It is of great significance to realize the accurate prediction of the temperature of power equipment to ensure the safety of the power system and to improve maintenance efficiency. Traditional forecasting methods cannot meet the requirements of high-precision forecasting. A temperature prediction method for power equipment based on an improved long short-term memory ( LSTM) neural network is proposed, which uses de-pooling convolutional neural networks (CNN) to extract local features of time series, and then use the recursive layer designed by LSTM to extract the long-term features of the time series to realize the temperature prediction of electrical equipment. Experimental results on the Power monitoring temperature data set of Capital International Airport show that the prediction accuracy of the temperature prediction value is better than 1 ℃ within 20 to 60 minutes, and the root mean square error (RMSE) 0. 12 is smaller than other temperature prediction models.