Kalman Filtering and LSTM-Based Fault Diagnosis and Localization Method for Low-Voltage Distribution Substations

The efficiency of fault disposal in low-voltage distribution substations has a significant impact on the analysis and detection of fault sections, phases and types. In view of this issue, a method for fault location and diagnosis in low-voltage distribution substations based on Kalman filter (KF) and long short-term memory (LSTM) theory is proposed in this paper. The KF is used to identify the fault mutation points and locate the fault position. The three-phase current of the fault nodes are used as inputs for LSTM neural network. The LSTM neural network model is trained and validated using simulated data by taking advantage of the LSTM’s memory capabilities in handling time series. The three-phase current of the mutation point is input into the established LSTM network to output the fault type. Finally, a series of case studies were conducted using KF-LSTM. The error rate between the mutation point identified by the KF method and the actual value was 0.01%, while the accuracy of the LSTM method in various fault experiments exceeded 99.5%. In conclusion, the proposed method effectively avoids interference during nonfault moments, and achieves fault accurately location and diagnosis in low-voltage distribution substations.