Scientific and accurate prediction of passenger transport volume can provide effective reference for transportation related departments. Taking passenger transport volume as the research object, based on SSA(singular spectrum analysis ), combined with LSTM(long-short term memory neural network ) and ARMA (auto-regressive moving average model), the time series of passenger transport volume was decomposed into signal sequence and noise sequence through SSA noise reduction processing, and LSTM and ARMA(2,3) modeling were carried out on them respectively. Based on this, its changing trend is predicted. By comparing the experimental results of single ARIMA(3,1,2) model and LSTM model, it shows that SSA-LSTM-ARMA has better prediction effect and higher prediction accuracy in passenger traffic volume.