Abstract: With the expansion of software scale expansion and logic complexity, software aging characteristics are more hidden, and the aging parameters timing signal is more complex. In view of the high requirements of the time series prediction method for sequence stationarity and the problems of slow convergence speed and easy trapping in local extremum of the BP neural network, a model of aging prediction software framework is proposed based on time domain convolution network (TCN). The available memory data was collected as the input of the framework and predicted by TCN model. The efficiency of the model was evaluated by checking the average error between the predicted output memory and the actual memory. Compared with ARIMA model and RNN（LSTM）model, TCN model has lower requirements on time series stability and better adaptability, and has no problems of gradient explosion or disappearance, and has the best prediction effect on acquired aging data.