A Novel Power System Anomaly Data Identification Method Based on Neural Network and Affine Propagation

Abstract

Identification of anomaly data is very important for power system state estimation. In this paper, a method of power system anomaly data identification based on neural network and affine propagation is proposed. In this first step, a 3-layer neural network is trained as a predictor using normal data. In the second step, data to be detected is preprocessed using the trained neural network, and predicted residuals are obtained. In the third step, these predicted residuals are clustered using the affine propagation clustering algorithm, and in the final step, anomaly data is identified based on the clustering results. As the neural network training process is easy to fall into local minimum, which reduces the prediction accuracy of the neural network, in this paper a novel chaotic particle swarm optimization algorithm is proposed to train the neural network. From the experimental results it can be seen that, compared with previous anomaly data identification method using the BP neural network and the gap statistic algorithm or the K-mean clustering algorithm, the proposed method can effectively improve the accuracy of anomaly data identification.