Transformer bushing is one of the transformer current-carrying components. Transformer bushing failure will cause the oil tube lead role to be damaged, which is not conducive to the normal operation of the transformer. In order to accurately predict the bushing temperature and detect potential transformer faults in a timely manner, a temperature prediction model based on historical temperature monitoring data was proposed using Particle Swarm optimization (PSO) to optimize Long and Short Term Memory Network (LSTM). The real data was collected continuously for 13 days under the same sensor on a high voltage bushing. After cleaning, the data were used as the input data set for training and test purpose of PSO-LSTM. PSOLSTM method was used to automatically optimize the LSTM parameters, which avoids the problem of low prediction accuracy due to empirical selection of parameters, thus improving the convergence speed and prediction accuracy. The simulation results show that the PSO-LSTM algorithm can complete the bushing temperature prediction within 7s. In terms of temperature prediction accuracy, root mean square error and average absolute error, it outperforms Back Propagation (BP) neural networks, Wavelet Neural Networks (WNN) and single LSTM algorithms, which shows its good application prospect in online monitoring and fault diagnosis for electric transformers.