15th Indonesian Scientific Conference in Japan Proceedings. ISSN:1881-4034 Generator Angle Difference Monitoring System to Ensure Power Transmission Steady State Stability Based on Neural Network Adi Soeprijanto1, Ardyono Priyadi1,2, Riyan Danisaputra1, Naoto Yorino2, Yoshifumi Zoka2 1 Electrical Engineering Department, Sepuluh Nopember Institute of Technology, Keputih-Sukolilo, Surabaya-60111 2 Graduate School of Engineering, Hiroshima University, 1-4-1 Kagamiyama, Higashihiroshima, 7398527 Japan Abstract. Power angle difference between groups of generators is an important measure of steady state stability in power system. Using neural network, a model to predict power angle difference between groups of generators at Java-Bali 500 kV is established. Neural Network is built with system loads as inputs and the average angles of group of generators as targets. This target data is provided through calculation based on Center of Angles (COA) criterion. Instead of measurement, Newton Raphson power flow solution result is used as initial values of COA criterion method. The 32 varied load data were used for learning process of the model. Backpropagation with momentum method is applied with input layers consist of 20 neurons representing real power of 20 buses, while hidden layers consist of 10 neurons. Output layer consists of only one neuron is represented as difference of average power angle between west group and east group. Keywords. Power angle stability, COA criterion, and neural network. 1 Introduction Steady state stability is the ability of synchronous machine of a power system to remain in synchronism after a very minute disturbance. Small disturbances are always present in a power system; for example there are gradual changes in load, manual or automatic changes of excitation, irregularities in prime-mover input, and so forth. Obviously these small disturbances cannot cause loss of synchronism unless the system is operating at, or very near, its steady state stability limit. Continuous electricity supply is conducted if only certain synchronous generators are ready to supply. Once the generator is out of synchronism, significant voltage and current fluctuation is occurred. These could initiate trouble in transmission system. Consequently, a monitoring system to ensure power transmission steady state stability is required. Power transmission steady state stability shows maximum transmitted power. Power transmission capacity of a generator group depends on power angle difference with another generator group. Neural Network is built with system loads as inputs and the average angles of group of generators as targets. This target data is provided through calculation based on COA criterion. Instead of measurement, Newton Raphson power flow solution result is used as initial values of COA criterion method. 2 Theory 2.1 Newton Raphson Power Flow Solution Power flow solution is necessary for planning, economic scheduling, and control of an existing system as well as planning its future expansion. The problem consists of determining 88 15th Indonesian Scientific Conference in Japan Proceedings the magnitudes and phase angle of voltages at each bus and active and reactive power flow in each line. N Pi = ∑ ViVnYin cos (δ i − δ n − θ in ) (1) n =1 N Qi = ∑ ViVn Yin sin (δ i − δ n − θ in ) (2) n =1 The procedure for power flow solution by the Newton Raphson method is as follows (Saadat 1999, Grainger and Stevenson 1994): 1. For load bus, where Pi sch and Qi sch are specified, voltage magnitude and phase angles are set equal to the slack bus values, or 1.0 and 0.0. For voltage-regulated buses, where voltage magnitude and Pi,calc are specified, phase angle are set equal to the slack bus angle, or 0. 2. Pi(k) and Qi,(k) are calculated from equation (1) and (2) 3. Calculate ∆P in each buses 4. Calculate Jacobian matrix, through estimation value or partial differentiation equation from equation (1) and (2) 5. Inverse the Jacobian matrix to calculate ∆δi and ∆│Vi │ in each buses 6. Calculate the new δi and ∆│Vi │by adding ∆δi and ∆│Vi │ to the latest value 7. The process is continued until residuals ∆P and ∆Q are less than the specified accuracy. 2.2 Two Machine System The type of power system whose steady-state stability limit can be found most simply consists of two synchronous machines, a generator and a motor, connected through a network of pure reactance. Power angle curve is established based on equation below: P= E1 .E 2 . sin δ x (3) where: P : Power transmitted from machine 1 to machine 2, E1: Internal voltage of machine 1, E2: Internal voltage of machine 1, x :Transfer reactance between the internal voltages, δ:Angle by which E1 leads E2 . Figure 1. Power Angle Curve Generator is stable if only transmitted power is less than Pm (δ<900). 89 Indonesian Student Association in Japan 15th Indonesian Scientific Conference in Japan Proceedings 2.3 Generator Power Angle Calculation (δ) Before initiating power angle difference between groups of generator, power angle of each generator must be defined. Power angle of each generator (δ) is angle by which E1 leads E2 (Taylor 1994, Gross 1986, Kundur 1994). Equivalent circuit of generator connected through infinite bus is shown below: E1 I E2 Xs Figure 2. Equivalent Circuit of Generator-Infinite Bus According to the circuit showed in figure 2., the value of E1 is initiated by: E1 = E 2 + jX s .I I= (4) S* * E2 (5) where: E1: Internal voltage of generator, E2: bus voltage, Xs: synchronous generator reactance, I: current flows (from generator) through bus, S: Apparent Power flows in bus, * : conjugate According to equation (4), phase diagram of voltage and current is established and the δ can be calculated (Kimbark 1947). Figure 3. Three Phase Diagram of Voltage and Current 2.4 Average Power Angle of Generator Groups Based on COA criterion, average power angle of generator groups is initiated (Kimbark 1999). The equations are as follow: Ma = ∑M l∈A (6) l δ a = M a−1 ∑ M l δ l (7) l∈ A where: l : generator number (1,2,3…n), Ml: inertia moment of each generator, Ma :Totalinertia moment of group ‘a’, δl : power angle of each generator, δa: average power angle of group ‘a’ Indonesian Student Association in Japan 90 15th Indonesian Scientific Conference in Japan Proceedings 2.5 Basic Concept of Neural Network To build a model of Neural Network, several input-output data are needed as data couple. Figure 4 shows a simple neural network (Momoh and El-Hawary 1999). Input Layer Hidden Layer Output layer X1 Z1 Y1 X2 Z2 Y2 Xi vij Zj weight wij Yk weight Figure 4. A Simple Neural Network Learning process of Neural Network is a process to search the best weight by training the network appropriated with the wanted performance, so that given input vectors produce aimed output vectors. The training is built by applying input vectors scheme orderly, and also to control network’s weight to search output vectors in line with a learning algorithm. While learning process, weight is convergent smoothly to certain value. In addition, output scheme produce aimed output vectors (Warwick et al. 1997, Kreshna and Srivastava 2006). The 32 varied load data were used for learning process of the model. Backpropagation with momentum method is applied with input layers consist of 20 neurons representing real power of 20 buses, while hidden layers consist of 10 neurons. Output layer consist only 1 neuron representing average power angle difference between west group and east group. 3 Simulation 3.1 Simulation Simulation step of the steady state stability monitoring system in Java-Bali 500 kV Transmission System is shown below: Figure 5. Simulation Design 91 Indonesian Student Association in Japan 15th Indonesian Scientific Conference in Japan Proceedings The step of monitoring system design is started by collecting secondary data of line. The bus data as inputs for Newton Raphson power flow. The solution of the power flow is used – instead of direct measurement - as inputs to COA method (Kimbark 1999) in calculating power angle difference between generator groups of East Java and West Java. The additional data needed in this step is inertia of the generators. Neural Network based monitoring system then is trained using the variation of load data as input and calculated steady state stability as target. This model assists steady state stability monitoring whether is stable, critical stable or unstable. As those data are included, the simulation is ready to start. Power Flow simulation is firstly done to find out power flow in each buses of 500 kV Java-Bali System. 14 11 9 1 10 13 12 8 2 3 4 15 16 17 18 19 20 7 5 6 Figure 6. A 500 kV Java-Bali Transmission System Indonesian Student Association in Japan 92 15th Indonesian Scientific Conference in Japan Proceedings Table 1. Load in 500 kV Java-Bali System Table 2. The 500 kV Java-Bali System Data Table 3. Inertia Moment of West Java Generators 93 Indonesian Student Association in Japan 15th Indonesian Scientific Conference in Japan Proceedings Table 4. Inertia Moment of East Java Generators Newton Raphson Power Flow solution results power angle of each generator. Then, those are processed to find power angle difference between West Java and East Java groups using the COA method. 3.2 Power Angle Difference (∆δ) According to table 1 until table 4, equation (6) and (7), average power angle of each generator group (Peak Load, 2004) is found. In addition, Power Angle Difference (∆δ) between East Group and West Group of generator is 22.430 – 3.6160 = 18.810. 3.3 Applying Neural Network Equal calculation is applied to determine the 32 of power angles difference by varying the load. The 32 varied load data which results 32 power angle difference data are prepared as input and target data for learning process in Neural Network. Learning process is finished when maximum error (10-4) or maximum epoch (9000 epochs) is reached. Learning process after 9000 epochs is shown in figure 7. Performance is 0.000127315, Goal is 0.0001 1 10 0 Training-Blue Goal-Black 10 -1 10 -2 10 -3 10 -4 10 0 1000 2000 3000 4000 5000 9000 Epochs 6000 7000 8000 9000 Figure 7. Learning Process (after 9000 epochs) After learning process, the built neural network model is tested by 5 similar data. Then, the output is benchmarked with the target. Table 5. shows error between neural network output and the target. Table 5. Error Percentage of Neural Network Indonesian Student Association in Japan 94 15th Indonesian Scientific Conference in Japan Proceedings 4 Conclusion 1. The 500 kV Java–Bali Transmission System is still in power transmission steady state stability when peak load was reached in 2004 (∆δ = 18.81 degrees) 2. If Western and Eastern Java Loads are significantly different, the 500 kV transmission system is closing critical steady state stability 3. Neural Network could be applied to monitor power angle difference between groups of generator References [1] Chapman, S.J. (1999). “Electric Machinery Fundamentals Third Edition”, McGraw-Hill Inc. [2] Grainger, J.J. and Stevenson, W.D. (1994). “Power System Analysis”, McGraw Hill Inc. [3] Gross, C.A. (1986). “Power System Analysis Second Edition”, John Wiley & Sons. [4] Kimbark, E.W. (1947). “Power System Stability Volume I”, John Wiley & Sons. [5] Kimbark, E.W. (1948). “Power System Stability Volume III”, John Wiley & Sons, [6] Kreshna, J. and Srivastava, L. (2006). ”Counterpropagation Neural Network for Solving Power Flow Problem”, International of Intelligent Technology Vol. 1, Number 1. [7] Kundur, P. (1994). “Power System Stability and Control”, McGraw Hill Inc. [8] Momoh, J.A. and El-Hawary, M.E. (1999). ”Electric Systems, Dynamics, and Stability With Artificial Intelligence Applications”, Technology . [9] Pai, M.A. (1981). “Power System Stability Volume 3”, North-Holland Systems and Control Series. [10] Saadat, H. (1999). ”Power System Analysis”, McGraw Hill Inc. [11] Semitekos, D. and Avouris, N. (2002). "Power Systems Contingency Analysis using Artificial Neural Networks", semitekos-power. [12] Taylor, C.W. (1994). “Power System Voltage Stability”, McGraw-Hill Inc. [13] Warwick, K., Ekwue, A., and Aggarwal, B.L. (1997 ). ”Artificial Intelligence Techniques in Power Systems”, Technology. . 95 Indonesian Student Association in Japan