In view of the shortage of the existing freight volume forecasting method that can only consider the single linear or nonlinear characteristic. Four combined forecasting models were proposed based on ARIMA-LSTM. the freight volume was forecasted considering the linear and nonlinear characteristic. The results show that prediction accuracy of freight volume resulting from any kind of ARIMA-LSTM combined forecasting model is better that from single model. And error reciprocal to determine the weight coefficient of parallel combination model is optimal. Compared with the ARIMA， the root mean square error，mean absolute error，the mean absolute percentage error reduces 40.66%， 29.76% and 51.45%，respectively. Compared with the LSTM，the value is the 13.67%，10.75% and 36.32%，respectively. This study can provide an effective basis for freight volume forecast.