A Novel EEMD-LSTM Combined Model for Transformer Top-Oil Temperature Prediction

Considering the potential overheating risks associated with the widely deployed transformers, how to accurately predict the top oil temperatures of transformers in real-time has become a critical but challenging task, raising extensive research attentions. This paper proposes a combined model integrating the ensemble empirical mode decomposition (EEMD) and the long short term memory (LSTM) for accurately predicting the top oil temperature of transformers, where the LSTM network model is trained based on the components decomposed by EEMD for avoiding the potential delay issues. Case studies using a practical distribution transformer station in Jiangxi Province are conducted to validate the effectiveness of the proposed EEMD-LSTM model. It is revealed that the proposed model shows significant superiorities over the BP neural network model and the single LSTM model in terms of prediction accuracy. The proposed prediction method based on data decomposition and component characteristics is conducive to monitoring the thermal state of transformers in advance, which could effectively improve the power system stability and facilitate the operation and maintenance schedule.