Prediction of Top Oil Temperature for Oil-immersed Transformers Based on PSO-LSTM

Abstract

The evaluation of the thermal state of transformers is of great significance for monitoring the operating state of transformers and power systems. The prediction of transformer top oil temperature based on historical data has become an effective way to evaluate the thermal state. To improve the prediction accuracy of top oil temperature of transformers, a new top oil temperature prediction model based on Long Short-Term Memory (LSTM) is proposed in this paper. The strong correlation factors affecting the top oil temperature are determined by the heat map and are selected as the model input. In addition, the concept of sliding time window is introduced, and the optimal time step is determined by particle swarm optimization (PSO) to improve the prediction accuracy of top oil temperature. The test results show that the multi-factor LSTM prediction model with time step optimized by PSO has a higher prediction accuracy.