Mechanical Vibration Information Prediction Model for GIS Equipment Based on Long and Short Term Memory Network and Time Domain Partitioning Method

Abstract

The Gas insulated metal enclosed switchgear (GIS) equipment mechanical defects seriously threaten its stable operation, and the prediction of the mechanical state can help to avoid the occurrence of accidents. So this paper proposes an open-loop prediction model GIS mechanical vibration information based on long and short term memory (LSTM) network and time domain partitioning. First, the time-series database was constructed based on the GIS mechanical defects simulation experiments for both instantaneous vibration waveform and long time interval partitioning. Then, the LSTM network algorithm was used to cascade to form a deep prediction network, and the open-loop prediction model is constructed by associating GIS disconnector poor contact defect to instantaneous vibration data and long time interval peak and harmonic distortion rate parameters. Finally, multi-step prediction was carried out for GIS vibration information with different currents and different defect severity, and the modal effectiveness was verified by comparison analysis. The results show that the model can achieve effective prediction of transient vibration signal and long interval state feature information, and the accuracy is much higher than that of the closed-loop prediction method, and the error is reduced by 36%, which is beneficial to the prevention GIS mechanical defect in advance.