Bad Data Detection and Identification of Hybrid AC/DC Power Systems with Voltage Source Converters Using Deep Belief Network and K-Means Clustering

Abstract

The widespread use of Voltage Source Converter based High Voltage Direct Current Transmission (VSC-HVDC) technology has significantly increased the complexity of grid configuration and operation, which demands higher quality of AC/DC state estimation. The bad data detection and identification plays a vital role in ensuring the accuracy of state estimation outcomes. This paper presents a new approach to bad data in hybrid AC/DC grids based on the combined deep belief network (DBN) and K-means clustering method. First, the DBNs are trained separately for active and reactive power given the characteristics of VSC-HVDC by which the bad data can be detected. Then, an improved K-means is used for clustering the DBN outputs by setting the mean and the number of the samples within the clusters as two metrics for bad data identification. Finally, the case study is performed in a modified IEEE 14-bus system and the results demonstrate the effectiveness of the proposed method in terms of both the accuracy and efficiency.