A 6mart 9oltage 6tability 0aneuver $lgorithm IRU9oltage &ollapses 0itigation Mohamed Usama 1, Hoda K.Mohamed 1, Islam A.M. El-Maddah 1, M.Amer Shedied 2 1 (Computer and Systems Engineering Department, Ain Shams University, Cairo-Egypt) 2 (Computer Science Department, Future Academy, Cairo-Egypt) 1 moh.usama88@yahoo.com , hoda.korashy@eng.asu.edu.eg , islam.elmadah@eng.asu.edu.eg 2 mo_amer2002@yahoo.com decades [1]. In general, the stability of a system can be defined as the ability of a system to return to its original condition after being subjected to a disturbance and remain its condition over time [2]. There are several types of instability that could face the system operator such as voltage instability, frequency instability, and inter-area oscillations. In V.S.A.M.A we only concerned about handling the voltage instability issue using machine learning techniques by developing an early warning system and an automatic maneuver system that can mitigate the power outage. We can define the voltage stability [3] as the process of maintaining acceptable voltages level at all buses after being subjected to a disturbance. We can achieve that stability by balancing between the supplies from the generation stations and the load demands of the consumers. This balance could be lost when a heavy reactive loaded system happens. To evaluate the voltage stability of the power grid, there are several mathematical indices [8] that can indicate how close the system to the failure or voltage collapses. In [4] voltage stability indices are used to find out the weak buses and use these results as an overall evaluation of the power grid .this could be used as an early warning system for blackouts. After obtaining an overall evaluation of the power grid, we can then propose different control conditions and evaluate them to choose the most suitable one to enhance the voltage stability. In [5] a novel index had proposed to evaluate the power grid and proposed voltage stability control strategy to improve the system-level voltage stability. We can use artificial intelligence techniques for voltage stability evaluation such as the artificial neural network (A.N.N), Deep learning and machine learning instead of mathematical indices. In [6] a neural network was used instead of ordinary mathematical methods to evaluate the voltage stability of the power grid. Abstract— In the power system, the instantaneous and permanent stability is a major requirement cannot be overlooked. Because of the power grid large-scale systems, any disturbance anywhere on the power grid could pose a reason of overall dynamic imbalances. Major consequences could be occurred to the electricity feeding across wide areas of country which is called partial blackout, even entire country which is called overall blackout. It is perhaps for this reason that the existence voltage stability indices which indicate the power grid system stability level is very essential .With knowing the voltage stability level of the transmission lines that involves the power grid in real time ( online operation) , the voltage stability of the entire power grid could be obtained easily . There are several mathematical base voltage indices. But in this proposal, another voltage stability index will be build based on the machine learning techniques to mitigate the voltage collapse phenomenon. This novel predictor is proposed in transient stability analysis based on machine learning techniques such as (Linear regression, neural network, and Decision tree). This predictor is built after a comparison was made between the impacts of various machine learning algorithms using different datasets. Three different mathematical voltage stability indices (FVSI, Lmn, and NLSI) had been used to prepare datasets for the training purpose. An early warning system had been built based on the proposed predictor. This early warning system could be used to inform the system operator with the hazards of voltage instability issues face the electric power grid and visualize these hazards. The E.W.S had then been used as a kernel to build V.S.A.M.A (Voltage Stability Automatic Maneuver Algorithm) that can handle the voltage instability issue. Index Terms— Decision tree, Early warning, Line stability index, Linear regression, Maneuver algorithm, Neural network, Predictive analysis , Predictive modeling , Predictors , RapidMiner programming, Voltage stability. I. INTRODUCTION In V.S.A.M.A, A novel voltage instability predictor will be proposed as a kernel of an early warning system and automatic maneuver algorithm that target to mitigate the power outage via suggesting suitable operating conditions. The impacts of these operating conditions are valuated using the predictor which is based on machine learning algorithms. Section two will discuss the background needed to complete this work including voltage stability indices used to evaluate the voltage stability, machine learning use in this proposal and some In the modern power system, real-time monitoring and operating of the electric power grid are the main tasks of the system operator. These tasks had become more complex and difficult because of the wide geographical regions covered by the power grid, a massive calculations were done, the difficulty of the decision taken and also the hardness of action rollback. One of the major concern issues in the electric power system is the grid stability. The power instability issue caused many notable wide-scale power outages or Blackouts in recent ,((( Authorized licensed use limited to: Cape Peninsula University of Technology. Downloaded on September 05,2022 at 10:10:05 UTC from IEEE Xplore. Restrictions apply. other utilities such as load flow algorithms, early warning and visualizer. Section three will discuss methodology used to build the voltage instability predictor, E.W.S, Visualizer, and the V.S.A.M.A. In section four and five, the results had been achieved and the conclusion will be demonstrated. 3) Novel line stability index (NLSI) Yazdanpanah-Goharrizi derived a line stability index based on the same concept as Lmn and FVSI [8], [10], [12]. The NLSI equation is illustrated below in eq-3. As Lmn and FVSI, NLSI must remain less than 1 to ensure the stability and avoid the voltage collapses. II. BACKGROUND ܰ ܫܵܮൌ A. Voltage Stability Indices To build the predictor in V.S.A.M.A which is the main component, firstly we need to train the selected machine learning algorithm using a bulk dataset. These dataset can be obtained by evaluating the voltage stability of the power grid transmission lines In several situations mathematically via voltage stability indices such as fast voltage stability indices (FVSI), line stability index (Lmn), line stability factor (LQP), novel line stability index (NLSI),…etc. These indices will be applied separately to each transmission line of an interconnected network. The representation of a transmission line between two buses is illustrated in figure 1. In V.S.A.M.A we will use three different indices (FVSI, Lmn, and NLSI) and compare the predictor result for each index. 1) Fast voltage stability indices (FVSI) Musirin proposed a novel Fast Voltage Stability Index (FVSI) [7], [8], [9], and [10] simplified from the voltage quadratic equation at the sending end of a representation of a 2bus system. The FVSI equation is illustrated below in equation-1. The value of the FVSI must remain less than 1 to ensure the stability and avoid the voltage collapses. Ͷܼ ଶ ܳݎ ܸ ݏଶ ܺ C. Early warning and Visualization The United Nations International Strategy for Disaster Reduction (UNISDR) defined the early warning system in [13]. It is the set of capacities needed to generate and disseminate timely and meaningful warning information to enable individuals threatened by a hazard to prepare and to act appropriately to reduce the possibility of harm or loss. In the power system, the operator in the control center is the one who concerned with this utility. The system operator needs to be informed acknowledgment of hazards and risks connected with the voltage instability issue. This will help the operator to take suitable actions to avoid the power grid failure. These hazards can be visualized to the system operator in percentage form or gradually colored form to make the extraction of data easier and taking the decision faster and more accurate. ሺͳሻ 2) Line stability index (Lmn) Moghavemmi proposed a novel method [8], [11] that examined the line voltage stability by an index which is varied from zero to one. This index formulated using power transfer concept in the single line power transmission network. The novel index equation is illustrated below in eq-2. Also, the value of the Lmn index must remain less than 1 to ensure the stability and avoid the voltage collapses. ݊݉ܮൌ Ͷܳݎ ሺܸ݊݅ܵݏሺˁ െ ߜሻሻଶ ሺ͵ሻ B. Machine Learning One of the most important targets of V.S.A.M.A is to replace the mathematical voltage stability index with predictor based on machine learning algorithm and evaluate the stability of the electric power grid with this predictor. Different algorithms had been used to produce several predictors and compare results to obtain most accurate predictor. The algorithms of machine learning can be classified into three main categories 1) Clustering algorithms 2) Classification algorithms 3) Regression algorithms. The predictor needed to be built has continues output from 0 to 1. That is mean we can classify our problem as a regression problem. There are several regression machine learning algorithms [12] such as 1) Linear regression 2) Polynomial regression 3) Vector linear regression 4) Local polynomial regression 5) Neural network 6) Deep learning. Because of the narrow output range of our proposed predictor, we can convert our problem from regression problem to classification problem by discretizing the output with some acceptable resolution. Now we can also use the classification algorithms to build the predictor. There are several classification machine learning algorithms [12] such as 1) Decision tree 2) K-Nearest Neighborhood 3) Neural network. In V.S.A.M.A, three different indices (FVSI, Lmn and NLSI) will be used to prepare the training datasets. These datasets will be used to train three different learning machine algorithms (Linear regression, neural network, and Decision tree). A competition will be held between these three datasets once for each algorithm to produce three predictors. The most accurate predictor will be chosen as a kernel of the V.S.A.M.A. Figure 1: “The two bus representation of a power system” ܫܸܵܨൌ ܲ ܴݎ ܳܺݎ ͲǤʹͷܸ ݏଶ ሺʹሻ Authorized licensed use limited to: Cape Peninsula University of Technology. Downloaded on September 05,2022 at 10:10:05 UTC from IEEE Xplore. Restrictions apply. D. Load Flow The power flow or load flow analysis could be defined as [13] the study to predict the effect of some certain loading condition on the power system grid parameters. In proposed system we can look to the power flow analysis as block gets the line data, bus data and loading conditions and this block outputs the voltage magnitude and angle at each bus as illustrated in figure 2. that will solve the issue of massive computation resources needed and the limited time available during online analysis. The block diagram of the proposed system is illustrated in figure 4. Figure 4: “V.S.A.M.A block diagram” The methodology section consists of four subsections (Select case study, Prepare data, Build the predictor and V.S.A.M.A). The first section discusses the structure of appropriate electric power grid that will be used to test and evaluate our algorithm. The second section discusses how to extract the datasets from the testing environment for the algorithm training. The third section discusses three algorithms used to build three predictors and how to implement these algorithms on RapidMiner Studio to build the predictors. The last section discusses the procedure of building the V.S.A.M.A and developing it using Matlab program. Figure 2: “Load Flow Block Diagram” The rest of system parameters such as the real power and reactive power flows, current flow, voltage drop, power flow losses, .etc. can be driven easily from the voltage magnitude and angle. There are several methods to perform the load flow study such as Gauss iterative method and Newton- Rapson method. In V.S.A.M.A, we used Gauss-Seidel method [13] [14]. The flowchart of this method is illustrated in figure 3. A. Select case study The IEEE 9-Bus system selected as a test case. It represents a simple approximation of a portion of the Western System Coordinating Council (WSCC). It consists of 3 generators, 3 transformers, 3 loads, 9 buses and 6 lines. The bus data and line data are illustrated in table 1 and table 2. Also, the structure of the network illustrated in figure 5. Figure 5: “The IEEE 9-Bus system structure” Figure 3: “Gauss iterative method flowchart” Table 1: “Line data” III. METHODOLOGY The target of V.S.A.M.A is to design a predictor that can evaluate the voltage stability of the power network and use this predictor to build an early warning system for the outage and over blackout. Starting from this, we can build an automatic maneuver algorithm that will be a kernel for a fully automatic maneuver system. The maneuver algorithm can handle the voltage instability issue under the system operator supervision. The proposed system will be based on predictive techniques Line From Bus To Bus R X B 1 4 1 0 0.0576 0 2 2 7 0 0.0625 0 3 9 3 0 0.0586 0 4 5 4 0.010 0.0680 0.176 5 6 4 0.017 0.0920 0.158 6 7 5 0.032 0.1610 0.306 7 9 6 0.039 0.1738 0.358 8 7 8 0.0085 0.0576 0.149 9 8 9 0.0119 0.1008 0.209 Authorized licensed use limited to: Cape Peninsula University of Technology. Downloaded on September 05,2022 at 10:10:05 UTC from IEEE Xplore. Restrictions apply. Table 2: “Bus data” Table 3: “A sample of the labeled dataset” # Nom KV PU Volt Angle (Deg) Gen MW Gen Mvar Load MW Load Mvar 1 16.5 1.040 0 45.44 4.26 - - 2 18.0 1.043 8.86 158 4.9 - - 3 13.8 1.039 5.22 90 -11.45 - - 4 230 1.037 -1.39 - - - - 5 230 1.027 -1.77 - - 75 35 6 230 1.026 -2.95 - - 90 30 7 230 1.045 3.67 - - - - 8 230 1.031 0.84 - - 125 40 9 230 1.047 2.44 - - - - R (p.u) X (p.u) B (p.u) Nom Vs Mag Vs Angle Vs Nom Vr … VSI 0 0 0 0.01 0.017 0.032 0.039 0.0085 0.0119 0 0 0 0.0576 0.0625 0.0586 0.068 0.092 0.161 0.1738 0.0576 0.1008 0.0576 0.0625 0.0586 0 0 0 0.176 0.158 0.306 0.358 0.149 0.209 0 0 0 230 18 230 230 230 230 230 230 230 230 18 230 1.02313 1.025 1.03201 0.99241 1.01089 1.02504 1.03201 1.02504 1.01594 1.02133 1.025 1.03149 -3.797 6.531 0.067 -7.009 -5.427 0.966 0.067 0.966 -1.333 -3.805 6.53 0.053 16.5 230 13.8 230 230 230 230 230 230 16.5 230 13.8 … … … … … … … … … … … … 0.072146948 0.056632164 0.040984084 0.112662127 0.055924315 0.147371653 0.134235962 0.031732019 0.091555124 0.078840694 0.052572214 0.03894044 C. Build the predictor The predictor is an iterative trained machine learning algorithm. The concept of building predictor is to learn machine learning algorithm from a large number of inputs and its corresponding outputs and try to configure a relation between these data to estimate the output for new inputs as illustrated below in figure 7. Both input data gathered with the corresponding output data is called labeled dataset. B. Prepare data In machine learning, it is very important to feed our supervised algorithms with a huge amount of labeled data set to achieve better results. Labeled data can be defined as input data and its corresponding outputs. In V.S.A.M.A the output data is the evaluation of voltage stability and the inputs is the system parameters such as resistance, reactance, voltage magnitude and its angle from and to bus, active and reactive power from and to bus,…etc. For the predictor training, the input data can be collected through two stages. The first stage is to simulate the network in different loading conditions and get the corresponding system parameters for each run as illustrated in figure 6. The Power World Simulation tool is used to simulate two hundred different loading conditions to produce around one thousand three hundred different examples as a dataset. We can obtain these loading conditions by changing the reactive power and active power of loads and generators. The second stage is to calculate the corresponding outputs by ordinary mathematical methods and we use three different voltage stability indices (FVSI, Lmn and NLSI). The labeled dataset used for the predictor training are illustrated in table 3. Figure 7: “Machine learning workflow” A sophisticated platform is used for predictive analysis and modeling called RapidMiner Studio. It is a graphically programming language that analyses data using connected blocks without code. 1) Linear regression The first machine learning algorithm used is the linear regression. It is a statistical method that allows us to discover the relationship between one dependent variable called labeled attribute and a series of other changing variables. In V.S.A.M.A, the predicted variable is the voltage stability index and it is continuous variable from 0 to 1. A competition between three indices had been done to select the most compatible index with the linear regression algorithm. The process of building the predictor design using linear regression algorithm is illustrated in figure-8 and the predictor equation for the winning index is illustrated in figure 9. Figure 6: “The 9-Bus system implementation in power-world” Figure 8: “The process of building linear regression predictor in RapidMiner” Authorized licensed use limited to: Cape Peninsula University of Technology. Downloaded on September 05,2022 at 10:10:05 UTC from IEEE Xplore. Restrictions apply. Figure 12: “The process of building the decision tree predictor in RapidMiner” Figure 9: “The Linear regression predictor equation” 2) Neural Network The second machine learning algorithm used is the artificial neural network. Our predictor had been learned by means of a feed-forward neural network trained by a back propagation algorithm (multi-layer perceptron). A competition between three indices had been done to select the most compatible index with the neural network algorithm. The process of building the predictor based on neural network algorithm is illustrated in figure 10 .A sample of the network weights of the predictor for the winning index is illustrated in table 4. Figure 13: “Graphical representation of the decision tree predictor” D. V.S.A.M.A After we had built our predictor that can predict the level of voltage stability for a transmission line using machine learning technique instead of the ordinary mathematical method, we can now use it as a kernel of the V.S.A.M.A. That algorithm can handle the voltage instability issue by suggesting suitable loading conditions. Also, the predictor can be applied on all over the network and visualize the stability evaluation of it to the system operator. By setting certain setpoint we can build an early warning system that can notify the system operator with abnormal behavior and trigger the maneuver algorithm. The flow chart of the V.S.A.M.A is illustrated bellow in figure 13. Figure 10: “The process of building A.N.N predictor in RapidMiner” Figure 11: “The weighting arrays for the A.N.N model” 3) Decision tree The third machine learning algorithm used is the decision tree. As we mentioned before, the output of the predictor bounded between small ranges from zero to one. That is why we can consider our problem as a classification problem. The decision tree algorithm is a supervised classification algorithm in the form of a tree structure by starting at the root of the tree and moving through it until a leaf node. The predictor is a set of rules that can be graphically represented. A competition between three indices had been done to select the most compatible index with the decision tree algorithm. The process of building the predictor designed using decision tree algorithm is illustrated in figure 11 and the predictor graphical representation of the winning index is illustrated in figure 12. Figure 14: “The flow chart of the V.S.A.M.A” Authorized licensed use limited to: Cape Peninsula University of Technology. Downloaded on September 05,2022 at 10:10:05 UTC from IEEE Xplore. Restrictions apply. IV. RESULTS A. Predictor results Three different datasets used to train three different machine learning algorithms. All datasets have the same data except the voltage stability indices (FVSI, Lmn and NLSI) used as labeled attribute. The table-5 illustrates the accuracy by the RMS (Root Mean Square) value of each predictor using the three datasets. The graphs below (14, 15, and 16) show the actual value of voltage index on the x-axis and its predicted value on the y-axis for each winning index in each Predictor. As shown in the table 5, the more accurate predictor is the A.N.N with the NLSI labeled attribute dataset. Table 4: “Predictors accuracy Comparison” FVSI Lmn NLSI Linear Reg. 0.036 0.043 0.038 A.N.N 0.020 0.010 0.008 Decision tree 0.033 0.047 0.065 Figure 17: “Decision Tree Accuracy” B. Case study For V.S.A.M.A testing, we choose a snapshot of the IEEE 9-Bus system operating conditions that have one line of nine exceeded the preset instability warning boundary which is equal 0.5. As shown in figure 15, the visualizer informs the operator that line number four has voltage instability issue and the V.S.A.M.A had to start solving this problem. After the V.S.A.M.A had been triggered by the warning system, the V.S.A.M.A suggested ten different loading conditions and used the predictor to evaluate the stability of the overall network for these different scenarios. The V.S.A.M.A suggests ten different loading conditions each time until it achieves a loading condition that guarantees the stability of the overall network with fewer changes in the operation conditions. In our test case, the voltage stability evaluation of the network for each scenario is shown in Table 6. Scenario number five shows that all lines will be stable and the stability issue for line four will be solved by this suggestion and the operator should apply these loading conditions to achieve the stability. Figure 15: “Linear Regression Accuracy” Table 5: “Evaluation of the V.S.A.M.A Suggestions” 1 2 3 4 5 6 7 8 9 10 Figure 16: “A.N.N Accuracy” Line1 Line2 Line3 Line4 Line5 Line6 Line7 Line8 Line9 0.035 0.035 0.035 0.035 0.035 0.035 0.035 0.035 0.035 0.265 0.260 0.260 0.265 0.256 0.256 0.266 0.251 0.253 0.224 0.225 0.227 0.225 0.228 0.230 0.226 0.231 0.234 0.513 0.505 0.505 0.513 0.496 0.496 0.513 0.487 0.488 0.129 0.146 0.147 0.130 0.164 0.165 0.130 0.181 0.182 0.346 0.336 0.334 0.344 0.324 0.320 0.342 0.312 0.305 0.291 0.277 0.278 0.293 0.263 0.266 0.294 0.250 0.254 0.005 0.006 0.005 0.005 0.005 0.005 0.005 0.005 0.005 0.079 0.072 0.072 0.079 0.065 0.065 0.079 0.058 0.058 0.035 0.264 0.223 0.513 0.129 0.348 0.290 0.006 0.079 Authorized licensed use limited to: Cape Peninsula University of Technology. Downloaded on September 05,2022 at 10:10:05 UTC from IEEE Xplore. Restrictions apply. [5] Houhe Chen, Tao Jiang, Haoyu Yuan, Hongjie Jia, Linquan Bai, Fangxing Li. "Wide-area measurement-based voltage stability sensitivity and its application in voltage control." Elsevier, 2016. [6] "Voltage stability monitoring of power systems using reduced network and artificial neural network." Elsevier, 2016. [7] I. Musirin; T. K. Abdul Rahman. "Novel fast voltage stability index (FVSI) for voltage stability analysis in power transmission system." IEEE Conference Publications. IEEE, 2002. [8] Danish, Mir Sayed Shah. Voltage Stability in Electric Power System: A Practical Introduction. Logos Verlag Berlin, 2015. [9] Aziz Oukennou,Abdelhalim Sandali. "Assessment and analysis of Voltage Stability Indices in electrical network using PSAT Software." Power Systems Conference (MEPCON), 2016 Eighteenth International Middle East. IEEE, 2017. [10] M. Cupelli, C. Doig Cardet, A. Monti. "Comparison of line voltage stability indices using dynamic real time simulation." Innovative Smart Grid Technologies (ISGT Europe), 2012 3rd IEEE PES International Conference and Exhibition on. IEEE, 2013. [11] K. R. Hridya, V. Mini, R. Visakhan, Asha Anu Kurian. "Analysis of voltage stability enhancement of a grid and loss reduction using series FACTS controllers." Power, Instrumentation, Control and Computing (PICC), 2015 International Conference on. IEEE, 2015. [12] Isaac A. Samuel, James Katende, Claudius O. A. Awosope, Ayokunle A. Awelewa. "Prediction of Voltage Collapse in Electrical Power System Networks using a New Voltage Stability Index." International Journal of Applied Engineering Research, 2017. [13] Das, J. C. Power System Analysis Short-Circuit Load Flow and Harmonics. Taylor & Francis Group, LLC, 2012. [14] C.P. Salomon, Germano Lambert-Torres. "Hybrid Particle Swarm Optimization Approach for Load-Flow Computation." International journal of innovative computing, information & control: IJICIC, 2013. [15] A.Yazdanpanah-Goharrizi, R.Asghari. "A Novel Line Stability Index (NLSI) for Voltage Stability Assessment of Power Systems." 7th WSEAS International Conference on Power Systems. 2007. [16] Mohssen Mohammed, Muhammad Badruddin Khan, Eihab Bashier Mohammed Bashier. Machine Learning: Algorithms and Applications. CRC Press, 2016. [17] Ashbindu Singh, Zinta Zommers. Reducing Disaster: Early Warning Systems for Climate Change. Springer, 2014. [18] Glenn W. Stagg, Ahmed H.El-Abiad. Computer methods in power system analysis. McGraw-Hill, 1987. Figure 18: “Visualizer” V. CONCLUSION A voltage stability automatic maneuver algorithm based on machine learning had been presented. The main contribution is to propose a novel voltage stability predictor by introduce a comparison between three different machine learning algorithms (Linear regression, neural network, and Decision tree) which are trained using three different datasets based on different voltage indices (FVSI, Lmn and NLSI) for each algorithm , that produces nine predictors. The best results achieved by predictor based on A.N.N algorithm trained by NLSI dataset. This predictor used to build an early warning system and maneuver algorithm that can handle the voltage stability issue under the system operator supervision. VI. FUTURE WORK The future work will focus on how to build an early warning system based on learning machine algorithm that can choose a trigger to V.S.A.M.A for each line based on the historical data instead of the fixed triggers value used in this work which is preset by the system operator. Next we will develop a fully automated maneuver system that can dispense of the system operator. VII. REFERENCES [1] Arulampalam Atputharajah, Tapan Kumar Saha. "Power system blackouts - literature review." Industrial and Information Systems (ICIIS), IEEE, 2009. [2] Pansini, Anthony J. Power Systems Stability Handbook. The Fairmont Press, 1992. [3] P. Kundur ; J. Paserba ; V. Ajjarapu ; G. Andersson ; A. Bose ; C. Canizares ; N. Hatziargyriou ; D. Hill ; A. Stankovic ; C. Taylor ; T. Van Cutsem ; V. Vittal. "Definition and classification of power system stability IEEE/CIGRE joint task force on stability terms and definitions." (IEEE Transactions on Power Systems) 2004. [4] Pasllavi Choudekar, Divya Asija and Ruchira. "Prediction of Voltage Collapase in Power System Using Voltage Stability Indices." Proceeding of International Conference. Springer, 2017. Authorized licensed use limited to: Cape Peninsula University of Technology. Downloaded on September 05,2022 at 10:10:05 UTC from IEEE Xplore. Restrictions apply.