Abstract: To cope with the significantly increased demand for railway freight operations and improve the efficiency of railway freight yard operations, this paper designed a prediction model of dwell time for final arrival vehicles in railway freight yard based on the collection of relevant influencing factors. This model predicted the empty and heavy status of the vehicle's departure based on statistical data, and then predicted the duration of stay. The paper constructed three sub models by simultaneously training a random forest and a BP (Back Propagation) neural network, and selecting the optimal results. Through data verification, the mean square error and mean absolute error of the prediction model are superior to models that only use random forest algorithm or BP neural network algorithm. It can effectively predict the dwell time for final arrival vehicle and provide technical support for the planning and efficiency analysis of freight station operations.