An Improved VMD-GRU Model for Forecasting the Container Freight Volume of Inland Ports

Accurate forecasting of container freight volume demand at ports is crucial to the social production and investment decisions of businesses. When dealing with complex, nonlinear fluctuations and ambiguous demand time series, the current methods for forecasting freight volume have demonstrated poor accuracy. In view of this, this study proposes a container freight volume forecasting model using Particle Swarm optimization (PSO)-based Variational Mode Decomposition (VMD) and Gated Recurrent Units (GRU). Within the proposed design. In particular, VMD is used to decompose the original time series, which reduces the prediction error caused by disturbances and abrupt changes in the original series. In the meantime, PSO is utilized to optimize the parameter selection of VMD in order to attain the optimal decomposition effect. After decomposition, GRU is utilized to accurately predict each intrinsic mode function (IMF). The final prediction result is obtained by superimposing the predictions of each IMF component. Compared to a BP neural network, a single GRU model, and a support vector machine (SVM), the root mean square error (RMSE), the mean absolute error (MAE), and the mean absolute percent error (MAPE) of the proposed model are smaller, and its prediction accuracy is greater than that of the comparison model.