Aimed at the problems that the prediction of aero-engine exhaust gas temperature is low in precision, and the mining of the time dimension information of flight data by traditional RNN neural networks is inadequate, an exhaust gas temperature prediction model is proposed in combination with Autoencoder neural network and temporal convolutional network (TCN). First, the features related to the exhaust gas temperature (EGT) from the flight data are extracted by the Autoencoder method, and then the reduceddimensional features are taken as the input to build up a TCN network deep learning model. Secondly, the exhaust gas temperature of the aeroengine is used as the output to fully mine the flight data dimensional information, improving model accuracy. Finally, the actual flight data is selected to be an experiment. The results show that compared BP to LSTM, the average absolute percentage error is reduced from 13.035%, 9.593% to 3.369% respectively, effectively improving the prediction accuracy of the model.