NEURAL NETWORK AND SPEED DEVIATION BASED GENERATOR OUT-OF-STEP PREDICTION

Abstract

A multi-layer perceptron neutral network (MLPNN) has been used as a decision tool to predict out-of-step conditions. Rotor speed deviations were sampled and the maximum speed deviation in 1 cycle is obtained, and used as input to the MLPNN. Each generator has one trained MLPNN assigned to it to predict whether or not that generator will go out of step following a disturbance. The trained neural networks responded to the 88 individual generator out-of-step (OOS) cases with 100% accuracy while the responses to 512 synchronism cases were 98.05% accurate. The 340 predictions for 34 simulations with all 10 generators in synchronism were 100% accurate. The study used the IEEE 39-bus as the test system.