Neural Networks Applied in the Prediction of Top Oil Temperature of Transformer

Abstract

Top oil temperature (TOT) is an important indicator reflecting the load capacity and insulation aging of the transformer. In order to predict the TOT accurately, this paper propose a transformer top oil temperature prediction method based on BP neural networks optimized by Adam. Firstly, we use the grey relational analysis method to calculate the correlation between other state variables of the transformer and the TOT, select state variables with larger correlation as the inputs and the TOT as the output to establish the neural networks prediction model (NNPM) of the TOT. Next NNPM of TOT is trained using historical data of transformer and Adam optimization algorithm. Then the case studying for historical data suggests that the prediction results of NNPM optimized by Adam of TOT are in accordance with measured results. Comparing with D Susa thermal circuit model and NNPM trained by SGD, the prediction accuracy of NNPM optimized by Adam is improved by 78.1% and 33.95% respectively. Finally, we choose different transformers to model and predict, and the results show that NNPM of TOT based on Adam has applicable ability to different transformers. The top oil temperature prediction method proposed in this paper provides a more accurate calculation basis for prediction of TOT of transformers and is of great significance for the safe and stable operation of the power transformers.