|Year : 2011 | Volume
| Issue : 1 | Page : 4-9
Power System Steady State Monitoring Using Artificial Neural Network
SG Ankaliki1, SG Gollagi2
1 Dean Academic, E and E Engineering, Hirasugar Institute of Technology, Nidasoshi-591236, Karnataka, India
2 Computer Science and Engineering, Hirasugar Institute of Technology, Nidasoshi-591236, Karnataka, India
|Date of Web Publication||4-Jan-2011|
S G Ankaliki
Dean Academic, E and E Engineering, Hirasugar Institute of Technology, Nidasoshi-591236, Karnataka
Source of Support: None, Conflict of Interest: None
| Abstract|| |
This article presents the application of Artificial Neural Network (ANN) for steady state monitoring of a power system. In steady state monitoring of a power system, it is important to predict the line flows and bus voltages for different operating conditions. In this article ANN has been proposed as an alternative method to solve the power system problems where the desired speed has not been achieved by conventional methods. The proposed method describes an adoptive pattern recognition approach based on highly parallel information processing. We provide a pattern of system description parameters to a neural network and net returns an estimate of line flows and bus voltages. Training data were obtained by Newton Rapson load flow simulation using Mipower Software Simulation Package for different system topologies over a range of load levels and the results were compiled to form the training data set. A back propagation algorithm was used for training ANN. Results of this approach help the power system operator to successfully handle the topologically independent steady state security assessment. To illustrate the proposed approach, IEEE-14 Bus system was considered. The difference between the actual and the estimated power flows and bus voltages was found to be good in terms of accuracy.
Keywords: Artificial neural network, and back propagation algorithm, NR load flow, steady state
|How to cite this article:|
Ankaliki S G, Gollagi S G. Power System Steady State Monitoring Using Artificial Neural Network. J Eng Technol 2011;1:4-9
| 1. Introduction|| |
In steady state security assessment of a power system, it is important to predict the line flows and bus voltages for different operating conditions and network topologies of a power system , . In the literature, several approaches, such as DC power flows  and the distribution factor  method have been proposed to estimate the line flows for real time applications, which are in general less accurate. With the development of artificial intelligence and artificial neural network (ANN) in recent years, there is a growing interest in proposing these tools for use in different areas of power system ,,, . In Ref.  , ANN was used for dynamic security assessment of power system. The multilayer perception model with back propagation (BP) algorithm was applied to predict the critical clearing time under different operating conditions and topology of the power system. Hsu et al.  developed a fast voltage estimation method using four-layered ANN. In the design of ANN, sets of variables that affect bus voltage most were selected as inputs to the ANN using entropy function. Ghosh et al.  designed a multilayer neural network for line flow contingency ranking. A regression-based correlation technique was used for feature selection to train the ANN using BP.
| 2. Artificial Neural Network|| |
A typical neural network consists of a large number of elementary processing units termed as "neurons," with each such unit connected to many others. If the numerical strength or weight of a connection between a pair of units is positive, the connection is excitatory, weight zero implies no connection and if the weight is negative, it is an inhibitory connection. Each neuron could receive several inputs (X1, X2, …, Xn) through the connections with associated strengths (W1 , W2 , …, Wn). These weighted inputs of a particular neuron are combined to produce a net input to the unit. Thus the input to the unit j is given by
This net signal is further processed by the activation function (F) of the neuron j to produce the output given by
An ANN generally consists of an input layer, an output layer, and one or more hidden layers with each layer having a number of neurons. The activation function of the neurons in the hidden and output layers of the network is taken as sigmoid. Training of an ANN implies the automatic adjustment of weights so that the application of a set of inputs produces the desired set of outputs. One of the popular learning algorithms is the error
BP algorithm  , which is based on the gradient descent technique for error reduction. In this algorithm, before starting the training process, after initializing all the weights, a set of patterns, each comprising a normalized input vector and the corresponding desired normalized output vector are shown to the ANN. The ANN output vector obtained is then compared with the desired output vector and the weights and the bias are adjusted until the error function  becomes acceptably small. The training time required by the BP algorithm is decreased by adding a bias to each neuron in the network using momentum coefficient and learning rate parameters  .
| 3. Proposed Approach|| |
The proposed approach is the application of ANN for steady state monitoring of a power system. ANN is more ideal for this type of problems because of its ability to augment new training data without the need for retraining. Here feed forward network is used for the accurate estimation of line flows and bus voltages. The supervised learning has been applied to ANNs. Although one ANN could be used to solve this problem, we used three ANNs: two for line flows and the other for bus voltage magnitudes. These produces better generalization results and reduces overall number of weights needed to represent the overall relationships. The inputs to the Neural network are the system variables, that is, the complex bus power of all the buses and self and mutual admittances of the lines that affect the line flows and bus voltages. The Complex bus power injection represents possible variations in generation and forecasted load. While self and mutual admittances of the lines represent the network topology.
| 4. Design of the ANN Model|| |
Design of the ANN model involves the following steps:
4.1 Preparation of Training Patterns
- Preparation of training patterns
- Selection of ANN structures
- Training of selected ANN structures
- Evaluation of the ANN model
A large number of offload patterns have been generated in a wide range of system operating conditions (60%-120%) and Newton Rapson (NR) load flow simulation using Mipower Software Package (Bangalore, India) has been performed to obtain the line flows and bus voltages in each case. The Input and Desired output patterns for Neural Network were obtained form Mi-Power simulator.
4.2 Selection of ANN structures
The selection of suitable ANN structure for proposed approach includes the selection of a number of layers, choice of transfer function, number of inputs, and number of neurons in each layer. A three-layered feed forward network can model complex mapping functions reasonably well and therefore it is suggested for this application. A sigmoid non-linear mapping function is employed in this application. The number of neurons in the input layer and hidden layer are decided by the input features and experimentation, which involves training and testing different network configurations. The neural network literature , provides guidelines for selecting the number of neurons for starting a network.
4.3 Training and Evaluation
Training of the selected network is done using training patterns and BP algorithm. The training is stopped when the mean squared error between the actual output and the desired output is acceptably small. Testing of trained ANN model is done using testing patterns generated by load flow simulation study [Figure 1].
| 5. Case Study|| |
To illustrate the proposed approach, an IEEE-14 Bus system was considered as shown in [Figure 2]. In this work, the goal was to examine the generalization capability of the ANN in the scope of being able to deal with a large range of operating conditions. This system has four generators at buses 2, 3, 6, and 8 and loads at buses 4, 5, 7, 9, 10, 11, 12, 13, and 14. To generate the training and test the data patterns, different loading conditions were selected with the loading level of the system in the range of 0.6-1.2 relative to the nominal operating point. The power factors of the loads were maintained at their nominal values. The training data set consisted of 52 × 15 dimensional patterns labeled with their corresponding line flow values (real or reactive power) or bus voltages. The training and testing data were obtained by the conventional NR load flow method using the commercial Mipower Software Package.
Although one ANN model could be used to solve this problem, in this work three ANNs were used; two for line flow estimation and the other for bus voltage magnitude estimation. These produces better generalization results and reduces overall number of weights needed to represent the overall relationships. The line flow neural network maps the complex bus power of all the buses and self and mutual admittances of the lines to line flows, while the bus voltage neural network maps the same inputs to the bus voltage. The bus admittances represent the network topology, whereas the busload and generation injection vectors represent possible variation in the load and generation distribution patterns based on forecast and generation scenarios.
5.1 Inputs to the Neural Network
Input patterns to the neural network are obtained from NR load flow solution. The following are the input variables:
An ANN having three layers with number of neurons in the input layer Ni = 15, chosen to be the same as that of the input variables, number of neurons in the hidden layer Nh = 6 and output layer No = 1 neuron was selected. For training ANN, BP algorithm was used. After training, the least squared error (E) was reduced to 0.000158 in 150,000 presentations of the training data set. The parameters of the learning process are momentum gain α = 0.25; threshold θ = 0.87768089; and adaptation gain η = 0.8960400.
- Self- and Mutual Admittances
- Complex Bus Loads
- Complex Bus Powers
| 6. Estimation of Power Flows and Bus Voltages|| |
[Table 1] shows the comparison of estimated and actual active power flow obtained by ANN-based algorithm and conventional NR load flow method in different lines and operating conditions.
The graphical comparison between the NR Load Flow and ANN-based algorithm results for Line Flows under different operating conditions is shown in [Figure 3].
|Figure 3: Line Flows in different lines by NR load Flow method and ANN-based algorithm at system load level of 0.6 pu|
Click here to view
[Table 2] shows the comparison of the estimated and actual reactive power flow obtained by ANN-based algorithm and conventional NR load flow method in different lines and operating conditions.
The graphical comparison between the NR Load Flow and the ANN-based algorithm results for Bus Voltage s under different operating conditions are shown in [Figure 4].
|Figure 4: Estimation of bus voltages by NR load flow method and ANN algorithm at system load level of 0.6 pu|
Click here to view
[Table 3] shows the comparison of estimated and actual bus voltages obtained by ANN-based algorithm and conventional NR load flow method at different buses at 0.6 pu load level.
| 7. Conclusions|| |
The designed ANN model can be applied to predict the line flows and bus voltages under changing operating conditions of the power system. Once the ANN is trained, it predicts quick results for unknown load patterns. The computation of line flows by load flow analysis takes long time, as it should be run for any change in load/generation. On the other hand, by the proposed method, once the training of the ANN is successfully completed, the prediction of the line flows and the bus voltages is almost instantaneous. This can be used for real time application. The outcome of this work can be used to examine the performance of a power system. In operation, this analysis assists engineers to operate the power system at a secure operating point where the equipment are loaded within their safe limits and power is delivered to customers with acceptable quality standards.
| References|| |
|1.||B. Stott, O. Alsac, and A. J. Monticelli, "Security Analysis and Optimization" IEEE Proceedings75, pp. 1623-1644, Dec. 1987. |
|2.||A. J. Wood, and B. F. Wollenberg, "Power Generation, Operation and Control" New York: Wiley; 1984. |
|3.||S. N. Singh, and S. C. Srinivas, "Improved Voltage and Reactive Power Distribution Factor for Outage Studies" IEEE Transaction on Power Systems, Vol. 12, No. 3, pp. 1085-1093, Aug 1997. |
|4.||T. S. Dillon, "Artificial Neural Network Applications to Power Systems and their Relationship to Symbolic Methods" Electric Power and Energy Systems, Vol. 13, No. 2, pp. 66-72, Apr 1991. |
|5.||P. K. Karla, S. C. Srivastava, and S. K. Joshi, "Synergism of Expert System and Artificial Net for Power System Applications" in proceedings of Fifteenth National System Conference, Roorkee (India), Mar 1992. |
|6.||V. S. Vankayala, and N. D. Rao, "Artificial Neural Networks and their Applications to Power Systems-a Bibliographical Survey" Electric Power System Research, Vol. 28, pp. 67-79, 1993. |
|7.||L. Srivastava, S.N. Singh, and J. Sharma, "ANN Applications in Power Systems; an Overview and Key Issues" Proceedings of the International Conference on Computer Applications in Electrical Engineering, Recent Advances (CERA 97), U.O.R., Roorkee (India), pp. 397-403, 1997. |
|8.||D. J. Sobajic, and Y. H. Pao, "Artificial Neural Network Based Dynamic Security Assessment for Electric Power Systems" IEEE Transaction on Power Systems, Vol. PWRS-4, pp. 220-228, 1989. |
|9.||Y. Y. Hsu, and C. C. Yang, "Fast Voltage Estimation Using an Artificial Neural Network" Electric Power System Research, Vol. 27, pp. 1-9, 1993. |
|10.||S. Ghosh, and B. H. Chowdhury, "Design of an Artificial Neural Network for Fast Line Flow Contingency Ranking" Electrical Power and Energy Systems, Vol. 18, No. 5, pp. 271-277, 1996. |
|11.||S. Haykin, "Neural Network-A Comprehensive Foundation" New York: Maxwell MacMillan International Publications; 1994. |
|12.||R. P. Lippman, "An Introduction to computing with Neural Nets" IEEE ASSP Magazine, pp. 4-22, Apr. 1987. |
|13.||D. Hammerstram, "Working with Neural Networks" IEEE Spectrum, pp. 46-53, Jul. 1993. |
| Authors|| |
Prof. S. G. Ankaliki obtained his BE (E and EE) and M.Tech (Power System) from Mysore University, INDIA. He is life member of ISTE, ISWE, and LMIE. He is presently working in HIRASUGAR INSTITUTE OF TECHNOLOGY, NIDASOSHI, INDIA, as a Professor and Dean (Academic). He is pursuing Ph.D. from Visvesvaraya Technological University, Belgaum, INDIA. His research interests include AI application to Power systems, Energy audit, Energy Management, and Demand side Management. He has published 13 articles in International Journals and 28 articles in Conferences.
Prof. S. G. Gollagi obtained his BE (CSE) and M.Tech (Computer Engineering) from Karnataka University, Dharawad, and University of Pune, INDIA, respectively. He is life member of ISTE. He is presently working in HIRASUGAR INSTITUTE OF TECHNOLOGY, NIDASOSHI, INDIA, as Assistant Professor. His research interests include Security in Wireless Network, AI application to power system, Algorithms, and Image processing. He has published 4 articles in International Journals and 5 articles in Conferences.
[Figure 1], [Figure 2], [Figure 3], [Figure 4]
[Table 1], [Table 2], [Table 3]