A novel hybrid deep-learning framework for medium-term container throughput forecasting: an application to China’s Guangzhou, Qingdao and Shanghai hub ports

Accurately forecasting container port throughput over the medium-term is crucial for making informed investment decisions by companies. However, current methods suffer from poor accuracy in dealing with complex, non-linear fluctuations and time series of uncertain demand. This research presents a novel hybrid framework that combines deep learning techniques with the distinctive features of seaborne transport to tackle this problem. The proposed approach employs variational mode decomposition (VMD) for decomposing the initial time series, thereby addressing prediction errors caused by disturbances and abrupt changes in the original data. In addition, Particle Swarm Optimization (PSO) is utilized to optimize the selection of parameters for VMD, aiming to achieve optimal outcomes in the decomposition process. Gated Recurrent Units (GRU) are then employed to accurately predict each intrinsic mode function (IMF) generated by the VMD. To obtain the final forecasts, the predictions of each IMF component are aggregated. Our results indicate that the proposed model performs better in demand forecasts, compared to traditional methods. According to our experimental results, the VMD has better numerical stability, noise removal, physical significance, and computational efficiency compared to other decomposing methods such as Ensemble Empirical Mode Decomposition (EEMD). The forecast results can approximate the development paths of the Guangzhou, Qingdao and Shanghai ports. This can help port operators and policymakers to prepare themselves for possible market fluctuations in the medium term, and make comprehensive adjustments and managemental decisions, such as capacity planning, on time.