WAVELET ANALYSIS AND NEURAL NETWORK TECHNIQUE FOR PREDICTING TRANSIENT STABILITY STATUS

Abstract

This paper presents a method based on wavelet analysis (WA) and Multilayer perceptron neural network (MLPNN) to predict transient stability status (TSS) after a disturbance. It uses as input data, generator terminal frequency deviations extracted at a rate of thirty-two samples per cycle. Only the first eight frequency deviation samples per machine are needed. The eight samples are sub-divided into two sets, one set consisting of the first four samples and the other set consisting of the last four samples. Each set of samples is decomposed into 2 levels using the Daubechies 8 mother wavelet and the absolute peak value of detail coefficients obtained. The absolute peaks of detail coefficients of the first sample sets of all generators are added and so are the absolute peaks of detail coefficients of the second sample sets. The two summed values are then used as inputs to a trained MLPNN which predicts the TSS. The method was evaluated using dynamic simulations carried out on the New England test system. The method was found to be accurate and can be implemented in real-world systems to provide system operators advance information on system stability, following disturbances, to aid the deployment of needed emergency control measures.

Keywords

• neural network
• out-of-step
• power system stability
• stability prediction
• transient stability