Access the full text.
Sign up today, get DeepDyve free for 14 days.
Improvement of Identification Procedure Using Hybrid Cuckoo Search Algorithm for Turbine- Governor and Excitation System Teimour Hosseinalizadeh, S. Mahmoud Salamati, S. Ali Salamati, and G. B. Gharehpetian 𝐾 Load limiter proportional gain for PI controller (%*.- Abstract— In this paper a new method is introduced in order to 𝐾 Load limiter integral gain for PI controller ,%*.- modify identification process of a gas power plant using a meta- 𝐿 Load limiter reference value -1$4 heuristic algorithm named Cuckoo Search (CS). Simulations play 𝐷𝑚 Speed sensitivity coefficient a significant role in dynamic analyses of power plants. This paper 𝐾 Power controller gain ,9: points out to a practical approach in model selection and 𝑃 Power controller set-point parameter estimation of gas power plants. The identification and 9:<$/ validation process concentrates on two subsystems: governor- 𝐾 Acceleration limiter gain turbine and exciter. Standard models GGOV1 and STB6 are 𝑇 Acceleration limiter time constant preferred for the dynamical structures of governor-turbine and 𝐼 Synchronous machine field current >? exciter respectively. Considering definite standard structure, main 𝐻𝑉 𝐺𝑎𝑡𝑒 Model block with two inputs and one output, parameters of dynamical model are pre-estimated via system the output always corresponding to the higher identification methods based on field data. Then obtained of the two inputs parameters are tuned carefully using an iterative Cuckoo 𝐿𝑉 𝐺𝑎𝑡𝑒 Model block with two inputs and one output, algorithm. Models must be validated by results derived via a trial the output always corresponding to the lower and error series of simulation in comparison to measured test data. of the two inputs The procedure gradually yields in a valid model with precise estimated parameters. Simulation results show accuracy of 𝐸 Exciter output voltage >? identified models. Besides, a whiteness analysis has been 𝐼 Exciter output current limit reference HI performed in order to show the authenticity of the proposed 𝑉 Output of terminal voltage transducer and load method in another way. Despite various detailed models, practical compensation elements. attempts of model selection, identification, and validation in a real 𝑉 Voltage regulator reference voltage IK> gas unit could rarely be found among literature. In this paper, 𝑉 Available exciter voltage Chabahar power plant in Iran, with total install capacity of 320 𝐾 Exciter output current limit adjustment MW, is chosen as a benchmark for model validation. JM 𝐾 Forward gain constant of the inner loop field Index Terms— Cuckoo Search (CS), excitation system, IEEE regulator. standard model, gas power plant, governor-turbine, model 𝐾 , 𝐾 Voltage regulator proportional and integral #O MO selection and identification, parameter estimation. gains 𝐾 Pre-control gain constant >> 𝐾 Exciter output current limiter gain HI NOMENCLATURE 𝑅 Permanent droop 𝑇 Electrical power transducer time constant #$%$& I. INTRODUCTION 𝐾 Governor proportional gain ()*+ YNAMIC performance analyses of power system have 𝐾 Governor integral gain ,)*+ been planned by engineers since 1970’s . Gas power 𝐾 Governor derivative gain -)*+ plants consist of many complicated subsystems dealing with 𝑇 Governor derivative controller time constant -)*+ high installation and maintenance expenditures. Practical 𝑇 Actuator time constant .&/ evaluation, experimental labor, and engineering costs in the 𝐾 Turbine gain /012 improvement of such systems are expected to invest large 𝑊 No load fuel flow 45% amount of money. Therefore, subsystem developments 𝑇 Turbine lag time constant sometimes are found impossible due to economic reasons. On 𝑇 Turbine lead time constant the other hand, operational limitations may yield in 𝑇 Transport time delay for diesel engine $5) technological infeasibility of new updates in control structures. 𝑇 Load limiter time constant 4%*.- Hence, a simulator tool capable of realistic approximation and Fig. 1. GGOV1 IEEE standard model. Thermal governor modeling approaches are common research transient performance analyses is of researchers’ interest. topic in the literature. However, practical model validation of Dynamic models of different parts in a gas unit comprise of presented schemes has been rarely performed. Western differential and algebraic equations derived according to basic Electricity Coordinating Council (WECC) has started an principles governing system phenomenon. important research on validation of governor-turbine dynamic Various approaches with a wide range of complexity are models. A new modeling approach is proposed and has been available in this field. Some of these approaches present a extensively validated against recording from WECC complete model of total behavior of power plant -. A benchmark systems. It is concluded based on results that deductive approach is used in  for turbine modeling as a part presented approach could be implemented in un-responsive of integrated power system model for steady state performance characteristics of frequency control simulation via past models. verification. In , a multi-temporal simulation model is Therefore, this model has been approved for use in all operation implemented in order to carry out integrated analysis of and planning studies in WECC. electricity, heat, and gas distribution. The main disadvantage of There exist different excitation system models considering the model proposed in , is the complexity of relevant coupled detailed system properties in the literature from direct current electrical, heat and gas flow equations requiring to be solved (DC) to static (ST) types; however, in a practical model simultaneously via Newton-Raphson approach. In , similar derivation and validation, facility in obtaining model attempts as  are performed for biomass power plant using parameters from field test data is very important. micro gas turbine. In , a power generating unit dynamic The paper is organized as follows: governor-turbine and exciter standard models are discussed in Section II and III, model is proposed in order to investigate boiler pressure effects, respectively. In section IV an easily powerful metaheuristic load-frequency control, boiler-turbine and coordinated control algorithm is described. System setup and proposed tuning. Different subsystems are modeled and identified identification procedure using field data are discussed in section individually via second order transfer function approach, not V. Also model training results and validations are given in this considering any special standard in model presentation. Partial section. At last, section VI concludes the paper and modeling of a special subsystem like governor-turbine , recommends some practical remarks in order to improve the exciter , and generator  are also found in the literature. GGOV1 model. By the way, modeling of integrated thermal power plants or their individual equipment could be categorized in two groups; II. GOVERNOR-TURBINE MODELING white-box and black-box models. White-box models deal with Several documents published in the last decade give various dynamic equations , while black-box models are employed modeling approaches for dynamics behavior of governor- when access to dynamic equations is impossible . Due to turbine. A hierarchy of models are presented for heavy-duty gas disadvantages like over parameterization, training complexity, turbines in . Models categorized in this standard include and lack of practicality, we utilize a white-box scheme in order GAST, the most simplistic representation of a gas-turbine, to model the subsystems of the thermal power plant. In fact, GAST2A, considering a proportional speed governor control, IEEE standard models are used in this paper and their GGOV1, developed as a general purpose of dynamic simulation parameters are identified using identification techniques. studies. Models for two main parts of a gas power plant including Fig. 1 demonstrates the IEEE standard GGOV1 governor- governor-turbine and exciter are identified and validated in this turbine model which is our chosen model for parameter work. IV. CUCKOO SEARCH ALGORITHM Metaheuristic algorithms such as Genetic and Particles Swarm Optimization (PSO) are nature based algorithms which have been used in wide range of problems . Cuckoo Search algorithm is one of these metaheuristic methods which has been shown to be effective in solving even non-polynomial time (NP) problems. One of the main advantages of this algorithm is that there are fewer parameters needed to be tuned in CS than in PSO and GA. Furthermore, it has been shown for multimodal objective function CS has better performance than GA , . Although CS has other details but pseudo code for its main form is as follows Fig. 2. ST6B IEEE Exciter standard model. CS algorithm via Lévy flights  identification. GGOV1 is flexible to model various governor Begin model and feedback signal. This is the main advantage of ( ) ( ) Objective function 𝑓 𝑥 , 𝑥 = 𝑥 , … , 𝑥 GGOV1 in comparison to GAST and GAST2A. This model is W - suitable for grid operation studies in large networks and load Produce initial population of 𝑛 host nests 𝑥 𝑖 = (1, … , 𝑛) While(𝑡<MaxGeneration) or (stop criterion) rejection analyses that were impractical in previous models. In this standard model, electrical power, governor output, or Get a Cuckoo randomly by Lévy flights Evaluate it fitness 𝐹 valve stroke can be utilized as feedback droop signal. The input of valve model is derived from a Low-Select block which Choose a nest among 𝑛 (say 𝑗) randomly selects between “FSRN”, “FSRT”, and “FSRA” signals If (𝐹 > 𝐹 ) , ` denoting speed governor, temperature, and acceleration control Replace 𝑗 by the new solution signal respectively. GGOV1 represents all fundamental end elements of a gas-turbine controller. Speed/power control A fraction (𝑝 ) of worst nests are abandoned and new ones normally is performed between about 70-100 percent of are built. nominal load. The other two control loops are active in Keep the best solutions maximum load enforcement. Detailed structure of model is Rank the solutions and find the current best explained in . Acceleration controller loop is not end while implemented in some gas power plants such as Chabahar power end plant; so there is no discussion about its parameters in this paper. Also droop control loop gets electrical power (𝑝 ) to (/aW) Generating new solution 𝑥 is performed using Lévy adjust turbine speed. GGOV1 model is partitioned into four flight which provides a random walk with below equation different subsystems which are determined in Fig. 1. So, obtaining of these parts’ parameters are necessary in order to (/aW) (/) 𝑥 = 𝑥 + 𝑎⨁𝐿é𝑣𝑦(𝜆) (1) evaluate the gas turbine behavior. , , Where 𝛼 is the step size, which is set at 𝛼 = 1 in this paper, the III. EXCITATION SYSTEM MODELING product ⨁ means entrywise multiplication and 𝐿é𝑣𝑦 is a Various excitation models could be found in the previous distribution which has infinite mean and variance and its literature , . However, excitation system models distribution is suitable for use in large-scale system stability studies must be selected for practical attempt of power plant model validation. kl 𝐿é𝑣𝑦~𝑢 = 𝑡 , (1 < 𝜆 ≤ 3). (2) In 421.5-2005 Standard of IEEE recommended models for practical stability studies are listed. The model structures In CS algorithm 𝑝 is the probability that the egg laid by a presented are intended to facilitate the use of field test data as a cuckoo is discovered by the host bird and we set this parameter means of obtaining model parameters. In 421.5-2005 Standard, in this paper at 0.25. Maximum generation and stop criterion exciter models are presented in three categories : AC, DC, kr are set at 100 and 𝑒 for all simulation in the next section. and Static. Exciter system in Chabahar power plant is of static Also appropriate initial population will be useful in reducing type. Hence, we must choose an appropriate model for experimental study of static category. The model selected for excitation system in this paper is ST6B. The structure of ST6B is illustrated in Fig. 2. Low-value Gate block in the model has two inputs which one of them is constructed by 𝐼 and 𝐼 is >? HI not active during normal condition. So the main focus in this paper is identifying of the exciter model in normal operation. In some power plant this model is called AVR mode. unitized by their ranges to have per-unitized parameters. A low-pass Butterworth filter with cutoff frequency 40𝐻𝑧 is used for filtering data. 3. In this stage least square method is used for pre- identifying desired parameters. In this approach parameters are estimated by Y kW Y 𝜃 = (𝑋 𝑋) 𝑋 𝑌 (3) Where 𝑌 is 𝑁 × 1 output vector for each subsystem, 𝑋 is 𝑁 × 𝑡 regressor matrix and 𝜃 is 𝑡 × 1 vector of subsystem model parameters. 4. Use identified parameter in the previous step as Fig. 3. Test setup in Chabahar gas power plant. (a) high speed data acquisition initial population for CS algorithm. Error index system (b) PLC setup (c) HMI of gas turbine and governor system (d) analog signal generator used for excitation in test time (e) generator room. which is used as objective function in CS is mean algorithm convergence time. This algorithm is used in the square error type. proposed identification procedure in the next section. V. SYSTEM IDENTIFICATION |( ) 𝑒𝑟𝑟𝑜𝑟 = 𝑦 − 𝑦~ (4) , , , W A. System setup To identify and validate parameters in GGOV1 and ST6B Where 𝑁 is data total number, 𝑦 is 𝑖th sample of standard models different tests have been performed on the gas recorded output data and 𝑦~ is 𝑖th sample of power plant while it was connected to the main grid. Excitation simulated output data. signal has a great influence in quality of identified model parameters. Although it has been recommended to use signal with rich frequency content (known as persistency of excitation) such as pseudo random binary signal (PRBS) and white noise , these signals are not applicable in identification of power plant subsystems because their sudden changes in value and frequency raise practical issue about system protection. So instead of using PRBS, square pulse wave with almost 5MW changes in power set point was applied to the power reference and all input and output of the subsystem has been recorded by 1 ms sample rate. Furthermore generator and gas turbine main specification are listed in Appendix. Fig. 3 Shows different part of the Chabahar gas power plant test setup. Data acquisition system which is used for recording data with high sample rate is shown in part (a) of this figure. Part (b) is power plant central controller, part (c) is human machine interface (HMI) which is used for monitoring of power plant. Analog signal generator used for system excitation is shown at (d) and generator with gas turbine overall room is displayed at (e). Using above setup two different data sets were collected for parameter estimation by proposed algorithm and validation of obtained parameters. Fig. 4. Flowchart of proposed identification method B. Identification method The proposed identification procedure is explained in the following steps. If the results is satisfying, end the algorithm and use different data set to validate identified parameters, otherwise 1. Data collection. In this part, input and output data for apply CS results as initial population and go to step 4. each subsystem has to be gathered using appropriate Fig. 4 demonstrates the used flowchart for parameter sample rate. All test are performed with square pulse estimation in this paper. We should notice although algorithm signal generated by industrial signal generator system. can work without linear identification section but convergence 2. Collected data should be processed. In most cases time for parameter estimation will be longer in this way. So one there is environmental noise which has to be removed can consider this algorithm as a hybrid optimization method in from the actual signal and also data has to be per the context of metaheuristic approaches in solving similar TABLE I problems. ERROR INDEX VALUES FOR TRAINING DATA C. Model training results System part Error index(percent) As mentioned in part A of this section field data recorded when the power plant was connected to the main grid. So for Subsystem (1) 0.0239 applying proposed identification algorithm there is no need to Subsystem (2) 0.0268 plant isolation from the main grid. Subsystem training results Subsystem (3) 0.001 are shown in Fig.5. As one can notice proposed identification Subsystem (4) 0.1192 Exciter (5) 0.0293 D. Model validation results Fig. 6 indicates validation results of the identified parameters using different data sets. Although tests have been performed in different operating point of the gas turbine but Fig. 6 (a) and Fig 6 (b) show identified parameters have appropriate dynamical responses so that they are acceptable. Test results also show both GGOV1 and ST6B standard models have very good conformity with real gas turbine and exciter system. Fig. 5. Measured and simulated results for different part of the system using training data. (a) Valve position (b) electrical power (c) speed controller output (d) temperature controller output (e) exciter output. algorithm successfully estimated unknown parameters and also results are displayed in per unit based so results can be compared easily. Table I also indicates the error index values in percent for different subsystems. All subsystems have error indices less than 0.2 percent which this result confirms Fig. 6. Measured and simulated results for different part of the system using conclusions have been inferred from Fig. 5. validation data. (a) Valve position (b) electrical power (c) speed controller output (d) temperature controller output (e) exciter output. Error index for all subsystems are given in Table II. As expected error values for validation indices are higher than training counterparts but they are still below the 0.5 percent so we conclude identified models have not been experts just for training data. At the end, in our opinion, parameter identification for studied power plant was done successfully. TABLE II ERROR INDEX VALUES FOR VALIDATION DATA System part Error index(percent) Subsystem (1) 0.1176 Subsystem (2) 0.0343 Subsystem (3) 0.0614 Subsystem (4) 0.1210 Exciter (5) 0.4085 E. Whiteness test Although defined error indices indicate that identified models have good qualities, but they do not show anything about what couldn’t be identified i.e. if there is any lost data in the residuals (errors) or not. Auto correlation for the residuals is as follows 𝑅 (𝜏) = 𝑒(𝑡)𝑒(𝑡 − 𝜏) (5) $ /W Fig. 7. Autocorrelation using training data for (a) valve position (b) electrical power (c) speed controller (d) temperature controller (e) exciter, identified models. Where 𝑒(𝑡) = 𝑦(𝑡) − 𝑦~(𝑡). If (5) is small for 𝜏 ≠ 0 then one can conclude residuals are white noise so models’ quality can be proved in another way. To verify similarity between white noise and residuals using (5) one should check this inequality N r 𝑅 (𝜏) < 𝛽 . (6) W $ I () W k 𝛽 is computed using exp = 𝛼 where 𝛼 is defined r r confidence level that is set 𝛼 = 0.01. Details for this method can be found in . Fig. 7 and Fig. 8 show defined index and confidence levels for training and validation data, respectively. As we see residual autocorrelations for different models are located between confidence levels so (6) is confirmed and correctness of identified parameters are validated. As an alternative of CS algorithm, GA and PSO have been used in proposed identification method to compare their performances with CS algorithm in our defined problem. Fig. 8. Autocorrelation using validation data for (a) valve position (b) electrical power (c) speed controller (d) temperature controller (e) exciter, identified models. GA and PSO have many parameters as degrees of freedom VI. CONCLUSION that must be tuned independently for each problem. The best IEEE standards for large scale gas turbine-governor and obtained results according to error indices are shown in Fig. 9 exciter have been studied in this paper to model a high duty gas and estimated parameters are included at Table III for different turbine-governor and excitation system for an installed gas algorithms. power plant in Iran. Some practical issues about a gas power As it can be seen from Fig. 9, PSO algorithm has higher error plant modeling and identification are discussed. An indices in comparison to CS and GA, therefore, its estimated identification procedure based on Cuckoo Search optimization parameters are not acceptable in this problem. CS and GA method, as one of the metaheuristic algorithms, has been results are approximately similar to each other. Actually in this proposed. GGOV1 and ST6B standard models’ parameters problem there is not a major difference in CS and GA have been identified using proposed algorithm. Validation performances so one can use them for parameter identification results and error indices show that the proposed identification problems. Although, as described before, due to its relative method successfully estimates standard models’ parameters. simplicity in tuning of required parameters, CS can be the first As a recommendation, Inlet Guide Van (IGV) model can be choice for solving similar problems. included in GGOV1. Adding this part may improve performance of the overall model at lower loads. Actually, IGVs are fully open at the loads beyond the 70% of nominal load and have not any effect at this range. APPENDIX Gas turbine and electrical generator main specifications installed in Chabahar power plant are listed below TABLE IV GENERATOR AND GAS TURBINE SPECIFICATIONS Parameter Value Prime mover GT V94.2 Generator Rated power 200 MVA Turbine Rated Power 160 MW Rated Voltage 15.75 KV Fig. 9. Validation data error indices for CS, GA and PSO algorithms for Rated Current 7331 A defined system parts Rated Speed 3000 rpm Rated Frequency 50 Hz TABLE III Excitation System Type Static IDENTIFIED SYSTEM PARAMETER VALUES Excitation Voltage at 296 V System CS GA PSO Rated Load parameter Excitation Current at 1417 A 3.10 3.12 2.9 ()*+ Rated Load 𝐾 0.90 0.91 0.95 ,)*+ 𝐾 0.0 0.0 0.10 ACKNOWLEDGMENT -)*+ 𝑇 0.0 0.0 0.15 The primary author wishes to thank all utility members of the -)*+ 1.83 1.80 1.75 Chabahar power plant, and Iran Grid Management Center . &/ (IGMC) who supported this research project. 𝐾 0.31 0.31 0.31 /012 𝑇 0.79 0.78 0.65 𝑇 0.0 0.0 0.0 0.10 0.11 0.15 $5) REFERENCES 𝑇 3.0 3.02 2.90 . 4%*.-  WSCC Control Work Group and Modeling & Validation Work 𝐾 25.01 24.95 24.10 (%*.- Group model validation, (1997, Feb.). Test guidelines for 𝐾 0.10 0.09 0.18 synchronous unit dynamic testing and model validation ” [Online] ,%*.- Available: www.wecc.biz 1.10 1.12 1.01 #$%$&  A. Ray, “Dynamic modeling of power plant turbines for controller 𝐿 0.90 0.91 0.83 design,” Appl. Math. Modeling, vol. 4, pp. 110-112, Apr. 1980. -1$4  X. Liu, and P. Mancarella, “Modeling, assessment and Sankey 𝑟 0.05 0.05 0.05 diagrams of integrated electricity-heat-gas networks in multi-vector 𝑊 0.43 0.43 0.43 district energy systems,” Applied Energy, to be published, Sep. 45% 2015, doi:10.1016/j.apenergy.2015.08.089. 3.95 3.93 3.81 #O  S. Barsali, A. De Marco, R. Giglioli, G. Ludovici, and A. Possenti, 𝐾 2.84 2.82 2.71 MO “Dynamic modeling of biomass power plant using micro gas turbine,” Renewable Energy, vol. 80, pp. 806-818, Feb. 2015. 𝐾 1.10 1.09 1.23 1.30 1.33 1.45 >>  E. Najimi, and M. H. Ramezani, “Robust control of speed and temperature in a power plant gas turbine,” ISA Transactions, vol. 51, pp. 304-308, 2012.  W. Qiao, Z. Gao, R. G. Harley, and G. K. Venayagamoorthy, “Robust neuro-identiﬁcation of nonlinear plants in electric power systems with missing sensor measurements,” Engineering Applications of Artiﬁcial Intelligence, vol. 21, pp. 604–618, Jul.  S. Glickman, R. Kulessky, and G. Nudelman, “A New Identification-Based Power Unit Model for Load-Frequency Control Purposes,” European Journal of Control, vol. 10, pp. 341- 351, Apr. 2004.  H. Balaghi Enalou and E. Abbasi Soreshjani, "A Detailed Governor- Turbine Model for Heavy-Duty Gas Turbines With a Careful Scrutiny of Governor Features," in IEEE Transactions on Power Systems, vol. 30, no. 3, pp. 1435-1441, May 2015.  D. C. Aliprantis, S. D. Sudhoff and B. T. Kuhn, "Genetic algorithm- based parameter identification of a hysteretic brushless exciter model," in IEEE Transactions on Energy Conversion, vol. 21, no. 1, pp. 148-154, March 2006.  M. Asmine et al., "Model Validation for Wind Turbine Generator Models," in IEEE Transactions on Power Systems, vol. 26, no. 3, pp. 1769-1782, Aug. 2011. doi: 10.1109/TPWRS.2010.2092794  P. Pourbeik, R. Rhinier, S. M. Hsu, B. Agrawal, and R. Bisbee, “Semi automated Model Validation of Power Plant Equipment Using Online Measurements,”IEEE Trans. Energy Conversion, vol. 28, no. 2, pp. 308-316, Jun. 2013.  S. Massucco, A. Pitto and F. Silvestro, "A Gas Turbine Model for Studies on Distributed Generation Penetration Into Distribution Networks," in IEEE Transactions on Power Systems, vol. 26, no. 3, pp. 992-999, Aug. 2011. doi: 10.1109/TPWRS.2010.2091290  Task Force on turbine governor modeling, Power System Dynamic Performance Committee, Power System Stability Subgroup, “Dynamic Models for Turbine-Governors in Power System Studies,” IEEE Power Energy Soc., January 2013.  L. Díez-Maroto, L. Rouco and F. Fernández-Bernal, "Modeling, Sizing, and Control of an Excitation Booster for Enhancement of Synchronous Generators Fault Ride-Through Capability: Experimental Validation," in IEEE Transactions on Energy Conversion, vol. 31, no. 4, pp. 1304-1314, Dec. 2016.  I. N. Moghaddam, Z. Salami and L. Easter, "Sensitivity Analysis of an Excitation System in Order to Simplify and Validate Dynamic Model Utilizing Plant Test Data," in IEEE Transactions on Industry Applications, vol. 51, no. 4, pp. 3435-3441, July-Aug. 2015.  IEEE Recommended Practice for Excitation System Models for Power System Stability Studies," in IEEE Std 421.5-2005 (Revision of IEEE Std 421.5-1992) , vol., no., pp.1-93, April 21 2006 doi: 10.1109/IEEESTD.2006.99499  C. Blum, A. Roli, "Metaheuristics in combinatorial optimization: Overview and conceptural comparision", ACM Comput. Surv., vol. 35, pp. 268-308, 2003.  C. Mishra, S. P. Singh and J. Rokadia, "Optimal power flow in the presence of wind power using modified cuckoo search," in IET Generation, Transmission & Distribution, vol. 9, no. 7, pp. 615-626, 4 30 2015.doi: 10.1049/iet-gtd.2014.0285  M. K. Kim, "Short-term price forecasting of Nordic power market by combination Levenberg–Marquardt and Cuckoo search algorithms," in IET Generation, Transmission & Distribution, vol. 9, no. 13, pp. 1553-1563, 10 1 2015.  X. S. Yang and Suash Deb, "Cuckoo Search via Lévy flights," 2009 World Congress on Nature & Biologically Inspired Computing (NaBIC), Coimbatore, 2009, pp. 210-214.doi: 10.1109/NABIC.2009.5393690  Ljung, L. (1999). System identification: theory for the user (2nd ed.). Englewood Cliffs,New Jersey: Prentice Hall.
Electrical Engineering and Systems Science – arXiv (Cornell University)
Published: Dec 27, 2018
Access the full text.
Sign up today, get DeepDyve free for 14 days.