Fault prediction and health status evaluation method of GIS equipment based on time series neural network

This paper proposes a gas insulated switchgear (GIS) equipment fault prediction and health status evaluation method based on time series neural network (TSNN), which solves the shortcomings of existing methods in data fusion, model generalization ability and real-time performance. Firstly, a multi-source data fusion scheme based on feature level fusion is designed, which integrates multi-dimensional monitoring data such as partial discharge, temperature and vibration into a unified feature vector, and dynamically calculates the fusion weight of each sensor data through attention mechanism to improve data utilization efficiency and enhance the adaptability of the model to complex working conditions. Secondly, a time series prediction model based on LSTM and two-stage attention mechanism is constructed. The importance of different sensor data is evaluated by spatial attention mechanism, and the long-term dependence is captured by temporal attention mechanism, so as to accurately predict the faults of GIS equipment. Multi-dimensional health evaluation indexes are defined, including failure probability, load rate, mechanical operation times and insulation aging coefficient. The health index is calculated by using weighted fusion formula, and the health status of equipment is divided into four grades: normal, attention, warning and dangerous, which provides intuitive reference for operation and maintenance personnel. Experimental results show that the proposed method is superior to traditional methods such as ARIMA, SVM+Kalman and CNN-LSTM in fault prediction performance, with an average F1-Score of 82.0%, and performs well in health status classification, with classification accuracy of normal, caution, warning and dangerous status of 92%, 85%, 83% and 85% respectively. This method can effectively capture the deterioration trend of equipment, meet the requirements of IEC 62478 standard for GIS equipment maintenance cycle, and has high engineering practicability.