Detection of Latent Fault in Medium Voltage Distribution Cables Based on Guiding Learning Model

Abstract

The latent fault in cables cannot be identified by conventional relay protection due to its short duration and small fault current. In order to prevent it from developing into a permanent fault, it is necessary to detect and identify the latent fault of the cables. The traditional latent fault detection of cables based on the machine learning method needs to mine a large amount of data, but there are few abnormal power data samples. At the same time, this process is prone to overlearning problems. This paper introduces a guiding learning model based on the combination of domain empirical knowledge and machine learning algorithms; the overcurrent detection criterion is constructed by performing Wavelet transform on the fault current to extract the feature quantity, which is used as the empirical knowledge to combine with the Extreme Learning Machine (ELM) model to construct the knowledge function; then PSO algorithm is used to optimize the model, the improvement of the accuracy of the guided learning model proposed in this paper is verified by comparing with the ELM detection model directly optimized by the PSO algorithm.