Research on Transformer Temperature Rise Prediction and Fault Warning Based on Attention-GRU

In order to predict the temperature rise curve of power transformer groups and achieve temperature rise fault warning. Build a training and prediction set by measuring the load and top oil temperature of the transformer. The GRU time series model with improved attention mechanism is constructed to realize intelligent prediction of transformer oil temperature. The prediction performance of different algorithms was compared on the same test set. The research results show that the average absolute error, root mean square difference, and average absolute percentage error under the prediction set are the best performance of the algorithm, which indicates that the model has high prediction accuracy and reliability. Through the improvement of the attention mechanism, the algorithm's ability to dig key features is effectively improved, which is helpful to improve the early warning perception ability of transformer temperature rise fault.