperformed the conven- tional GM(1,1) model. Later, Wu and his peers successfully applied fractional accumulation to the fuel production of China [41], tourism demand [42] and electricity consumption [43]. Subsequently, Xiao et al. [44] studied the GM(1,1) model, in which they regarded the fractional accumulated generator matrix as a type of generalised accumulated generating operation. Gao et al. [45] presented a new discrete fractional accumulation GM(1,1) model known as FAGM(1,1,D) and applied it to China’s CO emissions. Mao et al. [46] investigated a novel fractional grey model FGM(q,1). Interested readers can refer to [47, 48, 49, 50] for further details on fractional accumulation grey models. A further significant issue in grey system theory is that the solution applied for prediction does not match the grey difference equation. In 2009, Kong and Wei [51] proposed a parameter optimisation technique to study the DGM(2,1) model. Later, Chen et al. used a similar technique to improve the GM(1,1) [52] and ONGM(1,1) models [53], in which the basic structure of the original models remain in the optimised ones. Recently, Ma and Liu [54] studied the exact non- homogeneous grey prediction model (ENGM) with an exact basic equation and background value. Thereafter, Ma and Liu [55] considered the GMC(1,n) model with optimised parameters and applied it to forecasting the urban consumption per capita and industrial power consumption of China. Following the concept of fractional accumulation and the parameter optimisation method, we propose a novel FAGMO(1,1,k) model. In this paper, we study the nuclear energy consumption of China by means 3 of the FAGMO(1,1,k) model. The computational results indicate that the proposed grey model outperforms the existing ENGM model, optimised non- homogeneous grey model abbreviated as the ONGM(1,1,k) model, FAGM(1,1) model and FAGM(1,1,k) model. The main contributions of our paper are listed below. 1) A fractional accumulation grey model with optimised parameters is developed. 2) Detailed properties of optimised parameters are studied ac- cording to two theorems. These indicate that the first parameter is the most important factor affecting the accuracy of the FAGM(1,1,k) model. 3) Simula- tion results and two practical cases are considered to assess the effectiveness of the FAGMO(1,1,k) model compared to other models. 4) The FAGMO(1,1,k) grey forecasting model is implemented to forecast the nuclear energy consump- tion of China. It is demonstrated in the results that the newly proposed model offers higher precision than other grey models. The remainder of this paper is organised as follows. Section 2 provides a compendium of China’s energy consumption. Section 3 discusses several prelim- inaries. A detailed discussion of the FAGM(1,1,k) model is provided in section 4. Section 5 discusses the optimised parameters. Modelling evaluation criteria and detailed steps are provided in section 6. Section 7 discusses the valida- tion of the FAGMO(1,1,k) model. Applications are explained in section 8 and conclusions are drawn in the final section. 2. Brief overview of China’s energy consumption This section presents a systematic and comprehensive investigation of China’s energy consumption using five fuels, namely coal, oil, natural gas, nuclear energy and renewables. In China, renewables include hydroelectricity, wind, solar, geothermal, biomass and others. According to the statistical data of British Petroleum (BP) Statistical Review of World Energy 2018 (www.bp.com/ statisticalreview), the International Energy Agency (IEA) World Energy Out- look 2017 (www.iea.org/weo2017), Asia-Pacific Economic Cooperation (APEC) Energy Overview 2017 (www.apec.org/Publications), and National Bureau of 4 Statistics of China (NBS) China Statistical Yearbook 2017 (www.stats.gov.cn/tjsj/ndsj), China’s primary energy consumption increased from 142.9 million tonnes oil equivalent (Mtoe) in the first year of the third Five-Year Plan (1966 to 1970) th to 3132.2 Mtoe in the second year of the 13 Five-Year Plan (2016 to 2020), and increased dramatically since the turn of the millennium owing to continu- ous economic growth. According to the statistical data of BP, China’s primary energy consumption from 1966 to 2017 is plotted in Fig. 1. 1000 the third Five-Year Plan (1966-1970) Figure 1: Total primary energy consumption of China from 1966 to 2017 It is well known that China is the world’s largest energy consumer, account- ing for 23% and 23.2% of the global energy consumption in 2016 and 2017, respectively. While coal remains the dominant fuel, its share of total energy th consumption was 62% in 2016 and 60.4% in 2017. China’s 13 Five-Year Plan set an ambitious target for adjusting the primary energy consumption structure. th The energy plan set by China for the 13 Five-Year Plan can met the adjust- ment target of the primary energy consumption structure. A brief overview of China’s primary energy consumption from the perspective of five fuel types is provided below. total primary energy consumption, Mtoe 2.1. Coal Since the foundation of the People’s Republic of China, coal has always been the primary energy fuel, owing to abundant domestic reserves and its low cost [15]. From Figs. 2 and 3, it can be observed that coal soared from 122.4 Mtoe in 1966 to 1892.6 Mtoe in 2017, although the percentage of coal in the total primary energy consumption decreased from 85.7% in 1966 to 60.4% in 2017. Specifically, despite a continuous increase in coal consumption during the third and fourth Five-Year Plan periods, the proportion of coal in the total primary energy consumption has decreased from 85.7% to 72.5%. Following China’s Reform and Opening-Up Policy in 1978, the coal consumption has expanded rapidly from 282.8 Mtoe in 1978 to 1685.8 Mtoe in 2010, while the share of coal in the total primary energy consumption is around at 73.8%. At the beginning th of the 12 Five-Year Plan (2011 to 2015), the Chinese government has been stepping up its efforts to reduce coal consumption to deal with air pollution and climate change. The “supply-side reform” removes unnecessary and out-dated production capacity to avoid supply overcapacity in the coal mining industry. During this period, the consumption of coal decreased from 1903.9 Mtoe in 2011 to 1892.6 Mtoe in 2017, while the share of coal decreased from 70.8% to 60.4%. 2.2. Oil Oil is a major component of primary energy resources globally and plays a strategic role in economic growth. It can be observed in Figs. 2 and 3 that the oil consumption in China increased from 14.3 Mtoe in 1966 to 608.4 Mtoe in 2017, with an average annual growth rate of 7.5%. Owing to China’s oil reserves accounting for only 2% of the global amount, China is highly dependent on overseas oil imports of more than 60%. China became a net importer of crude oil in 1993 and the world’s second largest oil consumer in 2002 [15]. In April 2015, China surpassed the US as the world’s largest oil importer, with imports of 7.4 million barrels per day (Mbbl/D), thereby exceeding the US imports of th 7.2 Mbbl/D. Following the 11 Five-Year Plan, the oil consumption increased rapidly owing to economic growth and the improved quality of life. 6 coal oil natural gas nuclear renewables 1000 the third Five-Year Plan (1966-1970) Figure 2: China’s primary energy consumption under fuel types 0.9 0.8 0.7 0.6 0.5 coal 0.4 oil natural gas 0.3 nuclear renewables 0.2 0.1 Figure 3: Percentage of China’s primary energy consumption under fuel types different fuel types of energy consumption,Mtoe percentage of different fuel types 2.3. Natural gas Natural gas is a fossil fuel for electricity generation, chemical feedstock, heating and cooking, among others. Chinese organisations have estimated that the technically and ultimately recoverable resources of natural gas are 6.1 trillion cubic meters (tcm) and 37 tcm [56], respectively. However, natural gas has not become a major energy resource in China because the domestic natural gas industry has developed slowly. In recent years, the Chinese government has set the stable natural gas supply as one of the country’s energy strategies and encourages gas transportation from areas with significant resources to East China. The National Development and Reform Commission constructed three west-east gas pipelines in 2004, 2007 and 2015, respectively. Furthermore, the “shifting from coal to gas” policy has a significant impact on the natural gas market. The natural gas consumption has increased from 116.2 Mtoe in 2011 to 206.7 Mtoe in 2017, with an average annual growth rate of 8.6%. 2.4. Nuclear energy Nuclear energy is almost always used to generate electricity. To reduce the air pollution from coal-fired power plants, nuclear energy is an inevitable strategic option for China. In fact, China began to develop nuclear energy in the 1980s and the Qinshan Nuclear Power Plant began operating in 1991. In 2012, the State Council set a goal of 58 GW nuclear capacity by 2020. At th the beginning of the 13 Five-Year Plan, 38 nuclear power reactors were in operation with a production of 213.3 TWh, while 19 nuclear power reactors were under construction. At present, the Chinese government focuses on fourth- th generation reactors with increased safety. From the 11 Five-Year Plan (2006 to2010), nuclear energy consumption has soared rapidly from 12.4 Mtoe in 2006 to 56.2 Mtoe in 2017, with an average annual growth rate of 13.4%. 2.5. Renewables China’s renewable energy has been expanding rapidly in recent decades, owing to the development of the modern renewable energy industry. In 2017, 8 China’s renewables consumption accounted for 21.9% of the total global amount, increasing by 31% and accounting for 36% of the global renewables consumption growth. Meanwhile, the renewables consumption increased from 101.1 Mtoe in 2006 to 368.3 Mtoe in 2017; the share has increased from 5.1% to 11.8% with an th average annual growth rate of 11.4%. The 13 Five-Year Plan set targets for an installed wind power generation capacity of 250 GW, solar power generation capacity of 110 GW, and hydropower generation capacity of 350 GW by 2020. In summary, China’s primary energy consumption using five fuels for the period of 1966 to 2017 can be provided below. The coal consumption has grad- ually declined, the oil consumption has gradually increased, and the natural gas, nuclear energy and renewables have rapidly increased. China’s primary energy consumption structure exhibits a diversified trend, and the clean energy has increased yearly. 3. Definitions and properties of fractional accumulation This section provides the fractional accumulated generating operation (AGO), which can reduce the randomness of raw data in grey theory. Correspondingly, the inverse operation of accumulated generation is known as the inverse accumu- th th lated generating operation (IAGO). The r AGO and r IAGO are provided below, which can be found in paper [40, 46]. (0) (0) (0) (0) Definition 1. Let X = x (1), x (2), . . . , x (n) be an original sequence (r) th and X (r > 0) be the r accumulated generating operation (r-AGO) sequence (0) (r) (r−1) r of X , where x (k) = x (i), k = 1, 2, . . . , n. Denote by A the r-AGO i=1 (r) (0) r matrix that satisfies X = X A , and   r r r r [ ] [ ] [ ] ··· 0 1 2 n−1     r r r  0 [ ] [ ] ···  0 1 n−2     r r r A =  0 0 [ ] ···  , 0 n−3    . . . .  . . . . .   . . . .   0 0 0 ··· [ ] n×n r(r+1)···(r+i−1) r+i−1 (r+i−1)! 0 r 0 0 with [ ] = = = , = 0, = = 1. i i 0 i! i i!(r−1)! 0 (1) (0) Obviously, the 1-AGO sequence x (k) = x (i), k = 1, 2, . . . , n, namely i=1   1 1 1 ··· 1     0 1 1 ··· 1       (1) (0) X = X A with A = 0 0 1 ··· 1 .     . . . .  .  . . . . .  .  . . . .   0 0 0 ··· 1 n×n (r−1) Definition 2. The inverse accumulated generation is defined as x (k) = (r) (r) r th x (k) − x (k − 1), k = 1, 2, . . . , n. Denote by D the r inverse accumulated (0) (r) r generating operation (r-IAGO) matrix, which satisfies X = X D , and   −r −r −r −r ··· 0 1 2 n−1     −r −r −r  0 ···  0 1 n−2     −r −r D =  0 0 ···  , 0 n−3    . . . .  . . . . .   . . . .   −r 0 0 0 ··· n×n −r −r(−r+1)···(−r+i−1) i r(r−1)···(r−i+1) i r −r with = =(−1) =(−1) , =0, i> r. i i i! i! i (0) (1) (1) Similarly, the 1-IAGO sequence x (k) = x (k)−x (k−1), k = 1, 2, . . . , n;   1 −1 0 ··· 0     0 1 −1 ··· 0       (0) (1) that is, X = X D, with D = 0 0 1 ··· 0   .    . . . .  . . . . .   . . . .   0 0 0 ··· 1 n×n r + + Theorem 1. The expression [ ] , r ∈ R , i ∈ N is a function of r and i; for any value i, r ∈ (0, 1) , [ ] is a monotonically decreasing function of i; r = 1, [ ] ≡ 1; and r ∈ (1, +∞) , [ ] is a monotonically increasing function of i. 10 Proof 1. We consider the difference (r + i − 1)! (r + i − 2)! r r r+i−1 r+i−2 [ ] − = − = − i i−1 i i−1 i! (r − 1)! (i − 1)! (r − 1)! (r + i − 2)! r + i − 1 = − 1 (i − 1)! (r − 1)! i (r + i − 2)! r − 1 (i − 1)! (r − 1)! i (r + i − 2)! = (r − 1). i! (r − 1)! From the difference results, we complete the proof. To gain an improved understanding of Theorem 1, two figures are displayed in the following Fig. 4 r=1.00 10 r=1.35 r=1.00 0.9 r=1.45 r=0.35 9 r=1.55 0.8 r=0.45 r=1.65 r=0.55 r=1.75 0.7 r=0.65 r=0.75 0.6 0.5 0.4 0.3 0.2 0.1 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 i i Figure 4: Function versus values r and i: left r ∈ (0, 1), right r ∈ (1, +∞) (r) (0) r It follows from X = X A that k k−1 X X (r) r (0) r (0) x (k) = x (i) = [ ]x (k − i), (1) k−i i i=1 i=0 (r) (0) which means that x (k) is the weight of x (i), i = 1, 2, . . . , k. From Theorem 1, when r ∈ (0, 1), the weight of the old data is smaller than that of the new data. When r = 1, the weights of the old and new data are all 1. When r ∈ (1, +∞), the weight of the old data is larger than that of the new data. the values of [ ] the values of [ ] i −1 r r r r Theorem 2. The values of r-AGO A and r-IAGO D satisfy (A ) = D . Proof 2. From the definition of A , it is easy to calculate the determinant r r det (A ) = 1, which means that A is reversible. Employing mathematical induction, when r = 1, we obtain    1 1 1 ··· 1 1 −1 0 ··· 0       0 1 1 ··· 1 0 1 −1 ··· 0          AD =  0 0 1 ··· 1  0 0 1 ··· 0  = I.     . . . .  . . . .  . . . . . . . . . . . .    . . . . . . . . . .    0 0 0 ··· 1 0 0 0 ··· 1 Assuming that the properties hold true when r = m, this means that    −m −m −m −m m m m m [ ] [ ] [ ] ··· ··· 0 1 2 n−1 0 1 2 n−1       −m −m −m m m m 0 [ ] [ ] ··· 0 ···    0 1 n−2 0 1 n−2       −m −m m m 0 0 [ ] ··· 0 0 ··· =I.    0 n−3 0 n−3     . . . .  . . . .  . . . . . . . . . . . .    . . . . . . . . . .    −m 0 0 0 ··· [ ] 0 0 0 ··· 0 0 Then, when r = m + 1, we obtain m+1 m+1 m m m m m m A D = A (AD)D = A ID = A D = I, −1 r r so the result (A ) = D is proven. 4. Fractional grey FAGM(1,1,k) model Definition 3. The first-order differential equation (r) dx (t) (r) + ax (t) = bt + c, r > 0 (2) dt is known as the whitening differential equation of the FAGM(1,1,k) model. The parameter a is a development coefficient, while bt+c is the grey action quantity. The discrete differential equation 2k − 1 (r−1) (r) x (k) + az (k) = b + c (3) 12 (r−1) (r) is referred to as the basic equation of the FAGM(1,1,k). x (k) = x (k) − (r) (r) (r) (r) x (k − 1), z (k) = 0.5 x (k − 1) + x (k) . The least-squares estimation for φ = (a, b, c) of the FAGM(1,1,k) model satisfies −1 T T φ = B B B Y, (4) where     (r) (r−1) −z (2) 1 x (2)         (r) (r−1) −z (3) 1 x (3)         B = , Y = ,  . . .   .  . . . .     . . . .     (r) 2ν−1 (r−1) −z (ν) 1 x (ν) in which ν is the number of samples used to construct the model. Theorem 3. The time response function of the FAGM(1,1,k) model is b b c b b c (r) (0) −a(k−1) xˆ (k) = x (1) − + − e + k − + , k = 2, 3, . . . , n, (5) 2 2 a a a a a a (0) and the restored value of xˆ (k) k = 2, 3, . . . , n can be expressed by (0) (r) r ˆ ˆ X = X D . (6) Proof 3. From Eq. (2), we have (r) dx (t) (r) = −ax (t) + bt + c. (7) dt (r) Let u(t) = −ax (t) + bt + c; then, Eq. (7) is transformed into (r) du(t) dx (t) = −a + b = −au(t) + b. (8) dt dt To perform the indefinite integral on Eq. (8) and reduce it, we obtain (r) −at κ −a −ax (t) + bt + c + b = e e , (9) where e is a constant to be determined. (r) (0) Substituting t = 1 and x (t) = x (1) into Eq. (9), we obtain t=1 κ a 2 (0) e = e a x (1) − ab − ac + b . (10) It follows from Eqs. (9) and (10) that b b c b b c (r) (0) −a(t−1) x (t) = x (1) − + − e + t − + . (11) 2 2 a a a a a a Thus, the time response function of the FAGM(1,1,k) model is b b c b b c (r) (0) −a(k−1) xˆ (k) = x (1) − + − e + k − + , k = 2, 3, . . . , n, 2 2 a a a a a a (0) and the restored value of xˆ (k) k = 2, 3, . . . , n can be expressed by (0) (0) (0) (r) (r) (r) r xˆ (1), xˆ (2), . . . , xˆ (n) = xˆ (1), xˆ (2), . . . , xˆ (n) D . Setting b = 0 in Eq. (2), the fractional FAGM(1,1,k) model is reduced to the fractional FAGM(1,1) model [40] with the form (r) dx (t) (r) + aX (t) = c. (12) dt Setting r = 1 in Eq. (2), the fractional FAGM(1,1,k) model is reduced to the GM(1,1,k,c) model [53] with the form (1) dx (t) (1) + ax (t) = bt + c. (13) dt Setting r = 1, c = 0 in Eq. (2), the fractional FAGM(1,1,k) model is reduced to the GM(1,1,k) model [23] with the form (1) dx (t) (1) + ax (t) = bt. (14) dt Setting r = 1, b = 0, c = 1 in Eq. (2), the fractional FAGM(1,1,k) model is reduced to the GM(1,1) model [29] with the form (1) dx (t) (1) + ax (t) = c. (15) dt Thereafter, the flaw of the FAGM(1,1,k) model is provided. Integrating both sides of Eq. (2) in the interval [k − 1, k], we obtain Z Z Z Z k k k k (r) (r) dx (t) + ax (t)dt = btdt + cdt. (16) k−1 k−1 k−1 k−1 14 R (r) (r) (r) (r−1) With the knowledge of dx (t) = x (k) − x (k − 1) = x (k), k−1 R R k k tdt = (2k − 1)/2 and dt = 1, the exact discrete differential equation k−1 k−1 is expressed by (2k − 1) b (r−1) (r) x (k) + a x (t)dt = + c. (17) k−1 A comparison between Eq. (17) and the basic Eq. (3) indicates that differ- (r) (r−1) (r) ences exist in the background value z (k) = 0.5 x (k − 1) + x (k) and (r) x (t)dt. It is highly inaccurate to compute the integration utilising the k−1 (r) trapezoid formula if x (t) is not a linear function. Thus, the basic form and whitenisation differential equation of the FAGM(1,1,k) model do not strictly match. 5. Parameter optimisation of FAGM(1,1,k) model It can easily be verified that the parameters φ = (a, b, c), derived by the least-squares estimation in Eq. (3) and the parameters of the time response (r) function xˆ (k), k = 2, 3, . . . , n derived by Eq. (2), have different meanings. When the response function dose not satisfy the basic equation, large errors may arise. To match the basic Eq. (3) and response function (5), the system parameters are optimised in this system. Setting the optimised parameters of the grey system as (α, β, γ) and replac- ing the parameters φ = (a, b, c) in Eq. (2), the whitening differential equation is rewritten as (r) dx (t) (r) + αx (t) = βt + γ, r > 0. (18) dt Similarly, the general solution of Eq. (18) is given by β β γ β β γ (r) (0) −α(t−1) x (t) = x (1) − + − e + t − + . (19) 2 2 α α α α α α Furthermore, we have β β γ β β γ (r) (0) −α(k−1) xˆ (k)= x (1) − + − e + k − + , k = 2, 3, . . . , n. (20) 2 2 α α α α α α 15 Substituting Eq. (20) into the left side of Eq. (3), we obtain (r) (r) (r−1) (r) L(t) = x (k) − x (k − 1) + x (k − 1) + x (k) a a (r) (r−1) = 1 + x (k) − 1 − x (k − 1) 2 2 a β β γ β β γ (0) −α(k−1) = 1 + x (1) − + − e + k − + 2 2 2 α α α α α α a β β γ β β γ (0) −α(k−2) − 1 − x (1) − + − e + (k − 1) − + 2 2 2 α α α α α α h    i a a β β γ α (0) −α(k−1) = 1 + − 1 − e x (1) − + − e 2 2 α α α β a β β γ + ak + 1 − − − a. (21) α 2 α α α Owing to the left side L(t) and right side R(t) equivalence, namely L(t) − R(t) = 0, it is implied that a a 1 + − 1 − e = 0, (22) 2 2 a = b, (23) a β β γ b 1 − − − a = c − . (24) 2 α α α 2 It follows from Eqs. (22) to (24) that 2 + a α = ln , (25) 2 − a b 2 + a β = ln , (26) a 2 − a αc αb β β β γ = − + + − . (27) a 2a α 2 a Thus, the optimised parameters (α, β, γ) are obtained by Eqs. (25) to (27), and they also indicate that the parameters (a, b, c) derived by the least- squares estimation satisfy the relationship in Eqs. (25) to (27). In this pa- per, the FAGM(1,1,k) model with optimised parameters is referred to as the FAGMO(1,1,k) model. (0) (0) (0) Theorem 4. Assuming that the original data x (1), x (2), . . . , x (n) sat- isfy Eq. (20) with the given parameters αˆ, β, γˆ , the parameters (α, β, γ) of 16 FAGMO(1,1,k) obtained by Eqs. (4) and (25) to (27) satisfy the relationship (0) (0) (0) αˆ = α, β = β, γˆ = γ, and the predicted values xˆ (1), xˆ (2), . . . , xˆ (n) are (0) (0) (0) equal to the given data x (1), x (2), . . . , x (n) . (0) Proof 4. Substituting the original data x (k), k = 1, 2, . . . , n into Eq. (4), the parameters αˆ, β, γˆ can be derived. The parameters (α, β, γ) of FAGMO(1,1,k) can be obtained from Eqs. (25) to (27). Obviously, αˆ = α, β = β, γˆ = γ. (0) Therefore, the predicted values xˆ (k), k = 1, 2, . . . , n of the FAGMO(1,1,k) are (0) equal to the given data x (k), k = 1, 2, . . . , n. Theorem 4 demonstrates that the FAGMO(1,1,k) model is accurate for predicting arbitrary sequences that can be modelled by Eq. (20), while the FAGM(1,1,k) model cannot describe the sequences accurately owing to there always being a non-zero difference between the real parameters and the param- eters (a, b, c). Theorem 5. The optimised parameters (α, β, γ) are approximately equivalent to the parameters (a, b, c) when the value of |a| is very small; that is, α ≈ a, β ≈ b, γ ≈ c. (28) Proof 5. We first consider the difference between parameter α and a, which is 2 + a ε (a) = α − a = ln − a. (29) 2 − a It is known that ε (a)| = 0 and the first-order derivative is a=0 2+a d ln − a 2 dε (a) 2−a 4 a = = − 1 = . (30) 2 2 da da 4 − a 4 − a When |a| < 2, the derivative of ε (a) is positive, which indicates that the function ε (a) is a monotonically increasing function in the interval [-2,2]. Thus, the value ε (a) approaches zero as |a| decreases. Therefore, α ≈ a when |a| is very small. Secondly, the difference between β and b is expressed as b 2 + a b ε (a) = β − b = ln − b = ε (a) . (31) 2 1 a 2 − a a 17 1 2+a 2−a 2−a+2+a 4 Owing to lim ln = lim = lim = 1, we know that 2 2 a 2−a 2+a (2−a) 4−a a→0 a→0 a→0 ε (a) → 0 when a → 0. The first-order derivative of ε (a) is dε (a) b 4a 2 + a = − ln , (32) 2 2 da a 4 − a 2 − a which is also positive when |a| < 2. Thus, ε (a) decreases when the value of |a| decreases and β ≈ b when |a| is very small. Thirdly, the difference between γ and c is expressed as αc αb β β β ε (a) b ε (a) = γ − c = − + + − − c = c − . (33) a 2a α 2 a a a It follows from α ≈ a and β ≈ b that γ ≈ c when |a| is very small. From Theorem 5, we know that the differences between the parameters (α, β, γ) and (a, b, c) are decrease along with smaller |a|. Table 1 provides the values of ε (a) and ε (a)/a under different values of |a|. 1 1 Table 1: Values of ε (a) and ε (a) /a under different values of |a| 1 1 |a| 0.1 0.2 0.3 0.5 0.7 1.0 1.3 1.6 1.9 ε (a) 0.0001 0.0007 0.0023 0.0108 0.0309 0.0986 0.2506 0.5972 1.7636 ε (a) /a 0.0008 0.0034 0.0076 0.0217 0.0441 0.0986 0.1928 0.3733 0.9282 6. Modelling evaluation criteria and detailed modelling steps To evaluate forecasting accuracy of the FAGMO(1,1,k) model, the root mean squared percentage error (RMSPE) is applied to the prior-sample period (RM- SPEPR) and post-sample period (RMSPEPO). In general, the RMSPEPR, RM- SPEPO and RMSPE are defined as ν (0) (0) u X 1 xˆ (k) − x (k) t 1 1 RMSPEPR = × 100%, (34) (0) x (k) k=1 1 (0) (0) u X 1 xˆ (k) − x (k) t 1 1 RMSPEPO = × 100%, (35) (0) n − ν x (k) k=ν+1 1 (0) (0) 1 xˆ (k) − x (k) t 1 1 RMSPE = × 100%, (36) (0) x (k) k=1 1 18 where ν is the number of samples used to construct the model and n is the total number of samples. The index of agreement of the forecasting results is defined as (0) (0) xˆ (k) − x (k) k=1 IA = 1 − , (37) (0) (0) xˆ (k) − x + x (k) − x k=1 which is also a useful performance measure for sensitivity to differences in the observed and predicted data, where x is the average sample value. The average forecasting error (AE) and the mean absolute forecasting error (MAE) are (0) (0) AE = xˆ (k) − x (k) , (38) k=1 (0) (0) MAE = xˆ (k) − x (k) , (39) k=1 where AE reflects the positive and negative errors between the predicted and ob- served values, while MAE is applied for estimating the change in the forecasting model. The detailed modelling steps of the fractional FAGMO(1,1,k) are provided below. (0) Step 1: Determine the original data series x (i), i = 1, 2, . . . , n, and r- (r) (0) r AGO series X = X A . Step 2: Calculate the matrices B and Y to determine (a, b, c) using Eq. (4). Step 3: Compute the parameters (α, β, γ) by employing Eqs. (25) to (27). (0) Step 4: Substitute the values of x (1) and (α, β, γ) into Eq. (20) to (r) compute the predicted values X . (0) Step 5: Apply the r-IAGO matrix to obtain the restored values X = (r) r X D . 7. Validation of FAGMO(1,1,k) model This section provides numerical examples to validate the accuracy of the FAGMO(1,1,k) model compared to the FAGM(1,1,k) model and others. 19 7.1. Validation of FAGMO(1,1,k) and FAGM(1,1,k) models This subsection presents a numerical example to validate the accuracy of the FAGMO(1,1,k) and FAGM(1,1,k) models. The values r and α are provided (r) in the interval [0.01, 2] and [-1.99, 1.99], respectively. The initial point x (1) is randomly generated in the interval [1, 2] by the uniform distribution, while the parameters β and γ are randomly generated in the intervals [0, 5] and [0, (r) 100], respectively, by the uniform distribution. The other x (i) (i > 1) are generated with the aid of Eq. (20). All data used for the example are explained in Fig. 5. given in an interval randomly generated ✁✕ ✁✕ ▼❇ ❑❆ ✁ ❇ (r) ❇ dx (t) (r) + αx (t) = βt + γ dt (r) x (1) randomly generated (r) x (i) calculated by Eq.(20) Figure 5: Diagram of data for validation We define the notation in the following analysis 2 2 2 ε = (p − α) + (l − β) + (q − γ) , (40) params where (α, β, γ) are the provided parameters of Eq. (20) and (p, l, q) are the estimated parameters of the FAGM(1,1,k) or FAGMO(1,1,k) model. When applying the above parameters, the graphs are displayed in Figs. 6 and 7. We observe from Fig. 6 that the maximum ε of FAGMO(1,1,k) and params −5 FAGM(1,1,k) are 5.4228×10 and 489.9434, respectively, where the magnitude is approximately 9034932. Furthermore, the ε of the FAGM(1,1,k) model params is very small when α is near zero, which is coincident with Theorem 5. From Fig. 20 7, the maximum RMSPEs of FAGMO(1,1,k) and FAGM(1,1,k) are 0.0103% and 814.3864%, respectively, where the magnitude is approximately 79100. (r) It is known that the parameters β, γ, and initial points x (1) are all ran- (r) domly generated, which implies that the values of parameters β, γ and x (1) have no influence on the output series. Here, the values r and α are the most important factors affecting the accuracy of the grey models. −5 x 10 6 500 0 0 2 2 1 2 1 2 1.5 1.5 0 0 1 1 −1 −1 0.5 0.5 −2 −2 0 0 α α r r Figure 6: Values of ε of FAGMO(1,1,k) (left) and FAGM(1,1,k) (right) models params 0.012 1000 0.01 0.008 0.006 0.004 0.002 0 0 2 2 1 2 1 2 1.5 1.5 0 0 1 1 −1 −1 0.5 0.5 −2 0 −2 0 α α r r Figure 7: Values of RMSPE of FAGMO(1,1,k) (left) and FAGM(1,1,k) (right) models 7.2. Validation of FAGMO(1,1,k) model and other grey models This subsection further demonstrates the advantage of the FAGMO(1,1,k) model using two real cases. RMSPE params RMSPE ε params Case 1: (Predicting cumulative oil field production). We consider an ex- ample from the paper [54] that provides sample data. The data from 1999 to 2009 are applied to construct the grey model, while the data from 2010 to 2012 are used for prediction. The values are listed in Table 2, indicating that the FAGMO(1,1,k) model outperforms the other models in this case. Table 2: Results of ENGM, FAGM(1,1), FAGM(1,1,k) and FAGMO(1,1,k) models Year Data ENGM FAGM(1,1) FAGM(1,1,k) FAGMO(1,1,k) r = 1 r = 0.1106 r = 0.4073 r = 0.4052 1999 73.8217 73.8217 73.8217 73.8217 73.8217 2000 136.8817 138.4900 138.1621 137.1758 136.4573 2001 195.0590 195.4541 195.5377 196.1598 195.7633 2002 247.8547 247.9776 247.7638 249.3183 249.1781 2003 297.0902 296.4067 295.7629 297.2895 297.2750 2004 342.6394 341.0604 340.1238 341.0008 341.0322 2005 382.4312 382.2332 381.2700 381.2882 381.3320 2006 420.0399 420.1964 419.5291 418.8204 418.8699 2007 454.0430 455.2001 455.1670 454.1099 454.1712 2008 485.1171 487.4752 488.4068 487.5452 487.6290 2009 519.8508 517.2342 519.4402 519.4217 519.5393 2010 552.6569 544.6734 548.4350 549.9665 550.1281 2011 581.6092 569.9736 575.5400 579.3572 579.5714 2012 608.1863 593.3015 600.8887 607.7346 608.0086 RMSPEPR 0.4521% 0.4582% 0.3539% 0.3259% RMSPEPO 2.0066% 1.0185% 0.3617% 0.3332% Case 2: (Predicting foundation settlement close neighbouring Yangtze River). We consider an example from the paper [53], which provides sample data to con- struct the grey model. The values are presented in Table 3, indicating that the FAGMO(1,1,k) model outperforms the other models in this case. 22 Table 3: Results of ONGM(1,1,k,c), FAGM(1,1), FAGM(1,1,k) and FAGMO(1,1,k) models Day Data ONGM(1,1,k,c) FAGM(1,1) FAGM(1,1,k) FAGMO(1,1,k) r = 1 r = 0.0065 r = 0.2266 r = 0.2295 10 23.36 23.3600 23.3600 23.3600 23.3600 20 43.19 42.1779 43.3517 43.0586 43.0644 30 58.73 59.2549 59.4403 58.8205 58.8124 40 70.87 72.8374 72.4009 71.9763 71.9545 50 83.71 83.6405 82.8451 83.0247 82.9932 60 92.91 92.2330 91.2620 92.2158 92.1789 70 99.73 99.0672 98.0442 99.6885 99.6491 80 105.08 104.5030 103.5079 105.5215 105.4805 90 109.73 108.8264 107.9079 109.7568 109.7127 100 112.19 112.2652 111.4497 112.4117 112.3598 110 113.45 115.0002 114.2991 113.4857 113.4181 RMSPE 1.2730% 1.3257% 0.6030% 0.6011% 8. Applications In this section, the FAGMO(1,1,k) model is applied to forecast the nuclear energy consumption of China. The computational results of the FAGMO(1,1,k) model are compared to the ENGM [54], ONGM(1,1,k) [53], FAGM(1,1) [40] and FAGM(1,1,k) models. 8.1. Raw data Raw data of the nuclear energy consumption of China were collected from the report of the BP Statistical Review of World Energy 2018. The first 10 samples th th belonging to the 11 and the 12 Five-Year Plans are applied to construct the th prediction model, while the remaining samples of the 13 Five-Year Plan are used to validate and compare the forecasting results (see Table 4). Table 4: Raw data of nuclear energy consumption of China, Mtoe Year Data Year Data Year Data 2006 12.4 2011 19.5 2016 48.2 2007 14.1 2012 22.0 2017 56.2 2008 15.5 2013 25.3 2009 15.9 2014 30.0 2010 16.7 2015 38.6 23 8.2. Simulation and prediction results The simulation and prediction results are listed in Table 5 and Fig. 8, while the errors are listed in Table 6 and Fig. 9. The nuclear energy consumption of China from 2016 to 2017 is predicted ac- cording to the established grey models. It can be observed in Table 5 and Fig. 8 that five grey models, namely ENGM, ONGM(1,1,k), FAGM(1,1), FAGM(1,1,k) and FAGMO(1,1,k), successfully identify the trend of China’s nuclear energy consumption. However, these grey models differ from one another in terms of the prediction values from 2016 to 2020. From Fig. 8, China’s nuclear energy consumption is overestimated by the ENGM, ONGM(1,1,k) and FAGM(1,1,k) models, and underestimated by the FAGM(1,1) model. The values predicted by FAGMO(1,1,k) are substantially closer to the raw data than those predicted by the other models. We can observe from Table 6 and Fig. 9 that the RMSPEPR, RMSPEPO and RMSPE of FAGMO(1,1,k) are 3.1409%, 4.1502% and 3.3304%, respectively. The RMSPEPR, RMSPEPO and RMSPE of ENGM are as high as 8.3788%, 30.3663% and 14.5667%, those of ONGM(1,1,k) are 2.0494%, 12.0510% and 5.2635%, those of FAGM(1,1) are 4.8680%, 11.7968% and 6.5529%, and those of FAGM(1,1,k) are 2.3299%, 6.3828% and 3.3636%, respectively. The IA, AE and MAE of FAGMO(1,1,k) are 0.9985, 0.2526 and 0.7513, those of ENGM are 0.9538, 4.0536 and 4.0536, those of ONGM(1,1,k) are 0.9911, 1.0225 and 1.1896, those of FAGM(1,1) are 0.9887, -1.1818 and 1.7105, and those of FAGM(1,1,k) are 0.9971, 0.2736 and 0.8043, respectively. The computational results indicate that the FAGMO(1,1,k) model outperforms ENGM, ONGM(1,1,k), FAGM(1,1) and FAGM(1,1,k), while ENGM exhibits the most inferior performance. 24 Table 5: Simulation and prediction results of nuclear energy consumption by grey models Year Data ENGM ONGM(1,1,k) FAGM(1,1) FAGM(1,1,k) FAGMO(1,1,k) r = 1 r = 1 r = 1.4127 r = 1.0593 r = 1.1595 2006 12.4 12.4000 12.4000 12.4000 12.4000 12.4000 2007 14.1 14.9788 14.4788 15.0242 14.7054 15.0891 2008 15.5 15.6744 15.1057 13.9808 15.0121 14.8608 2009 15.9 16.6846 16.0032 15.0566 15.8012 15.5886 2010 16.7 18.1520 17.2884 16.9219 17.0700 16.9760 2011 19.5 20.2831 19.1286 19.3953 18.9344 19.0534 2012 22.0 23.3785 21.7635 22.4687 21.5861 21.9432 2013 25.3 27.8741 25.5363 26.1951 25.3029 25.8633 2014 30.0 34.4036 30.9383 30.6625 30.4740 31.1013 2015 38.6 43.8871 38.6732 35.9872 37.6390 38.0473 2016 48.2 57.6610 49.7483 42.3129 47.5433 47.2178 2017 56.2 77.6662 65.6063 49.8133 61.2149 59.2933 2018 – 106.7219 88.3125 58.6959 80.0704 75.1679 2019 – 148.9226 120.8244 69.2074 106.0614 96.0147 2020 – 210.2150 167.3766 81.6403 141.8758 123.3723 raw data ENGM ONGM(1,1,k,c) FAGM(1,1) FAGM(1,1,k) FAGMO(1,1,k) data for building models Figure 8: Comparison among five grey models for nuclear energy consumption data of nuclear energy consumption Table 6: Relative error values of nuclear energy consumption by five grey models Year ENGM ONGM(1,1,k) FAGM(1,1) FAGM(1,1,k) FAGMO(1,1,k) 2006 0 0 0 0 0 2007 0.0623 0.0269 0.0655 0.0429 0.0701 2008 0.0112 0.0254 0.0980 0.0315 0.0412 2009 0.0493 0.0065 0.0530 0.0062 0.0196 2010 0.0869 0.0352 0.0133 0.0222 0.0165 2011 0.0402 0.0190 0.0054 0.0290 0.0231 2012 0.0627 0.0108 0.0213 0.0188 0.0026 2013 0.1017 0.0093 0.0354 0.0001 0.0222 2014 0.1468 0.0313 0.0221 0.0158 0.0367 2015 0.1370 0.0019 0.0677 0.0249 0.0143 2016 0.1963 0.0321 0.1221 0.0136 0.0204 2017 0.3820 0.1674 0.1136 0.0892 0.0550 RMSPEPR 8.3788% 2.0494% 4.8680% 2.3299% 3.1409% RMSPEPO 30.3663% 12.0510% 11.7968% 6.3828% 4.1502% RMSPE 14.5667% 5.2635% 6.5529% 3.3636% 3.3304% IA 0.9538 0.9911 0.9887 0.9971 0.9985 AE 4.0536 1.0225 -1.1818 0.2736 0.2526 MAE 4.0536 1.1896 1.7105 0.8043 0.7513 0.4 ENGM ENGM ENGM ONGM(1,1,k,c) ONGM(1,1,k,c) ONGM(1,1,k,c) 30.3663 0.35 FAGM(1,1) FAGM(1,1) FAGM(1,1) FAGM(1,1,k) FAGM(1,1,k) FAGM(1,1,k) FAGMO(1,1,k) FAGMO(1,1,k) FAGMO(1,1,k) 0.3 0.25 20 RMSPEPO 0.2 14.5667 12.051 11.7967 RMSPE 0.15 0.1 6.3828 6.5529 5.2635 4.1502 3.3636 3.3304 0.05 0 0 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 1 2 3 4 5 6 7 8 9 10 year Figure 9: Errors among five grey models for nuclear energy consumption relative error 8.3. Further discussions As demonstrated by the case study, the novel FAGMO(1,1,k) model outper- forms other grey models. Moreover, it should be noted that in this paper we only conduct short-term forecasting, while it is well known that several exist- ing energy models can perform long-term forecasting, such as LEAP, TIMES and NEMS. We will discuss the difference between our model and these models further, following a very brief introduction to such models. • LEAP (long-range energy alternatives planning system) [14, 15] is a scenario- based energy environment modelling tool for climate change mitigation and energy policy analysis. It can be applied to examine energy production and consumption, as well as resource extraction in all sectors. The model studies the effects of various factors on energy consumption under different scenarios given an objective. LEAP is generally used for forecasting studies of between 20 and 50 years. • TIMES (The Integrated MARKAL-EFOM System) [16, 17] is an evolution of MARKAL, which was developed by the Energy Technology Systems Anal- ysis Programme of the IEA. It combines technical engineering and economic approaches, and uses linear programming to produce a least-cost energy system under numerous user-specified constraints. The software is used to analyse en- ergy, economic and environmental issues at different levels over several decades. • NEMS (National Energy Modeling System) [18, 19] is a long-standing US government policy model, which computes equilibrium fuel prices and quantities for the US energy sector. NEMS is used to model the demand side explicitly; in particular, to determine consumer technology choices in the residential and commercial building sectors. These models can perform long-term energy consumption projections. How- ever, they may require a large amount of data, such as population growth, GDP, urbanisation, energy policies and energy strategies. In numerous prac- tical situations, it is very difficult to obtain complete information because of time and cost limitations. The grey prediction model is an efficient method for conducting accurate forecasting with at least four samples. Compared to the 27 major energy models, the grey model is an effective choice for predicting China’s nuclear energy consumption. This paper collected 12 samples of China’s nuclear energy consumption from the BP Statistical Review of World Energy 2018. Thus, the LEAP, TIMES and NEMS models are all inapplicable owing to poor information. By employing th th the grey system theory and actual data from the 11 and the 12 Five-Year Plans, the FAGMO(1,1,k) model was constructed. It can be observed in Table 5 that the prediction value of FAGMO(1,1,k) is 123.3723 Mtoe in 2020, which is larger than the 84.6318 Mtoe provided in the BP energy outlook 2018. The main reasons for this are as follows. i) It is infeasible to consider factors such as energy policies and China’s energy strategies, which affect the current situation of China’s nuclear en- ergy consumption, in our proposed model because the FAGMO(1,1,k) model is univariate. However, the forecasting models of institutions including BP, the IEA and APEC are based on widely collected data. Furthermore, grey models are mainly used for short-term forecasting in the calculation process, such as [22, 23, 24, 27, 57]. Therefore, the forecasting results are relatively acceptable, reflecting the growth trend of future nuclear energy consumption in China. ii) In China’s nuclear energy market, 38 nuclear power reactors are in op- eration, 19 nuclear power reactors are under construction and more are to be constructed by the end of 2016. This is the reason for the increase in nuclear en- ergy consumption in recent years. However, no new nuclear projects have been approved for construction in 2016. Moreover, the State Council approved new safety rules and a nuclear power development plan following Japan’s Fukushima Daiichi crisis in 2011. These factors have also resulted in a slight slowdown in China’s nuclear energy consumption. In the future, nuclear energy could provide an important alternative to fos- sil fuels such as coal and oil, and its proportion of the total primary energy consumption will increase yearly. Based on our forecasting results using the FAGMO(1,1,k) model, the future nuclear energy consumption of China will in- crease rapidly if no certain restrictions are placed thereon. This implies that 28 higher management and technical levels are necessary to meet the safety and quality requirements. Therefore, China’s government and policy makers should pay additional attention to the safety and quality issues of nuclear energy to achieve long-term, environmentally friendly and low-carbon energy goals and lay the foundation for the sustainable development of China’s energy and economy. 9. Conclusions By applying the grey modelling technique and parameter optimisation method, the fractional FAGMO(1,1,k) model was proposed to predict China’s nuclear en- th ergy consumption of the 13 Five-Year Plan, based on the updated data from 2006 to 2015. The forecasting results provide the growth trend of the future nu- clear energy consumption of China, and also offer a guideline for policymaking and project planning. It can be observed that FAGMO(1, 1, k) is quite easy to use, with satisfac- tory accuracy in short-term nuclear consumption forecasting. For long-term prediction, its error will be larger because only 10 samples are used for mod- elling. This study is expected to be able to forecast the energy consumption of other countries that share similar patterns of economic development and energy consumption structures, among others. Furthermore, the optimised method applied to improve the FAGM(1,1,k) model can be used to improve other first- order grey models, such as NGBM(1,1), GMC(1,n) and RDGM(1,n). These are possible extensions and suggested directions for our future research. Acknowledgments This research was supported by the National Natural Science Foundation of China (No. 71771033), Longshan academic talent research supporting program of SWUST (No. 17LZXY20), Doctoral Research Foundation of Southwest Uni- versity of Science and Technology (No. 15zx7141, 16zx7140) and the Open Fund (PLN201710) of the State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation (Southwest Petroleum University). 29 References [1] F. M. Mirza, A. Kanwal, Energy consumption, carbon emissions and eco- nomic growth in Pakistan: dynamic causality analysis, Renewable & Sus- tainable Energy Reviews 72 (5) (2017) 1233–1240. [2] M. Shahbaz, T. H. V. Hoang, M. K. Mahalik, D. Roubaud, Energy con- sumption, financial development and economic growth in India: new evi- dence from a nonlinear and asymmetric analysis, Energy Economics 66 (3) (2017) 199–212. [3] H. A. Bekhet, A. Matar, T. Yasmin, CO emissions, energy consumption, economic growth, and financial development in GCC countries: dynamic simultaneous equation models, Renewable & Sustainable Energy Reviews 70 (4) (2017) 117–132. [4] M. Minda, C. Wei, Do commercial building sector-derived carbon emissions decouple from the economic growth in Tertiary Industry? A case study of four municipalities in China, Science of The Total Environment 650 (Part 1) (2019) 822–834. [5] J. Fan, X. Wang, L. Wu, H. Zhou, F. Zhang, X. Yu, X. Lu, Y. Xiang, Comparison of support vector machine and extreme gradient boosting for predicting daily global solar radiation using temperature and precipitation in humid subtropical climates: A case study in china, Energy Conversion and Management 164 (2018) 102–111. [6] Y. Wang, X. Yi, Transient pressure behavior of a fractured vertical well with a finite-conductivity fracture in triple media carbonate reservoir, Journal of Porous Media 20 (8) (2017) 707–722. [7] Y. Wang, X. Yi, Flow modeling of well test analysis for a multiple-fractured horizontal well in triple media carbonate reservoir, International Journal of Nonlinear Sciences and Numerical Simulation 19 (5) (2018) 439–457. doi:https://doi.org/10.1515/ijnsns-2016-0075. 30 [8] Y. Hu, E. Mackay, O. Vazquez, I. O, Streamline simulation of barium sul- fate precipitation occurring within the reservoir coupled with analysis of observed produced water chemistry data to aid scale management, SPE Production & Operations 33 (1) (2018) 85–101. [9] R. Brini, M. Amara, H. Jemmali, Renewable energy consumption, inter- national trade, oil price and economic growth inter-linkages: the case of Tunisia, Renewable & Sustainable Energy Reviews 76 (9) (2017) 620–627. [10] P. Du, J. Wang, W. Yang, T. Niu, Multi-step ahead forecasting in electrical power system using a hybrid forecasting system, Renewable Energy 122 (7) (2018) 533–550. [11] W. Cai, C. Liu, C. Zhang, M. Ma, W. Rao, W. Li, K. He, M. Gao, Devel- oping the ecological compensation criterion of idustrial solid waste based on emergy for sustainable development, Energy 157 (15) (2018) 940–948. [12] W. Cai, F. Liu, O. Dinolov, J. Xie, P. Liu, J. Tuo, Energy benchmarking rules in machining systems, Energy 142 (1) (2018) 258–263. [13] X. Wu, X. Ding, T. Lu, J. Wang, Topological dynamics of zadeh’s extension on upper semi-continuous fuzzy sets, International Journal of Bifurcation and Chaos 27 (10) (2017) 1–13. [14] S. P. Schnaars, How to develop and use scenarios, Long Range Planning 20 (1) (1987) 105–114. [15] K. Dong, R. Sun, H. Li, H. Jiang, A review of China’s energy consumption structure and outlook based on a long-range energy alternatives modeling tool, Petroleum Science 14 (1) (2017) 214–227. [16] J. Shi, W. Chen, X. Yin, Modelling building’s decarbonization with appli- cation of China Times model, Applied Energy 162 (15) (2016) 1303–1312. [17] H. Zhang, W. Chen, W. Huang, Times modelling of transport sector in China and USA: comparisons from a decarbonization perspective, Applied Energy 162 (15) (2016) 1505–1514. 31 [18] S. A. Gabriel, A. S. Kydes, P. Whitman, The national energy modeling system: a large-scale energy-economic equilibrium model, Operations Re- search 49 (1) (2001) 14–25. [19] R. Soroush, A. Koochi, M. Keivani, M. Abadyan, A bilayer model for in- corporating the coupled effects of surface energy and microstructure on the electromechanical stability of NEMS, International Journal of Structural Stability and Dynamics 17 (4) (2017) 1771005. [20] B. Zeng, Y. Tan, H. Xu, J. Quan, L. Wang, X. Zhou, Forecasting the electricity consumption of commercial sector in Hong Kong using a novel grey dynamic prediction model., Journal of Grey System 30 (1) (2018) 157–172. [21] B. Zeng, S. Liu, A self-adaptive intelligence gray prediction model with the optimal fractional order accumulating operator and its application, Math- ematical Methods in the Applied Sciences 40 (18) (2017) 7843–7857. [22] S. J. Feng, Y. D. Ma, Z. L. Song, J. Ying, Forecasting the energy con- sumption of China by the grey prediction model, Energy Sources Part B-economics Planning and Policy 7 (4) (2012) 376–389. [23] J. Cui, S. Liu, B. Zeng, N. Xie, A novel grey forecasting model and its optimization, Applied Mathematical Modelling 37 (6) (2013) 4399–4406. [24] H. Duan, G. R. Lei, K. Shao, Forecasting crude oil consump- tion in china using a grey prediction model with an optimal fractional-order accumulating operator, Complexity 2018 (2018) 1–12. doi:https://doi.org/10.1155/2018/3869619. [25] B. Zeng, H. Duan, Y. Bai, W. Meng, Forecasting the output of shale gas in china using an unbiased grey model and weakening buffer operator, Energy 151 (15) (2018) 238–249. 32 [26] B. Zeng, L. Chuan, Improved multi-variable grey forecasting model with a dynamic background-value coefficient and its application, Computers & Industrial Engineering 118 (4) (2018) 278–290. [27] L. Wu, Y. Gao, Y. Yang, X. Chen, Using a novel multi-variable grey model to forecast the electricity consumption of Shangdong province in China, Energy 157 (15) (2018) 327–335. [28] J. Wang, P. Du, H. Lu, W. Yang, T. Niu, An improved grey model optimized by multi-objective ant lion optimization algorithm for an- nual electricity consumption forecasting, Applied Soft Computing, doi: https://doi.org/10.1016/j.asoc.2018.07.022. [29] J. Deng, Control problems of grey systems, Systems & Control Letters 1 (5) (1982) 288–294. [30] N. Xie, S. Liu, Discrete grey forecasting model and its optimization, Ap- plied Mathematical modelling 33 (2) (2009) 1173–1186. [31] N. Xie, S. Liu, Y. Yang, C. Yuan, On novel grey forecasting model based on non-homogeneous index sequence, Applied Mathematical Modelling 37 (7) (2013) 5059–5068. [32] N. Xie, R. Wang, A historic review of grey forecasting models, Journal of grey System 29 (4) (2017) 1–29. [33] Z. Wang, D. Ye, Forecasting Chinese carbon emissions from fossil energy consumption using non-linear grey multivariable models, Journal of Cleaner Production 142 (20) (2017) 600–612. [34] Z. Wang, Q. Li, L. Pei, Grey forecasting method of quarterly hydropower production in China based on a data grouping approach, Applied Mathe- matical Modelling 51 (11) (2017) 302–316. [35] Z. X. Wang, H. H. Zheng, L. L. Pei, T. Jin, Decomposition of the factors influencing export fluctuation in China’s new energy industry based on a constant market share model, Energy Policy 109 (10) (2017) 22–35. 33 [36] X. Ma, Z. Liu, Application of a novel time-delayed polynomial grey model to predict the natural gas consumption in China, Journal of Computational & Applied Mathematics 324 (11) (2017) 17–24. [37] X. Ma, Z. Liu, Y. Wang, Application of a novel nonlinear multivari- ate grey bernoulli model to predict the tourist income of China, Jour- nal of Computational and Applied Mathematics 347 (2018) 84 – 94. doi:https://doi.org/10.1016/j.cam.2018.07.044. [38] X. Ma, Z. Liu, The kernel-based nonlinear multivariate grey model, Applied Mathematical Modelling 56 (4) (2018) 217–238. [39] X. Ma, Z. Liu, Predicting the oil production using the novel multivariate nonlinear model based on Arps decline model and kernel method, Neural Computing & Applications 29 (2) (2018) 579–591. [40] L. Wu, S. Liu, L. Yao, S. Yan, D. Liu, Grey system model with the fractional order accumulation, Communications in Nonlinear Science and Numerical Simulation 18 (7) (2013) 1775–1785. [41] L. Wu, S. Liu, W. Cui, D. Liu, T. Yao, Non-homogenous discrete grey model with fractional-order accumulation, Neural Computing & Applica- tions 25 (5) (2014) 1215–1221. [42] L. Wu, S. Liu, L. Yao, R. Xu, X. Lei, Using fractional order accumulation to reduce errors from inverse accumulated generating operator of grey model, Soft Computing 19 (2) (2015) 483–488. [43] L. Wu, S. Liu, Z. Fang, H. Xu, Properties of the GM(1,1) with fractional order accumulation, Applied Mathematics and Computation 252 (1) (2015) 287–293. [44] X. Xiao, H. Guo, S. Mao, The modeling mechanism, extension and opti- mization of grey GM (1,1) model, Applied Mathematical Modelling 38 (5-6) (2014) 1896–1910. 34 [45] M. Gao, S. Mao, X. Yan, J. Wen, Estimation of Chinese CO emission based on a discrete fractional accumulation grey model, The Journal of Grey System 27 (4) (2015) 114–130. [46] S. Mao, M. Gao, X. Xiao, M. Zhu, A novel fractional grey system model and its application, Applied Mathematical Modelling 40 (7-8) (2016) 5063– [47] Y. Shen, P. Qin, Optimization of grey model with the fractional order accumulation, Journal of Grey System 17 (3) (2014) 127–132. [48] Y. Yang, D. Xue, Continuous fractional-order grey model and electricity prediction research based on the observation error feedback, Energy 115 (1) (2016) 722–733. [49] Y. Yang, D. Xue, Modified grey model predictor design using optimal fractional-order accumulation calculus, IEEE/CAA Journal of Automat- ica Sinica 4 (4) (2017) 1–10. [50] Q. Liu, D. Yu, Non-equidistance and nonhomogeneous grey model NN- FGM(1,1) with the fractional order accumulation and its application, Jour- nal of Interdisciplinary Mathematics 20 (6-7) (2017) 1423–1426. [51] X. Kong, Y. Wei, Optimization of DGM(2,1), Journal of Grey System 12 (1) (2009) 9–13. [52] C.-S. Yu, S.-H. Chen, Applying genetic algorithm and GM(h, N) in sales forecasting, Journal of Grey System 20 (2) (2017) 109–116. [53] P. Chen, H. Yu, Foundation settlement prediction based on a novel NGM model, Mathematical Problems in Engineering 2014 (1) (2014) 1–8. [54] X. Ma, Z. Liu, Predicting the cumulative oil field production using the novel grey ENGM model, Journal of Computational & Theoretical Nanoscience 13 (1) (2016) 89–95. 35 [55] X. Ma, Z. Liu, The GMC(1, n) model with optimized parameters and its applications, Journal of Grey System 29 (4) (2017) 121–137. [56] Z. Hou, H. Xie, H. Zhou, P. Were, O. Kolditz, unconventional gas resources in china, Environmental Earth Sciences 73 (10) (2015) 5785–5789. [57] C.-I. Chen, H. Chen, S. Chen, Forecasting of foreign exchange rates of Taiwan’s major trading partners by novel nonlinear grey Bernoulli model NGBM(1, 1), Communications in Nonlinear Science and Numerical Simu- lation 13 (6) (2008) 1194–1204.

Journal

StatisticsarXiv (Cornell University)

Published: Sep 28, 2018

References