Detection and Identification of Bad Data Based on Neural Network and K-means Clustering

Abstract

Bad data detection and identification is very importance to improve the accuracy of state estimation. The paper propose a bad data detection and identification method based on BP neural network and K-means clustering . Firstly an active and a reactive BP neural network are trained to detect bad data considering the steady-state characteristics of power system. Secondly, for the issue of bad data misjudgment caused by BP output error detection criterion, an improved K-means algorithm based on distance cost function is applied to analyze the output data of active and reactive BP networks. After Kmeans clustering, two reference indicators of the cluster center distance and the number in a cluster are set to identify bad data. In this paper, active and reactive BP neural networks are classified to be trained, so that the network training scale is reduced and the speed of model training is improved; in addition, a K-means algorithm is used to narrow the scope of suspicious bad data and increase the reliability of the identification results. Finally the simulation results of the IEEE test systems and the actual grid operation data verify the effectiveness of the proposed method.