A Prediction Method for the Average Winding Temperature of a Transformer Based on the Fully Connected Neural Network

The average winding temperature of a transformer (AWTT), serving as a key indicator for assessing the running state of the transformer, is of utmost importance in determining a transformer’s electrical properties and the insulation longevity of the transformer. An accurate prediction of AWTT is essential for ensuring the safe operation of the transformer. A novel method for predicting AWTT is introduced based on the analysis of field monitoring data. Firstly, the thermal characteristics and operational mechanisms of oil-immersed transformers are examined. Secondly, a factor analysis model is developed to streamline the network structure, accounting for the strong correlations among ambient temperature, load current, and top oil temperature. Thirdly, the independent temperature factor and load factor are extracted as pivotal features, and then input into the fully connected neural network to predict AWTT. Through a case study involving a 110 kV/10 kV oil-immersed transformer, the results show that the proposed method reduces redundant correlation information compared to traditional methods and improves the prediction accuracy of AWTT, establishing a foundation for further transformer state assessments.