Time-series Prediction of Busbar Temperature Rise based on LSTM Neural Network

Switchgear is an essential component of electrical power systems and plays a vital role in ensuring their reliable and safe operation. Timely identification of abnormal conditions and accurate prediction of potential faults significantly enhance the system’s reliability and integrity. Thus, we developed a Long Short-Term Memory (LTSM) -based time series prediction model for busbar temperature rise with a real-time switchgear condition monitoring system. The monitoring system monitored and collected switchgear parameters such as load, current, ambient temperature, humidity, and busbar temperature rise. The system was deployed to an operational switchgear to collect data. Correlation analysis was performed, and variables with a strong influence on temperature rise were used to build a multivariate LSTM. Additionally, an ARIMA-based model was also built using the same collected data. Performance evaluation results showed that prediction error metrics (mean absolute error, mean absolute percentage error, and root mean square error) for LSTM were 0.029, 0.233, and 0.052, respectively. These values were significantly lower compared to those for ARIMA, which were 0.340, 2.853, and 0.362. The LSTM model reduced the prediction error by 91.5% compared to the conventional model as it predicted better busbar temperature rise.