Research on Transformer Winding Hot Spot Temperature Prediction Based on Improved PSO-LSTM Model

To address the issue of unsatisfactory diagnostic accuracy in the process of predicting the hot spot temperature of transformer windings, an improved particle swarm optimization-long short-term memory（PSO-LSTM） algorithm is proposed to optimize neural network training parameters. Firstly, the initial search range of particles in the PSO-LSTM algorithm is modified to improve the convergence accuracy and stability of the algorithm. Afterwards, the transformer winding hot spot temperature data is used as input, and the temperature prediction neural network is trained using an improved PSO-LSTM algorithm. Finally, the output results of the improved PSO-LSTM neural network are compared and analyzed with those of traditional LSTM neural networks and PSO-LSTM neural networks. The results showed that the improved PSO-LSTM algorithm improved the accuracy of transformer winding hot spot temperature prediction.