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

Learn More →

Enhanced Memetic Algorithm-Based Extreme Learning Machine Model for Smart Grid Stability Prediction

Enhanced Memetic Algorithm-Based Extreme Learning Machine Model for Smart Grid Stability Prediction Hindawi International Transactions on Electrical Energy Systems Volume 2022, Article ID 8038753, 10 pages https://doi.org/10.1155/2022/8038753 Research Article Enhanced Memetic Algorithm-Based Extreme Learning Machine Model for Smart Grid Stability Prediction 1 2 3 4 Manohar Mishra , Janmenjoy Nayak , Bignaraj Naik , and Bhaskar Patnaik Electrical and Electronics Engineering Department, Faculty of Engineering and Technology, Siksha O Anusandhan University, Bhubaneswar, India Department of Computer Science, Maharaja Sriram Chandra BhanjaDeo (MSCBD)University, Mayurbhanj, Odisha 757003, India Department of Computer Application, Veer Surendra Sai University of Technology, Burla, India Department of Electrical Engineering, Biju Patnaik University of Technology, Rourkela, India Correspondence should be addressed to Bhaskar Patnaik; bhaskar7310@gmail.com Received 3 January 2022; Revised 8 July 2022; Accepted 14 July 2022; Published 31 August 2022 Academic Editor: Harsh Dhiman Copyright © 2022 Manohar Mishra et al. +is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. +e smart grid is considered a conventional application domain of cyber-physical system (CPS) tools in the electrical utility industry. +e physical system dynamics of SG with the assistance of CPS are generally controlled by connected sensors and controllers via a communication link. +ese CPSs, which rely heavily on an expansive communication network and intelligent computing algorithms, are susceptible to cyber-physical attacks and are also sensitive to various technical, economical, and social factors compromising their stability. Assessment and prediction of the stability of CPSs are very vital in this context. In this work, a novel optimized (memetic algorithm-based) extreme learning machine model for smart grid-CPS stability prediction has been proposed. Here, the teaching-learning-based optimization and simulated annealing techniques are used to design the memetic algorithm. +e experimental result regarding the proposed model is then compared with other contemporary machine learning and deep learning models. so. However, it hardly suffices considering the new chal- 1. Introduction lenges in the form of changing grid dynamics and the need Use of the state-of-the-art technologies in the field of for enhanced grid performance in terms of efficiency, reli- communication and computational ability has helped evolve ability, and resiliency. Although the conventional electrical most industries as smarter in terms of improved efficiency, power system (EPS) remains saddled with the aforemen- productivity, quality of service, etc. However, the power tioned deficiencies, the changing scenarios ushered in due to system delivery grid (electric grid) has largely remained fast depleting conventional energy sources and resulting nonmodernized as compared to other sectors. +e reasons technological developments in the field of renewable energy can be attributed to the fact that grids have been built and sources (RESs) have brought out newer challenges. Pene- developed over the years primarily as mechanical systems tration of renewable energy resources at low voltage and without attention to the possible future need for refur- distribution level of the EPS convert the grid topology from a bishment or technological overhauling. With burgeoning centralized system to a distributed generation system. Al- demand for power, there has been a large-scale expansion of though DGs help to supplement the energy requirements, grid infrastructure, and manual monitoring of the grid has they also pose many technological challenges, primarily as been increasingly difficult and challenging [1]. +us, a the power flow gets bidirectional leading to relay mis- certain level of computation-based automation in moni- coordination. On the other hand, the availability of RESs toring and controlling has been witnessed over the decade or also has given a fillip to the concept of microgrid technology, 2 International Transactions on Electrical Energy Systems (vii) Help minimize the price of electricity which not only helps cater to the local load demand but also adds much to the system stability. Needless to say, micro- +e above functionalities of an industrial smart grid that grids being one of the key smart grid components are going ultimately ensures the stability of the network are possible by to be embedded prominently. However, the implementation using intelligent computing algorithms that comprise the of the same, as mentioned earlier, poses many challenges. backbone of the cyber technology in a CPS. Applications of +e issues and challenges associated with implementing these intelligent computational algorithms, such as ad- these cutting-edge technologies are enumerated in [2]. +e vanced optimization techniques, expert systems, fuzzy logic, complexities involved in the successful realization of a smart neural network, and deep neural network are used to design grid are very daunting, but the benefits accrued are also of schemes meant for control, power management, power immense importance. In a smart grid scenario, the moni- system protection, power trading, predisaster, and post- toring and control of the EPS rest heavily on an expansive disaster management in order to have enhanced power communication network to gather information generated by system resiliency and all other such aspects of the smart grid numerous sensors placed at various locations in the EPS. A concept that help to add to its stability. huge amount of data thus collected needs to be processed In this work, an optimized extreme learning machine and subjected to computational algorithms which can (ELM) is suggested to predict the stability of the industrial provide the appropriate and accurate decision in the shortest smart grid (SG). +e performance of the proposed model is possible time. In short, we are veering towards an EPS which then compared with other contemporary machine learning is heavily reliant on intelligent computational techniques (ML) and deep learning (DL) models. with fast communication networks serving as the backbone. +e steps involved in this work can be presented as To be precise, the present-day EPS is smartening up as cyber- follows: physical systems (CPSs). CPS is a strong linkage between cyber technologies and the physical system in order to deal (1) Initially, the SG data have been retrieved from the UCI-ML repository [6] and processed through a with the operational and controlling complexities in a smart grid that are otherwise manually impossible to execute. NIST preprocessing step such as normalization and label encoding. cyber-physical system website defines CPS as follows [3]: (2) A novel memetic algorithm-based optimized ex- “Cyber-physical systems (CPSs) comprise interacting treme learning machine (ELM) is proposed for the digital, analog, physical, and human components engi- training and testing of the data. Here, the TLBO and neered for function through integrated physics and logic.” SA techniques are used to design the memetic algorithm. A more generalized definition is given in [4] as follows: (3) +is extracted SG dataset is then processed through the proposed optimized ELM model for training and “CPS can be characterized as physical and engineered testing purposes. systems whose operations are monitored, controlled, (4) +e performance of the proposed ML model is then coordinated, and integrated by a computing and com- compared with other contemporary ML and DL municating core.” approaches using a few key indexes, such as accu- racy, precision, recall, and F-score. In the industrial smart grid perspective, the concept of CPS embodies physical systems such as the network of In short, our key contributions toward the prediction of power infrastructure strongly interfaced with control, in- the SG stability can be summarized as follows: telligence, processing, and information. +e CPS enables the (a) A novel memetic algorithm-based ELM model has smart grid to deliver its expected functional features, which been proposed to predict the stability status of SG. are as follows [5]: (b) +e collected SG data are then trained and tested (i) Minimizing the need for deployment of additional through the proposed ML model. power plants through effective power management (c) An accuracy of 99.75% is attained through the (ii) Adaptive and self-healing protection mechanism in suggested model. Comparative analysis with other order to ensure high reliability and resiliency of the contemporary ML and DL techniques indicates the EPS superiority of the proposed model. (iii) Facilitate plug-in of DGs to the grid to meet the +e organization of the remaining parts of this article is localized power demand presented as follows: Section 2 reviews the current (iv) Exhibit flexibility in the distribution process to literature associated with the application of intelligent supply the right amount of power to the various computational techniques in various aspects of SG types of loads connected to the grid. functionalities. Section 3 presents the proposed memetic (v) Predict the short-term and long-term power algorithm-based ELM architecture in detail. +e experi- demand mentation and result analysis has been carried out in (vi) Help reduce pollution through environment- Section 4. Section 5 highlights the concluding remark with friendly renewable energy sources the future direction. International Transactions on Electrical Energy Systems 3 hub. Many optimizations and intelligent techniques too are 2. Literature Survey engaged in this tier for the intended purpose. A smart grid vastly relies on the accumulation of a huge From a smart grid stability perspective, it is very essential amount of data procured from numerous sensors placed at that zone-specific accurate load forecasting must be carried different strategic locations of the EPS, fast and robust out in order to meet the dynamic energy demands. Various internet-based communication channels, and of course a ML techniques can be used for such purposes. Based on fast-acting intelligent computation algorithm. It is imper- historical data on weather, load variation pattern, and energy ative that handling and managing such huge data, protecting generation, these ML algorithms make an accurate pre- the communication channel from cyber infringements, and diction of load demand in specific regions [18, 19]. In [19],a engaging a fast-acting computational technique for deep neural network (DNN) model is employed for gen- accurate prediction of the set objective are vital for realizing eration and load demand forecasting, where DNN has a smart grid. Smart grid deployment thus involves huge proven better than the contemporary regression model. In complexity. Advanced intelligent systems, with techniques [20], the authors have offered a big data framework for such as ML, DL, reinforcement learning (RL), deep distributed processing to predict energy load demand. Here, reinforcement learning (DRL), and SG realization are the MLib-ML library is used for assessing the performance of becoming feasible [7]. different regression models. +e stability of SGs, which In the above context, the authors in [8] enumerate how enables smart cities, is challenged by the dynamic energy the use of big data analysis in conjunction with intelligent consumption due to the household appliances in these smart models helps to resolve the issue of processing these huge cities. IoT technologies along with ICTs present many energy data in an SG. Various applications of big data in the SG management techniques to address the issue. +e authors in perspective are listed in [9]. +reat to the communication [21] suggested a consumption prediction technique based on network in the form of covert data integrity assault (CDIA) a probabilistic data-driven prognostic technique developed can be detrimental to the reliability and safety of smart grid on a Bayesian network (BN) framework. Elsisi and Tran functionalities. +ese smartly designed CDIAs can easily proposed a unified Internet of thing (IoT) architecture to outwit the conventional bad-data detector employed in SG manage the issue of cyberattacks using a DNN model having control hubs. +e authors in [10] have proposed an unsu- a rectified linear unit [22]. +is stated system can supervise pervised ML-based model to identify CDIAs in SG com- the automated guided vehicles reliably and securely. Elsisi munication grids using an unlabeled dataset. +e ML et al. suggested an effective online fault diagnosis system algorithm employed is called isolation forest. +e security against cyberattacks and data uncertainties using an IoT system against CDIAs generally is a three-tier structure. enabled DL model [23]. ‘Protection,’ ‘intrusion detection system (IDS)’, and ‘alle- +e cost of power also plays an important role in en- viation,’ are the first, second, and third tier, respectively. suring the stability of the distributed power systems. +e +rough shielded communications and data safeguarding authors in [24] have proposed a decentralized SG control measures, the first tier ensures protection of the commu- model to ensure demand-side management in the grid by nication channel against the majority of CDIAs. In the event analyzing the electricity price versus grid frequency devia- of the protection tier getting violated, the IDS as the second tion. +e authors have also implemented an optimized data tier of defense detects the intrusion and generates precau- matching ML technique and the transparent open box tionary signals for the operators to take up preventive learning model to realize dynamic SG stability. measures against the CDIA. +e recent literature enumer- Against the backdrop of the above discussion on various ates many ML-based IDSs [11–17]. +e authors in [11, 12] factors of smart grid instability and the possible intelligent have depicted the application of ML algorithms in identi- methods to mitigate them, it is also essential that there fying unscrupulous user activities in smart grid commu- should be an efficient scheme for the estimation and pre- nication channels. +e authors in [13] have demonstrated diction of SG stability in place. +e authors in [25] have the utility of several ML algorithms in the detection of proposed an ML-based smart grid stability forecasting CDIAs at the physical layer of a smart grid. +e authors in scheme. +e proposed method employs three different ge- [14] have employed a support vector machine (SVM) netic algorithms in the future selection stage and four dif- classifier in the above context. +e authors in [15] have ferent ML classifiers including the GBM algorithm. +e proposed another ML-based model to detect time syn- authors in [5] proposed a few DL-based models such as chronization assault (TSA). +e authors in [16] proposed the recurrent neural network (RNN), long short-term memory Euclidean-distance-based ML model to predict the CDIAs. (LSTM), and gated recurrent unit (GRU) for prediction of +e authors in [17] have suggested a genetic algorithm (GA) SG stability. combined with SVM, for future selection and classification It can be inferred very well from the above survey that purposes, respectively, in order to detect CDIAs. Most of the there is no single comprehensive solution to tackle the CDIA recognition techniques using machine learning as challenges of stability in an SG. Also, there is a need for stated in the collected works have considered supervised further research in the area of stability prediction as evident learning on labeled data only. +e third defense tier, known from the scarcity of the literature in this regard. In the as alleviation, serves as a kind of restoration system, which present study, it is envisaged to design the novel memetic helps restore the reliable system operations once the CPS- algorithm-based ELM model for smart grid stability pre- assault recognition message is established at the SG control diction. +e proposed approach shows better results than the 4 International Transactions on Electrical Energy Systems traditional ELM technique and other advanced ML and DL (Equation (4)) is the weight matrix representing the weights techniques. between hidden layer neurons and out neurons, and Equation (5) is the prediction matrix. 3. Proposed TLBO-SA-ELM Approach st � Houtput ×β, (2) +e performance of the ML model is generally enhanced by f b + op × w 􏼁 . . . f b + op × w 􏼁 1 1 1 L 1 L integrating the optimization algorithm [26–29]. In this ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ Houtput �⎢ . . . . . . . . . ⎥ , work, a memetic algorithm based on TLBO and SA is ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ proposed to increase the performance of the ELM classifier. f b + op × w 􏼁 . . . f b + op × w 􏼁 1 N 1 L N L N×L Afterward, the proposed optimized ELM is trained and (3) tested using the SG dataset. A single metaheuristic algorithm cannot solve all opti- β �􏼂β ,β , . . . ,β 􏼃 , (4) 1 2 L L×1 mization problems and is usually less effective for high- dimensional SG datasets. +erefore, there always remains a 􏽢 􏽢 􏽢 􏽢 (5) possibility to design an improved search approach and st �􏼂st ,st , . . . ,st 􏼃 . 1 2 N N×1 develop novel optimization techniques which can be used to In this work, the hyperparameters of ELM are optimized solve gene selection problems in the SG dataset. +is is our with the memetic version of teaching-learning based opti- major inspiration behind the scheming of this proposed mization (TLBO) [31]. Here, the TLBO is integrated with memetic algorithm. +e proposed SG stability predictive simulated annealing [32] in order to avoid a locally optimal algorithm is designed using three basic soft computing solution. models such as TLBO, SA, and ELM. In this work, the domains of the considered hyper- In this work, an ELM-based model [30] with optimized parameters are as follows: activation function (f), alpha (α), parameters is developed for efficient prediction of smart grid and several hidden neurons (nh). Here, f is the mathematical status from smart grid operational features. +e proposed equation that is responsible to determine whether the neuron problem can be visualized as an optimization problem where input is significant for prediction. α is the controlling pa- the objective is to select the best p � 􏼈f , α , nh 􏼉 in P � i i i i rameter for the adjustment of weights and H is the number of p , p , . . . , p (population with ‘n’ number of hyper- 􏼈 􏼉 1 2 n th hidden neurons that highly impact the performance and parameter sets). Here, p is the i randomly generated network stability. +is work is focused on the process of hyperparameter value set which is drawn from an allowed finding optimal parameters (P ) ELM by using TLBO with range of values as follows: f ∈ list[1, 2, 3, 4, 5, 6, 7, 8], SA. +e objective function is used to evaluate the parameter nh ∈ range[1, 200], and α ∈ range[0.1, 1.0]. Here, f , nh , i i i i th combinations (P ) and output a fitness (accuracy) and α are i activation function, a selected number of the fit � ELM(D , P ) which indicates how well the set of hidden layer, and learning rate. +e activation function f is P Train i hyperparameters performs for the considered problem. For chosen as ‘1,’ ‘2,’ ‘3,’ ‘4,’ ‘5,’ ‘6,’ ‘7,’ and ‘8’ for sine, tanh, the present problem, we have considered ‘accuracy’ as the tribas, sigmoid, hand/lim, soft/lim, Gaussian, and multi- evaluation matrix, and it is the objective to maximize the quadric, respectively. +e performance of ELM on the objective function presented in Equation (2). +e architecture prediction of smart grid status is dependent on these pa- of the proposed algorithm is presented graphically in Figure 1. rameters f, nh, and α. Here, the studied problem can be +e proposed scheme for hyperparameter optimization visualized as an optimization problem to get optimal P � starts with random generation of the population of the f , nh , α in P, which is the optimal parameter set of ELM 􏼈 􏼉 i i i hyperparameter set P � 􏼈P 􏼉 , where P � 􏼈f , nh , α 􏼉 i i i i i for solving the identification of the various states of the i�1 th representing i instance of the ELM parameter set. +e smart grid. On the given operational SG properties, m n fitness of each P in P is evaluated (in Algorithm 1, i.e., fit � i P D � 􏼈op , st 􏼉 , P � 􏼈P 􏼉 , and modelELM(D , P ), i i i Train i i�1 i�1 ∗ ELM(D , P ) by setting the P in ELM and testing on the where the objective is to find optimal P which optimizes the Train i i data D . +en, the fittest P in P is P . +e teaching Train i Teacher following objective function: factor tf � rand(1 + rand(1)) and the population P � arg max 􏼈fit � score st, st � ELM D , P􏼁 􏼁􏼉 i P ∈P i Train i meanP are computed in order to use them for the Mean new generation of a new population. For each P , the P is i i new ⎧ ⎨ ⎫ ⎬ generated as P � P + r × (P − tf × P ), where r i i Teacher Mean 􏽢 􏽢 � arg max fit � score(st, st) � 􏽘 I st , st􏼁 . P ∈P i i i i ⎩ ⎭ is a random number. +e updating of P is performed by i�1 new comparing each of P in P with the corresponding P in i i (1) new P . +en, the resultant population is improvised by choosing two solutions P and P randomly from P and +e dataset D � op , st is the collection of smart i j 􏼈 􏼉 i i i�1 altering them as mentioned in the following equation: grid operational properties op with thirteen features and one class label st representing the status of the grid con- new ⎫ ⎪ dition, i.e., either ‘stable’ or ‘unstable.’ +e proposed ELM ⎪ P � P + rand(1) × 􏼐P − P 􏼑 if 􏼒fit < fit 􏼓 i j i P P i j model has been trained with these instances. In ELM, the . (6) new prediction of the class label is made using Equation (2), P � P + rand(1) × 􏼐P − P 􏼑 Otherwise i j i j where Houtput (Equation (3)) is the output matrix, β International Transactions on Electrical Energy Systems 5 Initiate the population Generate a random solution of ELM parameter P' from P and N(P ) n best best best P = {P } , P = {f ,  , Nh } i i=1 i i i i Compute the fitness of Evaluate each P in P i P and P'' as fit and fit best bset best bset fit = ELM(D , P ) i Train i Select P with highest fitness (Accuracy) as P teacher Yes No if Compute P Compute tf = round(1+rand(1)) mean fit > fit best best new Compute P for each P new P = p + r × (P − tf + P ) i i teacher mean compute p = e update P by comparing each P in P new new within each P in P Generate a random number r for each randomly choosen P and P i j Inprove them by equation (6) ' r < p P = P best best find the best P i.e. P best Decrease the update P temperature T No No If Max. Yes trail reached? Max. Generation Reached Yes Return P best Return best solution P* Figure 1: Proposed system architecture. After improvisation of P, the best solution P in P is After getting the update P from Algorithm 2 (sim- best best selected based on the highest fitness. +en, the simulated ulated annealing process), the population P has been annealing (Algorithm 2) has been applied on P to generate updated by replacing old P with new P (returned from best best best a new solution from P in order to avoid local optimal the procedure SA). If a maximum generation is reached or best solutions. +e proposed approach of applying simulated no further improvement on the performance, then we stop annealing with TLBO not only accepts the best solution but and assign P � P . Finally, Algorithm 3 returns the best best also considers the nearer solution to the best solution with hyperparameter set P from the entire final population P. some probability. +e major steps of applying simulated +en, this is set on ELM as fit ∗ � ELM(D , P ) and P Train annealing on P are as follows: (i) random generation of a fit ∗ � ELM(D , P ) to get the generalized performance best P Train solution P from P and N(P ), where it is the on training data D and test data D , respectively. +e best best best Train Test neighborhood operation; (ii) computation of the fitness fit complete process of getting an optimal hyperparameterP best ′ ′ and fit of P and P , respectively, by calling the can be realized in Algorithm 3. best best best procedure ELM(D , P ) and ELM(D , P ), re- Train best Train best ′ ′ spectively; (iii) replacement of P with P if fit > fit , best best best best 4. Results and Discussion θ/T or if r and(1)< p, where p � e ; (iv) decrement of the temperature: T � 0.93 × T and checking the maximum trial. In this section, the proposed SG stability prediction method If maximum trial is reached, then extract the P . is evaluated through proper experimental analysis. best 6 International Transactions on Electrical Energy Systems 4.1. Experimental Platform. +e experimental setup for the Predicted proposed work comprises an online graphical processing stable TS FU unit (GPU) enabled by Google called “Google Colab,” a supercomputer having a Windows 10 Operating system with a core I7 processor and a Python 3.9 programming tool. Unstable FS TU 4.2. Dataset Description. +e SG data used in this experi- ment have been extracted from the UCI-ML repository [6]. +is dataset comprised 10000 samples with fourteen fea- TP: Truly Stable tures. +e attributes are divided into 12 primary predictive FN: Falsely Unstable features and 2 dependent variables. +e predictive features FP: Falsely Stable provide information about the reaction time (tau[x]) of TN: Truly Unstable smart grid participants {range: 0.5–10s}, nominal power S: Stable (p[x]) consumed (negative)/produced (positive), and co- U: Unstable efficient related to elasticity price (g[x]). +e dependent Figure 2: An example of the confusion matrix. attributes can be described as follows: the first attribute indicates: the maximal real part of the characteristic equa- tion root (if positive - the system is linearly unstable) and the F2_Score (F1): in the calculation of F1_Score, less weight second dependent attribute states the stability label (class is given for precision compared to the recall value. +erefore, level) of the system (categorical: stable/unstable). in equation (10), a beta (β) value of 2 can be considered for the calculation of F1_measure. So, it can be represented as 4.3. Performance Evaluation Matrices. In this work, the follows: classification result corresponding to the proposed TLBO- 5 R∗ P SA-ELM is interpreted through the following performance (12) F2 � × . evaluation matrices. 4 R + P Confusion matrix (CM): it represents the actual value with respect to the predicted values corresponding to the class level. 4.4. Performance Evaluation of the Proposed TLBO-SA-ELM CM is a table that visualizes and compares the result Model. +e collected SG data are divided into 8 : 2 ratios for obtained from a classifier by presenting the actual values and training and testing, respectively. +e testing result in terms predicted values (number of samples) with respect to the of the confusion matrix is presented in Figure 3. class levels in terms of correct and incorrect prediction. A Figure 3 shows the confusion matrix for ELM and the sample of the confusion matrix is presented in Figure 2. proposed model for the classification of the SG stability From the above confusion matrix, the following six dataset. +e proposed TLBO-SA-ELM model detects 703 matrices are calculated as presented from the following data as truly stable and 1292 as truly unstable, whereas equations. traditional ELM detects 669 data as truly stable and 1223 as Accuracy (A): it signifies the correctness of the classifier. truly unstable. Only 5 data are misclassified through the It is mathematically represented as follows: proposed model. From these confusion matrixes, different TS + TU measuring indices such as precision, recall, accuracy, and (7) A � , Fβ-Score are calculated and tabulated in Table I. It can be TS + TU + FS + FU seen from the table that the proposed model can classify the TS stability of the SG data with 99.75% accuracy. +e other (8) Precision(P): P � , indices such as precision, recall, F1, F2 scores, and area TS + FU under the curve (AUC) are noted to be 1.0, 0.996144, TS 0.998068, 0.6197, and 0.998072, respectively. It can be seen (9) Recall(R): R � , TS + FS from the table that the proposed model outperforms tra- ditional ELM with respect to all the measuring indices. R∗ P +e proposed approach is a hybrid model where the Fβ Score: F � 􏼐1 + β 􏼑 × . (10) β (R + P) parameter of the ELM is optimized through a memetic algorithm comprised TLBO and SA. In order to check the F1_Score (F1): in the calculation of F1_Score, equal effectiveness of the proposed memetic algorithm, a few other weight is given for both precision and recall. +erefore, in optimization algorithms such as genetic algorithm (GA), equation (10), a beta (β) value of 1 can be considered for the particle swarm optimization (PSO), and TLBO are hy- calculation of F1_measure. So it can be represented as bridized with the ELM algorithm, and the corresponding follows : result is interpreted in Figure 4. Figure 4 shows the plot of R∗ P fitness (accuracy) versus the number of generations. It is (11) F1 � 2 × . seen from the figure that the proposed memetic algorithm- R + P stable Unstable True International Transactions on Electrical Energy Systems 7 Predicted Predicted stable 669 34 stable 703 0 Unstable 74 1223 Unstable 5 1292 stable Unstable stable Unstable (a) (b) Figure 3: Confusion matrix for (a) traditional ELM and (b) proposed TLBO-SA-ELM. 0 510 15 20 25 30 35 40 Generations ELM-GA ELM-TLBO ELM-TLBO-SA ELM-PSO Figure 4: Fitness changes in various generations. th Let D � op , st be the smart grid operational data, where op � op , op , . . . , op is the i instance of the smart grid 􏼈 􏼉 􏽮 􏽯 i i i�1 i i,1 i,2 i,N operational feature value and st is the status of the grid. (1) Randomly generate bias b , i � 0 to L, and weight w , i � 0 to L. i i (2) Calculate the hidden layer output function Houtput by using the selected activation function f(·). f(b + op × w ) ... f(b + op × w ) 1 1 1 L 1 L ⎢ ⎥ ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ Houtput � ⎣ ... ... ... ⎦ . f(b + op × w ) ... f(b + op × w ) 1 N 1 L N L N×L (3) Compute the output weight matrix β � Houtput × st, which maximizes the objective function 􏽢 μ ‖Houtput × β − st‖ � min ‖Houtput × β − st‖. Here, Houtput is the Moore–Penrose generalized inverse of the Houtput. μ −1 T Houtput � (Houtput × Houtput) × Houtput . k 1 if (st � k) 􏽢 􏽢 i 􏽢 􏽢 Perform the prediction by using β on the data st � α × ( Houtput × β), α is the learning rate. st � 􏼨 􏼩, −1 if (st ≠ k) k � 1, 2, . . . , c c k 􏽢 􏽢 Predict the final class label as st � arg max (st ) i k�1,2,...,c i Return score of the prediction fit � score(st, st) ALGORITHM 1: fit ⟵ ELM(X, P ). P i (1) Randomly generate a solution P from P and N(P ) (neighbor structure of P ) best best best best ′ ′ (2) Compute the fitness fit and fit of P and P , respectively, by calling the procedure ELM(D , P ) and best best best best Train best ELM(D , P ), respectively Train best ′ ′ (3) If fit > fit , then assign P � P best best best best θ/T Else calculate p � e . If r and(1)< p, then assign P � P best best (4) Decrease the temperature, T � 0.93 × T (5) If the maximum trial is reached, then return P best (6) Else go to step 1. ALGORITHM 2: P � SA(P , N(P ), D ). best best best Train Fitness (Accuracy) True True 8 International Transactions on Electrical Energy Systems th Let D � 􏼈op , st 􏼉 be the smart grid operational data, where op � 􏽮op , op , . . . , op 􏽯 is the i instance of the smart grid i i i�1 i i,1 i,2 i,N operational feature value and st is the status of the grid. th (1) Randomly generate the population P � 􏼈P 􏼉 , where P � 􏼈f , nh , α 􏼉 representing i is an instance of the ELM parameter set. i i�1 i i i i (2) Evaluate the fitness of each P in P. fit � ELM(D , P ). i P Train i (3) Select fittest P in P as P . i Teacher (4) Compute teaching factor tf � r and(1 + rand(1)) and population meanP . Mean new new (5) Generate P from P in P: P � P + r × (P − tf × P ), where r is a random number. i i i i Teacher Mean new new (6) Perform the updating P by comparison of each P in P with the corresponding P in P . i i (7) For each pair of randomly chosen P and P in P, improvise them by using the following equation: i j new P � P + r and(1) × (P − P ) if (fit < fit ) i j i P P i j 􏼩. new P � P + r and(1) × (P − P ) Otherwise i j i j (8) Find the best solution P in P based on the highest fitness. best (9) Generate a new solution from P by using a simulated annealing process P � SA(P , N(P ), D ). best best best best Train (10) Update the population by replacing old P with new P (returned from the procedure SA). best best (11) If a maximum generation is reached, then assign P � P . best Else, go to step 2. (12) Return P . ALGORITHM 3: P ←ELM − TLBO − SA(D , P). Train Table 1: Performance metrics of prediction models. Performance Metrics Prediction Models Precision Recall F1 Score F2 Score ROC-AUC Accuracy DT 0.7488 1.0 0.8871 0.702315 0.61 74.88 NB 0.6485 1.0 0.786775 0.585075 0.5 64.85 LR 0.6485 1.0 0.786775 0.585075 0.5 64.85 RF 0.946227 0.909020 0.927251 0.5795 0.906857 90.75 XGBoost 0.910437 0.995373 0.951012 0.5944 0.907359 93.35 ELM 0.972951 0.942945 0.957713 0.5986 0.947290 94.6 Proposed model 1.0 0.996144 0.998068 0.6197 0.996086 99.75 Table 2: Performance comparison with existing approaches. Model Accuracy 80 Recurrent neural network (RNN) [5] 96.60 Gated recurrent unit (GRU) [5] 97.30 Long short term memory (LSTM) [5] 97.13 Multidimensional long short term memory (M-LSTM) 99.07 [5] Proposed model (ELM-TLBO-SA) 99.75 based ELM converses quickly (within 14 generations) compared to other approaches. Figure 5: Overall comparison of studied models. 4.4.1. Comparative Analysis. In order to validate the per- formance of the proposed algorithm applied to the SG the training and testing dataset is considered during these experimentations. stability dataset, the corresponding experimental results are compared with the output of several other contemporary ML It can be analyzed from the table that NB and LR show the worst performance (accuracy: 64.85) compared to models. In this regard, the results corresponding to a de- cision tree (DT), na¨ıve bias (NB), linear regression (LR), XGBoost having an accuracy of 93.35%. It can also be an- alyzed that the performance of XGBoost is observed to be random forest (RF), and extreme gradient boosting (XGBoost) are depicted in Table 1. A similar ratio (8 : 2) of less than the proposed model with a very high margin Accuracy NB LR DT XGBoost RF ELM RNN GRU LSTM M-LSTM Proposed Method International Transactions on Electrical Energy Systems 9 (accuracy> 6.4%). Finally, it can be concluded that the TSA: Time synchronization assault proposed approach performs better than all other ML GA: Genetic algorithm models with respect to each performance indices. In addi- DNN: Deep neural network tion to this, the performance of the proposed approach is IoT: Internet of thing compared with the result of different deep learning models BN: Bayesian network (such as RNN, GRU, LSTM, and MLSTM) applied to a RNN: Recurrent neural network similar dataset [5]. +is comparative result analysis is pre- GRU: Gated recurrent unit sented in Table 2. It can be analyzed from the table that LSTM: Long short-term memory MLSTM performs better than RNN, GRU, and LSTM; CM: Confusion matrix however, the proposed algorithm (TLBO-SA-ELM) out- f: Activation function performs in all aspect. Figure 5 shows the overall com- α: Alpha parative results graphically. nh: Number of hidden neuron P : Hyperparameter 5. Conclusion D : Training data Train D : Test data. Test Smart grids are identified with the cyber-physical system used for intelligent management of power generation and Data Availability dissipation ensuring quality power supply at the most economical price. However, the threat to CPS and several +e data used in this study are openly available in (UCI-ML issues including a threat to CPSs severely affects the stability Repository) at (https://archive.ics.uci.edu/ml/datasets/ of a smart grid. Machine learning techniques play a vital role Electrical+Grid+Stability+Simulated+Data+) [6]. in predicting the stability of the smart grid. In this work, a novel memetic algorithm-based ELM model is introduced to predict the stability of the SG. +e proposed optimized ELM Conflicts of Interest algorithm is tested on the SG dataset extracted from the +e authors declare that they have no conflicts of interest. UCI-ML repository. +e proposed model achieved 99.75% accuracy and 100% precision in classifying the stability status of the smart grid. To justify the effectiveness of the References proposed model, a detailed comparative analysis is carried [1] Z. M. Fadlullah and N. Kato, Evolution of Smart Grids, out with other contemporary machine learning models Springer, Berlin, Germany, 2015. (such as decision tree, linear regression, random forest, na¨ıve [2] B. Patnaik, M. Mishra, R. C. Bansal, and R. K. Jena, “AC bias, XGBoost, and ELM) and previously published work microgrid protection–A review: current and future pro- based on deep learning (such as LSTM, GRU, RNN, and spective,” Applied Energy, vol. 271, Article ID 115210, 2020. MLSTM). 99.8072% of ROC is achieved through the pro- [3] 2021, https://www.nist.gov/el/cyber-physical-systems. posed algorithm which outperformed the other comparative [4] R. Rajkumar, I. Lee, L. Sha, and J. Stankovic, “Cyber-physical models. +e effectiveness of the proposed memetic algo- systems: the next computing revolution,” in Proceedings of the rithm based on TLBO and SA is additionally compared with Design Automation Conference, pp. 731–736, Anaheim, CA, several other optimization techniques (such as particle swam USA, June 2010. [5] M. Alazab, S. Khan, S. S. R. Krishnan, Q. V. Pham, optimization, genetic algorithm, and TLBO) hybridized with M. P. K. Reddy, and T. R. Gadekallu, “A multidirectional ELM. +e future scope of the work may deal with the LSTM model for predicting the stability of a smart grid,” IEEE analysis of dynamic power requirements, reliability, and Access, vol. 8, pp. 85454–85463, 2020. resiliency of SG. [6] V. Arzamasov, K. Bohm, ¨ and P. Jochem, “Towards concise models of grid stability,” in Proceedings of the 2018 IEEE Abbreviations International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm), SG: Smart grid pp. 1–6, IEEE, Aalborg, Denmark, October 2018. CPS: Cyber-physical system [7] D. Zhang, X. Han, and C. Deng, “Review on the research and EUI: Electrical utility industry practice of deep learning and reinforcement learning in smart grids,” CSEE Journal of Power and Energy Systems, vol. 4, ELM: Extreme learning machine no. 3, pp. 362–370, 2018. TLBO: Teaching-learning-based optimization [8] S. Zerdoumi, A. Q. M. Sabri, A. Kamsin et al., “Image pattern SA: Simulated annealing recognition in big data: taxonomy and open challenges: EPS: Electrical power system survey,” Multimedia Tools and Applications, vol. 77, no. 8, RES: Renewable energy resources pp. 10091–10121, 2018. DG: Distributed generation [9] M. Ghorbanian, S. H. Dolatabadi, and P. Siano, “Big data ML: Machine learning issues in smart grids: a survey,” IEEE Systems Journal, vol. 13, DL: Deep learning no. 4, pp. 4158–4168, 2019. CDIA: Covert data integrity assault [10] S. Ahmed, Y. Lee, S.-H. Hyun, and I. Koo, “Unsupervised IDS: Intrusion detection system machine learning-based detection of covert data integrity SVM: Support vector machine assault in smart grid networks utilizing isolation forest,” IEEE 10 International Transactions on Electrical Energy Systems Transactions on Information Forensics and Security, vol. 14, Intelligent Computer Communication and Processing (ICCP), no. 10, pp. 2765–2777, 2019. pp. 175–182, Cluj-Napoca, Romania, September 2019. [11] Z. M. Fadlullah, M. M. Fouda, N. Kato, X. Shen, and [26] M. Elsisi, “Optimal design of nonlinear model predictive Y. Nozaki, “An early warning system against malicious ac- controller based on the new modified multitracker optimi- zation algorithm,” International Journal of Intelligent Systems, tivities for smart grid communications,” IEEE Network, vol. 35, no. 11, pp. 1857–1878, 2020. vol. 25, no. 5, pp. 50–55, 2011. [27] M. Elsisi, “Improved grey wolf optimizer based on opposition [12] Y. Zhang, L. Wang, W. Sun, R. C. Green II, and M. Alam, and quasi learning approaches for optimization: case study “Distributed intrusion detection system in a multi-layer autonomous vehicle including vision system,” Artificial In- network architecture of smart grids,” IEEE Transactions on telligence Review, pp. 1–24, 2022. Smart Grid, vol. 2, no. 4, pp. 796–808, 2011. [28] R. M. Adnan, R. R Mostafa, O. Kisi, Z. M. Yaseen, S. Shahid, [13] M. Ozay, I. Esnaola, F. T. Yarman Vural, S. R. Kulkarni, and and M. Zounemat-Kermani, “Improving streamflow predic- H. V. Poor, “Machine learning methods for attack detection in tion using a new hybrid ELM model combined with hybrid the smart grid,” IEEE Transactions on Neural Networks and particle swarm optimization and grey wolf optimization,” Learning Systems, vol. 27, no. 8, pp. 1773–1786, 2016. Knowledge-Based Systems, vol. 230, Article ID 107379, 2021. [14] M. Esmalifalak, L. Liu, N. Nguyen, R. Zheng, and Z. Han, [29] L. Yang, Y. Li, and Z. Li, “Improved-ELM method for “Detecting stealthy false data injection using machine learning detecting false data attack in smart grid,” International in smart grid,” IEEE Systems Journal, vol. 11, no. 3, Journal of Electrical Power & Energy Systems, vol. 91, pp. 1644–1652, 2017. pp. 183–191, 2017. [15] J. Wang, W. Tu, L. C. Hui, S. Yiu, and E. K. Wang, “Detecting [30] G. B. Huang, Q. Y. Zhu, and C. K. Siew, “Extreme learning time synchronization attacks in cyber-physical systems with machine: theory and applications,” Neurocomputing, vol. 70, machine learning techniques,” in Proceedings of the IEEE 37th no. 1-3, pp. 489–501, 2006. International Conference on Distributed Computing Systems, [31] R. Rao, V. Savsani, D. Vakharia, and D. P. Vakharia, pp. 2246–2251, ICDCS, Atlanta, GA, USA, June 2017. “Teaching–learning-based optimization: a novel method for [16] S. Ahmed, Y. Lee, S.-H. Hyun, and I. Koo, “Feature selec- constrained mechanical design optimization problems,” tion–based detection of covert cyber deception assaults in Computer-Aided Design, vol. 43, no. 3, pp. 303–315, 2011. smart grid communications networks using machine learn- [32] L. Liu, H. Mu, J. Yang, X. Li, and F. Wu, “A simulated ing,” IEEE Access, vol. 6, pp. 27518–27529, 2018. annealing for multi-criteria optimization problem: Dbmosa,” [17] S. Ahmed, Y. Lee, S. H. Hyun, and I. Koo, “Covert cyber Swarm and Evolutionary Computation, vol. 14, pp. 48–65, assault detection in smart grid networks utilizing feature selection and euclidean distance-based machine learning,” Applied Sciences, vol. 8, no. 5, p. 772, 2018. [18] C. Chu, “Using machine learning to predict renewable energy generation and smart grid designs,” M.Sc +esis, Princeton University, Princeton, NJ, USA, 2019. [19] R. Vinayakumar, M. Alazab, K. P. Soman, P. Poornachandran, A. Al-Nemrat, and S. Venkatraman, “Deep learning approach for intelligent intrusion detection system,” IEEE Access, vol. 7, [20] D. Syed, S. S. Refaat, and H. Abu-Rub, “Performance eval- uation of distributed machine learning for load forecasting in smart grids,” in Proceedings of the IEEE Cybernetics & In- formatics (K&I), pp. 1–6, Velke Karlovice, Czech Republic, January 2020. [21] N. Bassamzadeh and R. Ghanem, “Multiscale stochastic prediction of electricity demand in smart grids using bayesian networks,” Applied Energy, vol. 193, pp. 369–380, 2017. [22] M. Elsisi and M. Q. Tran, “Development of an IoTarchitecture based on a deep neural network against cyber attacks for automated guided vehicles,” Sensors, vol. 21, no. 24, p. 8467, [23] M. Elsisi, M. Q. Tran, K. Mahmoud, D. E. A. Mansour, M. Lehtonen, and M. M. Darwish, “Effective IoT-based deep learning platform for online fault diagnosis of power trans- formers against cyberattacks and data uncertainties,” Mea- surement, vol. 190, Article ID 110686, 2022. [24] D. A. Wood, “Predicting stability of a decentralized power grid linking electricity price formulation to grid frequency applying an optimized data matching learning network to simulated data,” Technology and Economics of Smart Grids and Sustainable Energy, vol. 5, no. 1, pp. 3–21, 2020. [25] D. Moldovan and I. Salomie, “Detection of sources of in- stability in smart grids using machine learning techniques,” in Proceedings of the IEEE 15th International Conference on http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png International Transactions on Electrical Energy Systems Hindawi Publishing Corporation

Enhanced Memetic Algorithm-Based Extreme Learning Machine Model for Smart Grid Stability Prediction

Loading next page...
 
/lp/hindawi-publishing-corporation/enhanced-memetic-algorithm-based-extreme-learning-machine-model-for-E5wBAmGruF
Publisher
Hindawi Publishing Corporation
Copyright
Copyright © 2022 Manohar Mishra et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
eISSN
2050-7038
DOI
10.1155/2022/8038753
Publisher site
See Article on Publisher Site

Abstract

Hindawi International Transactions on Electrical Energy Systems Volume 2022, Article ID 8038753, 10 pages https://doi.org/10.1155/2022/8038753 Research Article Enhanced Memetic Algorithm-Based Extreme Learning Machine Model for Smart Grid Stability Prediction 1 2 3 4 Manohar Mishra , Janmenjoy Nayak , Bignaraj Naik , and Bhaskar Patnaik Electrical and Electronics Engineering Department, Faculty of Engineering and Technology, Siksha O Anusandhan University, Bhubaneswar, India Department of Computer Science, Maharaja Sriram Chandra BhanjaDeo (MSCBD)University, Mayurbhanj, Odisha 757003, India Department of Computer Application, Veer Surendra Sai University of Technology, Burla, India Department of Electrical Engineering, Biju Patnaik University of Technology, Rourkela, India Correspondence should be addressed to Bhaskar Patnaik; bhaskar7310@gmail.com Received 3 January 2022; Revised 8 July 2022; Accepted 14 July 2022; Published 31 August 2022 Academic Editor: Harsh Dhiman Copyright © 2022 Manohar Mishra et al. +is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. +e smart grid is considered a conventional application domain of cyber-physical system (CPS) tools in the electrical utility industry. +e physical system dynamics of SG with the assistance of CPS are generally controlled by connected sensors and controllers via a communication link. +ese CPSs, which rely heavily on an expansive communication network and intelligent computing algorithms, are susceptible to cyber-physical attacks and are also sensitive to various technical, economical, and social factors compromising their stability. Assessment and prediction of the stability of CPSs are very vital in this context. In this work, a novel optimized (memetic algorithm-based) extreme learning machine model for smart grid-CPS stability prediction has been proposed. Here, the teaching-learning-based optimization and simulated annealing techniques are used to design the memetic algorithm. +e experimental result regarding the proposed model is then compared with other contemporary machine learning and deep learning models. so. However, it hardly suffices considering the new chal- 1. Introduction lenges in the form of changing grid dynamics and the need Use of the state-of-the-art technologies in the field of for enhanced grid performance in terms of efficiency, reli- communication and computational ability has helped evolve ability, and resiliency. Although the conventional electrical most industries as smarter in terms of improved efficiency, power system (EPS) remains saddled with the aforemen- productivity, quality of service, etc. However, the power tioned deficiencies, the changing scenarios ushered in due to system delivery grid (electric grid) has largely remained fast depleting conventional energy sources and resulting nonmodernized as compared to other sectors. +e reasons technological developments in the field of renewable energy can be attributed to the fact that grids have been built and sources (RESs) have brought out newer challenges. Pene- developed over the years primarily as mechanical systems tration of renewable energy resources at low voltage and without attention to the possible future need for refur- distribution level of the EPS convert the grid topology from a bishment or technological overhauling. With burgeoning centralized system to a distributed generation system. Al- demand for power, there has been a large-scale expansion of though DGs help to supplement the energy requirements, grid infrastructure, and manual monitoring of the grid has they also pose many technological challenges, primarily as been increasingly difficult and challenging [1]. +us, a the power flow gets bidirectional leading to relay mis- certain level of computation-based automation in moni- coordination. On the other hand, the availability of RESs toring and controlling has been witnessed over the decade or also has given a fillip to the concept of microgrid technology, 2 International Transactions on Electrical Energy Systems (vii) Help minimize the price of electricity which not only helps cater to the local load demand but also adds much to the system stability. Needless to say, micro- +e above functionalities of an industrial smart grid that grids being one of the key smart grid components are going ultimately ensures the stability of the network are possible by to be embedded prominently. However, the implementation using intelligent computing algorithms that comprise the of the same, as mentioned earlier, poses many challenges. backbone of the cyber technology in a CPS. Applications of +e issues and challenges associated with implementing these intelligent computational algorithms, such as ad- these cutting-edge technologies are enumerated in [2]. +e vanced optimization techniques, expert systems, fuzzy logic, complexities involved in the successful realization of a smart neural network, and deep neural network are used to design grid are very daunting, but the benefits accrued are also of schemes meant for control, power management, power immense importance. In a smart grid scenario, the moni- system protection, power trading, predisaster, and post- toring and control of the EPS rest heavily on an expansive disaster management in order to have enhanced power communication network to gather information generated by system resiliency and all other such aspects of the smart grid numerous sensors placed at various locations in the EPS. A concept that help to add to its stability. huge amount of data thus collected needs to be processed In this work, an optimized extreme learning machine and subjected to computational algorithms which can (ELM) is suggested to predict the stability of the industrial provide the appropriate and accurate decision in the shortest smart grid (SG). +e performance of the proposed model is possible time. In short, we are veering towards an EPS which then compared with other contemporary machine learning is heavily reliant on intelligent computational techniques (ML) and deep learning (DL) models. with fast communication networks serving as the backbone. +e steps involved in this work can be presented as To be precise, the present-day EPS is smartening up as cyber- follows: physical systems (CPSs). CPS is a strong linkage between cyber technologies and the physical system in order to deal (1) Initially, the SG data have been retrieved from the UCI-ML repository [6] and processed through a with the operational and controlling complexities in a smart grid that are otherwise manually impossible to execute. NIST preprocessing step such as normalization and label encoding. cyber-physical system website defines CPS as follows [3]: (2) A novel memetic algorithm-based optimized ex- “Cyber-physical systems (CPSs) comprise interacting treme learning machine (ELM) is proposed for the digital, analog, physical, and human components engi- training and testing of the data. Here, the TLBO and neered for function through integrated physics and logic.” SA techniques are used to design the memetic algorithm. A more generalized definition is given in [4] as follows: (3) +is extracted SG dataset is then processed through the proposed optimized ELM model for training and “CPS can be characterized as physical and engineered testing purposes. systems whose operations are monitored, controlled, (4) +e performance of the proposed ML model is then coordinated, and integrated by a computing and com- compared with other contemporary ML and DL municating core.” approaches using a few key indexes, such as accu- racy, precision, recall, and F-score. In the industrial smart grid perspective, the concept of CPS embodies physical systems such as the network of In short, our key contributions toward the prediction of power infrastructure strongly interfaced with control, in- the SG stability can be summarized as follows: telligence, processing, and information. +e CPS enables the (a) A novel memetic algorithm-based ELM model has smart grid to deliver its expected functional features, which been proposed to predict the stability status of SG. are as follows [5]: (b) +e collected SG data are then trained and tested (i) Minimizing the need for deployment of additional through the proposed ML model. power plants through effective power management (c) An accuracy of 99.75% is attained through the (ii) Adaptive and self-healing protection mechanism in suggested model. Comparative analysis with other order to ensure high reliability and resiliency of the contemporary ML and DL techniques indicates the EPS superiority of the proposed model. (iii) Facilitate plug-in of DGs to the grid to meet the +e organization of the remaining parts of this article is localized power demand presented as follows: Section 2 reviews the current (iv) Exhibit flexibility in the distribution process to literature associated with the application of intelligent supply the right amount of power to the various computational techniques in various aspects of SG types of loads connected to the grid. functionalities. Section 3 presents the proposed memetic (v) Predict the short-term and long-term power algorithm-based ELM architecture in detail. +e experi- demand mentation and result analysis has been carried out in (vi) Help reduce pollution through environment- Section 4. Section 5 highlights the concluding remark with friendly renewable energy sources the future direction. International Transactions on Electrical Energy Systems 3 hub. Many optimizations and intelligent techniques too are 2. Literature Survey engaged in this tier for the intended purpose. A smart grid vastly relies on the accumulation of a huge From a smart grid stability perspective, it is very essential amount of data procured from numerous sensors placed at that zone-specific accurate load forecasting must be carried different strategic locations of the EPS, fast and robust out in order to meet the dynamic energy demands. Various internet-based communication channels, and of course a ML techniques can be used for such purposes. Based on fast-acting intelligent computation algorithm. It is imper- historical data on weather, load variation pattern, and energy ative that handling and managing such huge data, protecting generation, these ML algorithms make an accurate pre- the communication channel from cyber infringements, and diction of load demand in specific regions [18, 19]. In [19],a engaging a fast-acting computational technique for deep neural network (DNN) model is employed for gen- accurate prediction of the set objective are vital for realizing eration and load demand forecasting, where DNN has a smart grid. Smart grid deployment thus involves huge proven better than the contemporary regression model. In complexity. Advanced intelligent systems, with techniques [20], the authors have offered a big data framework for such as ML, DL, reinforcement learning (RL), deep distributed processing to predict energy load demand. Here, reinforcement learning (DRL), and SG realization are the MLib-ML library is used for assessing the performance of becoming feasible [7]. different regression models. +e stability of SGs, which In the above context, the authors in [8] enumerate how enables smart cities, is challenged by the dynamic energy the use of big data analysis in conjunction with intelligent consumption due to the household appliances in these smart models helps to resolve the issue of processing these huge cities. IoT technologies along with ICTs present many energy data in an SG. Various applications of big data in the SG management techniques to address the issue. +e authors in perspective are listed in [9]. +reat to the communication [21] suggested a consumption prediction technique based on network in the form of covert data integrity assault (CDIA) a probabilistic data-driven prognostic technique developed can be detrimental to the reliability and safety of smart grid on a Bayesian network (BN) framework. Elsisi and Tran functionalities. +ese smartly designed CDIAs can easily proposed a unified Internet of thing (IoT) architecture to outwit the conventional bad-data detector employed in SG manage the issue of cyberattacks using a DNN model having control hubs. +e authors in [10] have proposed an unsu- a rectified linear unit [22]. +is stated system can supervise pervised ML-based model to identify CDIAs in SG com- the automated guided vehicles reliably and securely. Elsisi munication grids using an unlabeled dataset. +e ML et al. suggested an effective online fault diagnosis system algorithm employed is called isolation forest. +e security against cyberattacks and data uncertainties using an IoT system against CDIAs generally is a three-tier structure. enabled DL model [23]. ‘Protection,’ ‘intrusion detection system (IDS)’, and ‘alle- +e cost of power also plays an important role in en- viation,’ are the first, second, and third tier, respectively. suring the stability of the distributed power systems. +e +rough shielded communications and data safeguarding authors in [24] have proposed a decentralized SG control measures, the first tier ensures protection of the commu- model to ensure demand-side management in the grid by nication channel against the majority of CDIAs. In the event analyzing the electricity price versus grid frequency devia- of the protection tier getting violated, the IDS as the second tion. +e authors have also implemented an optimized data tier of defense detects the intrusion and generates precau- matching ML technique and the transparent open box tionary signals for the operators to take up preventive learning model to realize dynamic SG stability. measures against the CDIA. +e recent literature enumer- Against the backdrop of the above discussion on various ates many ML-based IDSs [11–17]. +e authors in [11, 12] factors of smart grid instability and the possible intelligent have depicted the application of ML algorithms in identi- methods to mitigate them, it is also essential that there fying unscrupulous user activities in smart grid commu- should be an efficient scheme for the estimation and pre- nication channels. +e authors in [13] have demonstrated diction of SG stability in place. +e authors in [25] have the utility of several ML algorithms in the detection of proposed an ML-based smart grid stability forecasting CDIAs at the physical layer of a smart grid. +e authors in scheme. +e proposed method employs three different ge- [14] have employed a support vector machine (SVM) netic algorithms in the future selection stage and four dif- classifier in the above context. +e authors in [15] have ferent ML classifiers including the GBM algorithm. +e proposed another ML-based model to detect time syn- authors in [5] proposed a few DL-based models such as chronization assault (TSA). +e authors in [16] proposed the recurrent neural network (RNN), long short-term memory Euclidean-distance-based ML model to predict the CDIAs. (LSTM), and gated recurrent unit (GRU) for prediction of +e authors in [17] have suggested a genetic algorithm (GA) SG stability. combined with SVM, for future selection and classification It can be inferred very well from the above survey that purposes, respectively, in order to detect CDIAs. Most of the there is no single comprehensive solution to tackle the CDIA recognition techniques using machine learning as challenges of stability in an SG. Also, there is a need for stated in the collected works have considered supervised further research in the area of stability prediction as evident learning on labeled data only. +e third defense tier, known from the scarcity of the literature in this regard. In the as alleviation, serves as a kind of restoration system, which present study, it is envisaged to design the novel memetic helps restore the reliable system operations once the CPS- algorithm-based ELM model for smart grid stability pre- assault recognition message is established at the SG control diction. +e proposed approach shows better results than the 4 International Transactions on Electrical Energy Systems traditional ELM technique and other advanced ML and DL (Equation (4)) is the weight matrix representing the weights techniques. between hidden layer neurons and out neurons, and Equation (5) is the prediction matrix. 3. Proposed TLBO-SA-ELM Approach st � Houtput ×β, (2) +e performance of the ML model is generally enhanced by f b + op × w 􏼁 . . . f b + op × w 􏼁 1 1 1 L 1 L integrating the optimization algorithm [26–29]. In this ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ Houtput �⎢ . . . . . . . . . ⎥ , work, a memetic algorithm based on TLBO and SA is ⎢ ⎥ ⎢ ⎥ ⎣ ⎦ proposed to increase the performance of the ELM classifier. f b + op × w 􏼁 . . . f b + op × w 􏼁 1 N 1 L N L N×L Afterward, the proposed optimized ELM is trained and (3) tested using the SG dataset. A single metaheuristic algorithm cannot solve all opti- β �􏼂β ,β , . . . ,β 􏼃 , (4) 1 2 L L×1 mization problems and is usually less effective for high- dimensional SG datasets. +erefore, there always remains a 􏽢 􏽢 􏽢 􏽢 (5) possibility to design an improved search approach and st �􏼂st ,st , . . . ,st 􏼃 . 1 2 N N×1 develop novel optimization techniques which can be used to In this work, the hyperparameters of ELM are optimized solve gene selection problems in the SG dataset. +is is our with the memetic version of teaching-learning based opti- major inspiration behind the scheming of this proposed mization (TLBO) [31]. Here, the TLBO is integrated with memetic algorithm. +e proposed SG stability predictive simulated annealing [32] in order to avoid a locally optimal algorithm is designed using three basic soft computing solution. models such as TLBO, SA, and ELM. In this work, the domains of the considered hyper- In this work, an ELM-based model [30] with optimized parameters are as follows: activation function (f), alpha (α), parameters is developed for efficient prediction of smart grid and several hidden neurons (nh). Here, f is the mathematical status from smart grid operational features. +e proposed equation that is responsible to determine whether the neuron problem can be visualized as an optimization problem where input is significant for prediction. α is the controlling pa- the objective is to select the best p � 􏼈f , α , nh 􏼉 in P � i i i i rameter for the adjustment of weights and H is the number of p , p , . . . , p (population with ‘n’ number of hyper- 􏼈 􏼉 1 2 n th hidden neurons that highly impact the performance and parameter sets). Here, p is the i randomly generated network stability. +is work is focused on the process of hyperparameter value set which is drawn from an allowed finding optimal parameters (P ) ELM by using TLBO with range of values as follows: f ∈ list[1, 2, 3, 4, 5, 6, 7, 8], SA. +e objective function is used to evaluate the parameter nh ∈ range[1, 200], and α ∈ range[0.1, 1.0]. Here, f , nh , i i i i th combinations (P ) and output a fitness (accuracy) and α are i activation function, a selected number of the fit � ELM(D , P ) which indicates how well the set of hidden layer, and learning rate. +e activation function f is P Train i hyperparameters performs for the considered problem. For chosen as ‘1,’ ‘2,’ ‘3,’ ‘4,’ ‘5,’ ‘6,’ ‘7,’ and ‘8’ for sine, tanh, the present problem, we have considered ‘accuracy’ as the tribas, sigmoid, hand/lim, soft/lim, Gaussian, and multi- evaluation matrix, and it is the objective to maximize the quadric, respectively. +e performance of ELM on the objective function presented in Equation (2). +e architecture prediction of smart grid status is dependent on these pa- of the proposed algorithm is presented graphically in Figure 1. rameters f, nh, and α. Here, the studied problem can be +e proposed scheme for hyperparameter optimization visualized as an optimization problem to get optimal P � starts with random generation of the population of the f , nh , α in P, which is the optimal parameter set of ELM 􏼈 􏼉 i i i hyperparameter set P � 􏼈P 􏼉 , where P � 􏼈f , nh , α 􏼉 i i i i i for solving the identification of the various states of the i�1 th representing i instance of the ELM parameter set. +e smart grid. On the given operational SG properties, m n fitness of each P in P is evaluated (in Algorithm 1, i.e., fit � i P D � 􏼈op , st 􏼉 , P � 􏼈P 􏼉 , and modelELM(D , P ), i i i Train i i�1 i�1 ∗ ELM(D , P ) by setting the P in ELM and testing on the where the objective is to find optimal P which optimizes the Train i i data D . +en, the fittest P in P is P . +e teaching Train i Teacher following objective function: factor tf � rand(1 + rand(1)) and the population P � arg max 􏼈fit � score st, st � ELM D , P􏼁 􏼁􏼉 i P ∈P i Train i meanP are computed in order to use them for the Mean new generation of a new population. For each P , the P is i i new ⎧ ⎨ ⎫ ⎬ generated as P � P + r × (P − tf × P ), where r i i Teacher Mean 􏽢 􏽢 � arg max fit � score(st, st) � 􏽘 I st , st􏼁 . P ∈P i i i i ⎩ ⎭ is a random number. +e updating of P is performed by i�1 new comparing each of P in P with the corresponding P in i i (1) new P . +en, the resultant population is improvised by choosing two solutions P and P randomly from P and +e dataset D � op , st is the collection of smart i j 􏼈 􏼉 i i i�1 altering them as mentioned in the following equation: grid operational properties op with thirteen features and one class label st representing the status of the grid con- new ⎫ ⎪ dition, i.e., either ‘stable’ or ‘unstable.’ +e proposed ELM ⎪ P � P + rand(1) × 􏼐P − P 􏼑 if 􏼒fit < fit 􏼓 i j i P P i j model has been trained with these instances. In ELM, the . (6) new prediction of the class label is made using Equation (2), P � P + rand(1) × 􏼐P − P 􏼑 Otherwise i j i j where Houtput (Equation (3)) is the output matrix, β International Transactions on Electrical Energy Systems 5 Initiate the population Generate a random solution of ELM parameter P' from P and N(P ) n best best best P = {P } , P = {f ,  , Nh } i i=1 i i i i Compute the fitness of Evaluate each P in P i P and P'' as fit and fit best bset best bset fit = ELM(D , P ) i Train i Select P with highest fitness (Accuracy) as P teacher Yes No if Compute P Compute tf = round(1+rand(1)) mean fit > fit best best new Compute P for each P new P = p + r × (P − tf + P ) i i teacher mean compute p = e update P by comparing each P in P new new within each P in P Generate a random number r for each randomly choosen P and P i j Inprove them by equation (6) ' r < p P = P best best find the best P i.e. P best Decrease the update P temperature T No No If Max. Yes trail reached? Max. Generation Reached Yes Return P best Return best solution P* Figure 1: Proposed system architecture. After improvisation of P, the best solution P in P is After getting the update P from Algorithm 2 (sim- best best selected based on the highest fitness. +en, the simulated ulated annealing process), the population P has been annealing (Algorithm 2) has been applied on P to generate updated by replacing old P with new P (returned from best best best a new solution from P in order to avoid local optimal the procedure SA). If a maximum generation is reached or best solutions. +e proposed approach of applying simulated no further improvement on the performance, then we stop annealing with TLBO not only accepts the best solution but and assign P � P . Finally, Algorithm 3 returns the best best also considers the nearer solution to the best solution with hyperparameter set P from the entire final population P. some probability. +e major steps of applying simulated +en, this is set on ELM as fit ∗ � ELM(D , P ) and P Train annealing on P are as follows: (i) random generation of a fit ∗ � ELM(D , P ) to get the generalized performance best P Train solution P from P and N(P ), where it is the on training data D and test data D , respectively. +e best best best Train Test neighborhood operation; (ii) computation of the fitness fit complete process of getting an optimal hyperparameterP best ′ ′ and fit of P and P , respectively, by calling the can be realized in Algorithm 3. best best best procedure ELM(D , P ) and ELM(D , P ), re- Train best Train best ′ ′ spectively; (iii) replacement of P with P if fit > fit , best best best best 4. Results and Discussion θ/T or if r and(1)< p, where p � e ; (iv) decrement of the temperature: T � 0.93 × T and checking the maximum trial. In this section, the proposed SG stability prediction method If maximum trial is reached, then extract the P . is evaluated through proper experimental analysis. best 6 International Transactions on Electrical Energy Systems 4.1. Experimental Platform. +e experimental setup for the Predicted proposed work comprises an online graphical processing stable TS FU unit (GPU) enabled by Google called “Google Colab,” a supercomputer having a Windows 10 Operating system with a core I7 processor and a Python 3.9 programming tool. Unstable FS TU 4.2. Dataset Description. +e SG data used in this experi- ment have been extracted from the UCI-ML repository [6]. +is dataset comprised 10000 samples with fourteen fea- TP: Truly Stable tures. +e attributes are divided into 12 primary predictive FN: Falsely Unstable features and 2 dependent variables. +e predictive features FP: Falsely Stable provide information about the reaction time (tau[x]) of TN: Truly Unstable smart grid participants {range: 0.5–10s}, nominal power S: Stable (p[x]) consumed (negative)/produced (positive), and co- U: Unstable efficient related to elasticity price (g[x]). +e dependent Figure 2: An example of the confusion matrix. attributes can be described as follows: the first attribute indicates: the maximal real part of the characteristic equa- tion root (if positive - the system is linearly unstable) and the F2_Score (F1): in the calculation of F1_Score, less weight second dependent attribute states the stability label (class is given for precision compared to the recall value. +erefore, level) of the system (categorical: stable/unstable). in equation (10), a beta (β) value of 2 can be considered for the calculation of F1_measure. So, it can be represented as 4.3. Performance Evaluation Matrices. In this work, the follows: classification result corresponding to the proposed TLBO- 5 R∗ P SA-ELM is interpreted through the following performance (12) F2 � × . evaluation matrices. 4 R + P Confusion matrix (CM): it represents the actual value with respect to the predicted values corresponding to the class level. 4.4. Performance Evaluation of the Proposed TLBO-SA-ELM CM is a table that visualizes and compares the result Model. +e collected SG data are divided into 8 : 2 ratios for obtained from a classifier by presenting the actual values and training and testing, respectively. +e testing result in terms predicted values (number of samples) with respect to the of the confusion matrix is presented in Figure 3. class levels in terms of correct and incorrect prediction. A Figure 3 shows the confusion matrix for ELM and the sample of the confusion matrix is presented in Figure 2. proposed model for the classification of the SG stability From the above confusion matrix, the following six dataset. +e proposed TLBO-SA-ELM model detects 703 matrices are calculated as presented from the following data as truly stable and 1292 as truly unstable, whereas equations. traditional ELM detects 669 data as truly stable and 1223 as Accuracy (A): it signifies the correctness of the classifier. truly unstable. Only 5 data are misclassified through the It is mathematically represented as follows: proposed model. From these confusion matrixes, different TS + TU measuring indices such as precision, recall, accuracy, and (7) A � , Fβ-Score are calculated and tabulated in Table I. It can be TS + TU + FS + FU seen from the table that the proposed model can classify the TS stability of the SG data with 99.75% accuracy. +e other (8) Precision(P): P � , indices such as precision, recall, F1, F2 scores, and area TS + FU under the curve (AUC) are noted to be 1.0, 0.996144, TS 0.998068, 0.6197, and 0.998072, respectively. It can be seen (9) Recall(R): R � , TS + FS from the table that the proposed model outperforms tra- ditional ELM with respect to all the measuring indices. R∗ P +e proposed approach is a hybrid model where the Fβ Score: F � 􏼐1 + β 􏼑 × . (10) β (R + P) parameter of the ELM is optimized through a memetic algorithm comprised TLBO and SA. In order to check the F1_Score (F1): in the calculation of F1_Score, equal effectiveness of the proposed memetic algorithm, a few other weight is given for both precision and recall. +erefore, in optimization algorithms such as genetic algorithm (GA), equation (10), a beta (β) value of 1 can be considered for the particle swarm optimization (PSO), and TLBO are hy- calculation of F1_measure. So it can be represented as bridized with the ELM algorithm, and the corresponding follows : result is interpreted in Figure 4. Figure 4 shows the plot of R∗ P fitness (accuracy) versus the number of generations. It is (11) F1 � 2 × . seen from the figure that the proposed memetic algorithm- R + P stable Unstable True International Transactions on Electrical Energy Systems 7 Predicted Predicted stable 669 34 stable 703 0 Unstable 74 1223 Unstable 5 1292 stable Unstable stable Unstable (a) (b) Figure 3: Confusion matrix for (a) traditional ELM and (b) proposed TLBO-SA-ELM. 0 510 15 20 25 30 35 40 Generations ELM-GA ELM-TLBO ELM-TLBO-SA ELM-PSO Figure 4: Fitness changes in various generations. th Let D � op , st be the smart grid operational data, where op � op , op , . . . , op is the i instance of the smart grid 􏼈 􏼉 􏽮 􏽯 i i i�1 i i,1 i,2 i,N operational feature value and st is the status of the grid. (1) Randomly generate bias b , i � 0 to L, and weight w , i � 0 to L. i i (2) Calculate the hidden layer output function Houtput by using the selected activation function f(·). f(b + op × w ) ... f(b + op × w ) 1 1 1 L 1 L ⎢ ⎥ ⎡ ⎢ ⎤ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ Houtput � ⎣ ... ... ... ⎦ . f(b + op × w ) ... f(b + op × w ) 1 N 1 L N L N×L (3) Compute the output weight matrix β � Houtput × st, which maximizes the objective function 􏽢 μ ‖Houtput × β − st‖ � min ‖Houtput × β − st‖. Here, Houtput is the Moore–Penrose generalized inverse of the Houtput. μ −1 T Houtput � (Houtput × Houtput) × Houtput . k 1 if (st � k) 􏽢 􏽢 i 􏽢 􏽢 Perform the prediction by using β on the data st � α × ( Houtput × β), α is the learning rate. st � 􏼨 􏼩, −1 if (st ≠ k) k � 1, 2, . . . , c c k 􏽢 􏽢 Predict the final class label as st � arg max (st ) i k�1,2,...,c i Return score of the prediction fit � score(st, st) ALGORITHM 1: fit ⟵ ELM(X, P ). P i (1) Randomly generate a solution P from P and N(P ) (neighbor structure of P ) best best best best ′ ′ (2) Compute the fitness fit and fit of P and P , respectively, by calling the procedure ELM(D , P ) and best best best best Train best ELM(D , P ), respectively Train best ′ ′ (3) If fit > fit , then assign P � P best best best best θ/T Else calculate p � e . If r and(1)< p, then assign P � P best best (4) Decrease the temperature, T � 0.93 × T (5) If the maximum trial is reached, then return P best (6) Else go to step 1. ALGORITHM 2: P � SA(P , N(P ), D ). best best best Train Fitness (Accuracy) True True 8 International Transactions on Electrical Energy Systems th Let D � 􏼈op , st 􏼉 be the smart grid operational data, where op � 􏽮op , op , . . . , op 􏽯 is the i instance of the smart grid i i i�1 i i,1 i,2 i,N operational feature value and st is the status of the grid. th (1) Randomly generate the population P � 􏼈P 􏼉 , where P � 􏼈f , nh , α 􏼉 representing i is an instance of the ELM parameter set. i i�1 i i i i (2) Evaluate the fitness of each P in P. fit � ELM(D , P ). i P Train i (3) Select fittest P in P as P . i Teacher (4) Compute teaching factor tf � r and(1 + rand(1)) and population meanP . Mean new new (5) Generate P from P in P: P � P + r × (P − tf × P ), where r is a random number. i i i i Teacher Mean new new (6) Perform the updating P by comparison of each P in P with the corresponding P in P . i i (7) For each pair of randomly chosen P and P in P, improvise them by using the following equation: i j new P � P + r and(1) × (P − P ) if (fit < fit ) i j i P P i j 􏼩. new P � P + r and(1) × (P − P ) Otherwise i j i j (8) Find the best solution P in P based on the highest fitness. best (9) Generate a new solution from P by using a simulated annealing process P � SA(P , N(P ), D ). best best best best Train (10) Update the population by replacing old P with new P (returned from the procedure SA). best best (11) If a maximum generation is reached, then assign P � P . best Else, go to step 2. (12) Return P . ALGORITHM 3: P ←ELM − TLBO − SA(D , P). Train Table 1: Performance metrics of prediction models. Performance Metrics Prediction Models Precision Recall F1 Score F2 Score ROC-AUC Accuracy DT 0.7488 1.0 0.8871 0.702315 0.61 74.88 NB 0.6485 1.0 0.786775 0.585075 0.5 64.85 LR 0.6485 1.0 0.786775 0.585075 0.5 64.85 RF 0.946227 0.909020 0.927251 0.5795 0.906857 90.75 XGBoost 0.910437 0.995373 0.951012 0.5944 0.907359 93.35 ELM 0.972951 0.942945 0.957713 0.5986 0.947290 94.6 Proposed model 1.0 0.996144 0.998068 0.6197 0.996086 99.75 Table 2: Performance comparison with existing approaches. Model Accuracy 80 Recurrent neural network (RNN) [5] 96.60 Gated recurrent unit (GRU) [5] 97.30 Long short term memory (LSTM) [5] 97.13 Multidimensional long short term memory (M-LSTM) 99.07 [5] Proposed model (ELM-TLBO-SA) 99.75 based ELM converses quickly (within 14 generations) compared to other approaches. Figure 5: Overall comparison of studied models. 4.4.1. Comparative Analysis. In order to validate the per- formance of the proposed algorithm applied to the SG the training and testing dataset is considered during these experimentations. stability dataset, the corresponding experimental results are compared with the output of several other contemporary ML It can be analyzed from the table that NB and LR show the worst performance (accuracy: 64.85) compared to models. In this regard, the results corresponding to a de- cision tree (DT), na¨ıve bias (NB), linear regression (LR), XGBoost having an accuracy of 93.35%. It can also be an- alyzed that the performance of XGBoost is observed to be random forest (RF), and extreme gradient boosting (XGBoost) are depicted in Table 1. A similar ratio (8 : 2) of less than the proposed model with a very high margin Accuracy NB LR DT XGBoost RF ELM RNN GRU LSTM M-LSTM Proposed Method International Transactions on Electrical Energy Systems 9 (accuracy> 6.4%). Finally, it can be concluded that the TSA: Time synchronization assault proposed approach performs better than all other ML GA: Genetic algorithm models with respect to each performance indices. In addi- DNN: Deep neural network tion to this, the performance of the proposed approach is IoT: Internet of thing compared with the result of different deep learning models BN: Bayesian network (such as RNN, GRU, LSTM, and MLSTM) applied to a RNN: Recurrent neural network similar dataset [5]. +is comparative result analysis is pre- GRU: Gated recurrent unit sented in Table 2. It can be analyzed from the table that LSTM: Long short-term memory MLSTM performs better than RNN, GRU, and LSTM; CM: Confusion matrix however, the proposed algorithm (TLBO-SA-ELM) out- f: Activation function performs in all aspect. Figure 5 shows the overall com- α: Alpha parative results graphically. nh: Number of hidden neuron P : Hyperparameter 5. Conclusion D : Training data Train D : Test data. Test Smart grids are identified with the cyber-physical system used for intelligent management of power generation and Data Availability dissipation ensuring quality power supply at the most economical price. However, the threat to CPS and several +e data used in this study are openly available in (UCI-ML issues including a threat to CPSs severely affects the stability Repository) at (https://archive.ics.uci.edu/ml/datasets/ of a smart grid. Machine learning techniques play a vital role Electrical+Grid+Stability+Simulated+Data+) [6]. in predicting the stability of the smart grid. In this work, a novel memetic algorithm-based ELM model is introduced to predict the stability of the SG. +e proposed optimized ELM Conflicts of Interest algorithm is tested on the SG dataset extracted from the +e authors declare that they have no conflicts of interest. UCI-ML repository. +e proposed model achieved 99.75% accuracy and 100% precision in classifying the stability status of the smart grid. To justify the effectiveness of the References proposed model, a detailed comparative analysis is carried [1] Z. M. Fadlullah and N. Kato, Evolution of Smart Grids, out with other contemporary machine learning models Springer, Berlin, Germany, 2015. (such as decision tree, linear regression, random forest, na¨ıve [2] B. Patnaik, M. Mishra, R. C. Bansal, and R. K. Jena, “AC bias, XGBoost, and ELM) and previously published work microgrid protection–A review: current and future pro- based on deep learning (such as LSTM, GRU, RNN, and spective,” Applied Energy, vol. 271, Article ID 115210, 2020. MLSTM). 99.8072% of ROC is achieved through the pro- [3] 2021, https://www.nist.gov/el/cyber-physical-systems. posed algorithm which outperformed the other comparative [4] R. Rajkumar, I. Lee, L. Sha, and J. Stankovic, “Cyber-physical models. +e effectiveness of the proposed memetic algo- systems: the next computing revolution,” in Proceedings of the rithm based on TLBO and SA is additionally compared with Design Automation Conference, pp. 731–736, Anaheim, CA, several other optimization techniques (such as particle swam USA, June 2010. [5] M. Alazab, S. Khan, S. S. R. Krishnan, Q. V. Pham, optimization, genetic algorithm, and TLBO) hybridized with M. P. K. Reddy, and T. R. Gadekallu, “A multidirectional ELM. +e future scope of the work may deal with the LSTM model for predicting the stability of a smart grid,” IEEE analysis of dynamic power requirements, reliability, and Access, vol. 8, pp. 85454–85463, 2020. resiliency of SG. [6] V. Arzamasov, K. Bohm, ¨ and P. Jochem, “Towards concise models of grid stability,” in Proceedings of the 2018 IEEE Abbreviations International Conference on Communications, Control, and Computing Technologies for Smart Grids (SmartGridComm), SG: Smart grid pp. 1–6, IEEE, Aalborg, Denmark, October 2018. CPS: Cyber-physical system [7] D. Zhang, X. Han, and C. Deng, “Review on the research and EUI: Electrical utility industry practice of deep learning and reinforcement learning in smart grids,” CSEE Journal of Power and Energy Systems, vol. 4, ELM: Extreme learning machine no. 3, pp. 362–370, 2018. TLBO: Teaching-learning-based optimization [8] S. Zerdoumi, A. Q. M. Sabri, A. Kamsin et al., “Image pattern SA: Simulated annealing recognition in big data: taxonomy and open challenges: EPS: Electrical power system survey,” Multimedia Tools and Applications, vol. 77, no. 8, RES: Renewable energy resources pp. 10091–10121, 2018. DG: Distributed generation [9] M. Ghorbanian, S. H. Dolatabadi, and P. Siano, “Big data ML: Machine learning issues in smart grids: a survey,” IEEE Systems Journal, vol. 13, DL: Deep learning no. 4, pp. 4158–4168, 2019. CDIA: Covert data integrity assault [10] S. Ahmed, Y. Lee, S.-H. Hyun, and I. Koo, “Unsupervised IDS: Intrusion detection system machine learning-based detection of covert data integrity SVM: Support vector machine assault in smart grid networks utilizing isolation forest,” IEEE 10 International Transactions on Electrical Energy Systems Transactions on Information Forensics and Security, vol. 14, Intelligent Computer Communication and Processing (ICCP), no. 10, pp. 2765–2777, 2019. pp. 175–182, Cluj-Napoca, Romania, September 2019. [11] Z. M. Fadlullah, M. M. Fouda, N. Kato, X. Shen, and [26] M. Elsisi, “Optimal design of nonlinear model predictive Y. Nozaki, “An early warning system against malicious ac- controller based on the new modified multitracker optimi- zation algorithm,” International Journal of Intelligent Systems, tivities for smart grid communications,” IEEE Network, vol. 35, no. 11, pp. 1857–1878, 2020. vol. 25, no. 5, pp. 50–55, 2011. [27] M. Elsisi, “Improved grey wolf optimizer based on opposition [12] Y. Zhang, L. Wang, W. Sun, R. C. Green II, and M. Alam, and quasi learning approaches for optimization: case study “Distributed intrusion detection system in a multi-layer autonomous vehicle including vision system,” Artificial In- network architecture of smart grids,” IEEE Transactions on telligence Review, pp. 1–24, 2022. Smart Grid, vol. 2, no. 4, pp. 796–808, 2011. [28] R. M. Adnan, R. R Mostafa, O. Kisi, Z. M. Yaseen, S. Shahid, [13] M. Ozay, I. Esnaola, F. T. Yarman Vural, S. R. Kulkarni, and and M. Zounemat-Kermani, “Improving streamflow predic- H. V. Poor, “Machine learning methods for attack detection in tion using a new hybrid ELM model combined with hybrid the smart grid,” IEEE Transactions on Neural Networks and particle swarm optimization and grey wolf optimization,” Learning Systems, vol. 27, no. 8, pp. 1773–1786, 2016. Knowledge-Based Systems, vol. 230, Article ID 107379, 2021. [14] M. Esmalifalak, L. Liu, N. Nguyen, R. Zheng, and Z. Han, [29] L. Yang, Y. Li, and Z. Li, “Improved-ELM method for “Detecting stealthy false data injection using machine learning detecting false data attack in smart grid,” International in smart grid,” IEEE Systems Journal, vol. 11, no. 3, Journal of Electrical Power & Energy Systems, vol. 91, pp. 1644–1652, 2017. pp. 183–191, 2017. [15] J. Wang, W. Tu, L. C. Hui, S. Yiu, and E. K. Wang, “Detecting [30] G. B. Huang, Q. Y. Zhu, and C. K. Siew, “Extreme learning time synchronization attacks in cyber-physical systems with machine: theory and applications,” Neurocomputing, vol. 70, machine learning techniques,” in Proceedings of the IEEE 37th no. 1-3, pp. 489–501, 2006. International Conference on Distributed Computing Systems, [31] R. Rao, V. Savsani, D. Vakharia, and D. P. Vakharia, pp. 2246–2251, ICDCS, Atlanta, GA, USA, June 2017. “Teaching–learning-based optimization: a novel method for [16] S. Ahmed, Y. Lee, S.-H. Hyun, and I. Koo, “Feature selec- constrained mechanical design optimization problems,” tion–based detection of covert cyber deception assaults in Computer-Aided Design, vol. 43, no. 3, pp. 303–315, 2011. smart grid communications networks using machine learn- [32] L. Liu, H. Mu, J. Yang, X. Li, and F. Wu, “A simulated ing,” IEEE Access, vol. 6, pp. 27518–27529, 2018. annealing for multi-criteria optimization problem: Dbmosa,” [17] S. Ahmed, Y. Lee, S. H. Hyun, and I. Koo, “Covert cyber Swarm and Evolutionary Computation, vol. 14, pp. 48–65, assault detection in smart grid networks utilizing feature selection and euclidean distance-based machine learning,” Applied Sciences, vol. 8, no. 5, p. 772, 2018. [18] C. Chu, “Using machine learning to predict renewable energy generation and smart grid designs,” M.Sc +esis, Princeton University, Princeton, NJ, USA, 2019. [19] R. Vinayakumar, M. Alazab, K. P. Soman, P. Poornachandran, A. Al-Nemrat, and S. Venkatraman, “Deep learning approach for intelligent intrusion detection system,” IEEE Access, vol. 7, [20] D. Syed, S. S. Refaat, and H. Abu-Rub, “Performance eval- uation of distributed machine learning for load forecasting in smart grids,” in Proceedings of the IEEE Cybernetics & In- formatics (K&I), pp. 1–6, Velke Karlovice, Czech Republic, January 2020. [21] N. Bassamzadeh and R. Ghanem, “Multiscale stochastic prediction of electricity demand in smart grids using bayesian networks,” Applied Energy, vol. 193, pp. 369–380, 2017. [22] M. Elsisi and M. Q. Tran, “Development of an IoTarchitecture based on a deep neural network against cyber attacks for automated guided vehicles,” Sensors, vol. 21, no. 24, p. 8467, [23] M. Elsisi, M. Q. Tran, K. Mahmoud, D. E. A. Mansour, M. Lehtonen, and M. M. Darwish, “Effective IoT-based deep learning platform for online fault diagnosis of power trans- formers against cyberattacks and data uncertainties,” Mea- surement, vol. 190, Article ID 110686, 2022. [24] D. A. Wood, “Predicting stability of a decentralized power grid linking electricity price formulation to grid frequency applying an optimized data matching learning network to simulated data,” Technology and Economics of Smart Grids and Sustainable Energy, vol. 5, no. 1, pp. 3–21, 2020. [25] D. Moldovan and I. Salomie, “Detection of sources of in- stability in smart grids using machine learning techniques,” in Proceedings of the IEEE 15th International Conference on

Journal

International Transactions on Electrical Energy SystemsHindawi Publishing Corporation

Published: Aug 31, 2022

References