Fault‐cause identification method based on adaptive deep belief network and time–frequency characteristics of travelling wave

Abstract

Accurate fault‐cause identification is highly important to the fault analysis of overhead transmission lines (OTLs). In order to improve the efficiency and accuracy of fault identification, this study proposes a fault identification method based on the ADBN (adaptive deep belief network) model and the time–frequency characteristics of a travelling wave. According to the mechanisms of different OTL faults, the appropriate time–frequency characteristic parameters of the fault current travelling wave were selected as the input of the ADBN model, and the fault‐type labels were selected as the output. The ADBN model introduces the idea of adaptive learning rate into CD (contrastive divergence) algorithm and improves its performance with self‐adjusting learning rate. The parameters of the ADBN model were pre‐trained with the improved CD algorithm and adjusted by back propagation algorithm with the labels of the samples. The performance of the ADBN model was verified by field data, and the accuracy of fault identification was analysed under different model parameters, characteristic parameters, and sample sizes. The results showed that the model helps to characterise the inherent relationship between characteristic parameters and fault causes, and the proposed method can effectively identify different fault causes in OTLs.