Data-Driven Inversion and Prediction Method for Hot-Spot Temperature of Oil-Immersed Transformer

The control of hot-spot temperature of a transformer is the key to ensuring its safe operation. Therefore, operators are especially focused on the hot-spot temperature and its variations in relation to different operating conditions. The hot-spot temperature of large oil-immersed transformers is difficult to be measured directly. This study proposes a data-driven approach for inverting and predicting the hot-spot temperature of oil-immersed transformers. Combining the structural parameters of oil-immersed transformers with monitoring data, the inversion model for the hot-spot temperature of the operating transformer can be developed to achieve real-time calculation of the hot-spot temperature. Based on multidimensional time series data, the prediction of transformer hot-spot temperature is accomplished by leveraging the strengths of recurrent neural networks and convolutional neural networks. An operating oil-immersed transformer is considered as an example. The proposed hot-spot temperature inversion model demonstrates an absolute error of less than 0.3 °C in hot-spot temperature inversion when evaluated on the test dataset and a calculation time of less than 0.02 s. The prediction model exhibits an absolute error of less than 0.8 °C in hot-spot temperature predictions on the test dataset. This indicates the effectiveness of the method for real-time calculation and prediction of hot-spot temperatures in large oil-immersed transformers.