A fault location method for distribution system based on one-dimensional convolutional neural network

With the improvement of power users’ importance, it is increasingly important to find and eliminate the distribution network’s faults in time. Given the distribution network’s weak fault features, this paper presents a fault location method based on one-dimensional convolutional neural network (1D-CNN). The proposed method takes the raw measurement data as input and automatically learns the representation of the non-linear mapping relationship between the three-phase voltage and current signals of various fault locations and the estimated value of the fault location. In this study, extensive experiments are performed to demonstrate the effectiveness of the proposed model. The time-domain simulation is carried out in the PSCAD environment to generate large-scale fault samples, and then 1D-CNN is trained and tested based on these samples in Python and Keras library. The analysis of 5-fold cross-validation results proves that the proposed model showed impressive positioning results and has good adaptability. It provides an auxiliary means for power workers to diagnose distribution network faults.