Electric power system contingency ranking using artificial intelligence techniques

Abstract

This research attempts to implement an optimized Artificial Neural Network based newlinecontingency selection and ranking algorithm. Optimizing the ANN involves the weight newlineand bias updating iteratively to quicken the convergence while training ANN. Thus, newlinespeeding up the process of ANN training for every configuration used. The objectives of newlineresearch are as follows: newlineand#61623; Online Static Security Assessment using Multilayer Feedforward Artificial newlineNeural Network newlineThe proposed online static security assessment module utilizes multilayer feedforward newlineartificial neural network (MLFFNN). Real and reactive power, Voltage magnitude and newlinephase angle at various buses are taken as the inputs to the ANN. The outputs are set as newlinesecure or insecure, critical contingency screening and contingency ranking. The number newlineof inputs mainly depends upon the topology of the system under consideration. The newlineactivation function in the hidden layers is the hyperbolic tangent and at the output layer, newlinethe linear function is used. The network is trained using back propagation algorithm. newlineTraining and testing on neural network is done using bus quantities due line outages. newlineNewton Raphson method is used for load flow analysis. In the proposed approach, power newlinesystem security assessment against unplanned line outages are done by utilizing the high newlineadaption capability of ANNs, as these are better suited to deal with nonlinear problems. newlineand#61623; Online Static Security Assessment using Radial Basis Feedforward Network newlineA radial basis function network is an artificial neural network that uses radial basis newlinefunctions as activation functions. The output of the network is a linear combination of newlineradial basis functions of the inputs and neuron parameters. Radial basis function networks newlinehave many uses, including function approximation, time series prediction, classification, newlineand system control. The network is capable of performing nonlinear mapping. The hidden newlinelayer with Gaussian activation functions and linear activation function in the output layer. newlineRBF is trained and tested with back propagation algorithm. The RBFN gives faster newlineconvergence than MLFFNN. newlineand#61623; Online Static Security Assessment Module with PSO Trained RBFNN newlineThe online static security assessment module is modeled with RBF neural network and newline newlinetrained with PSO algorithm. RBFNN trained with PSO gives better results compared to newlineback propagation. The real and reactive powers, voltage magnitudes and phase angle for all newlinebuses are used for describing the system operating point and are chosen as the input. The newlineoutputs are set as secure or insecure, critical contingency screening and contingency newlineranking. PSO algorithm is a simple and faster algorithm for getting the optimal results in newlineoptimization techniques. This algorithm is inspired from food searching habit of birds. newlineThe time taken to train ANN is less when compared to back propagation training. newlineThe contingency analysis is performed on standard IEEE-118 and 30 bus system and newlinepractical Karnataka Power Transmission Corporate Limited (KPTCL) to test the proposed newlinemethod. The results obtained from the ANN is compared with conventional method. The newlinetime taken to train and test the neural network is less with ANN method. Hence, the newlinecomputation time taken to find the performance index while finding the contingency newlineranking is far better when compared to conventional methods. newline