Analyzing and Forecasting Container Throughput With a Hybrid Decomposition-Reconstruction-Ensemble Method: A Study of Two China Ports

Accurate container throughput forecasting is critical for enhancing port efficiency and ensuring global trade stability, particularly in the face of economic uncertainties, geopolitical tensions, and supply chain disruptions. Existing forecasting methods often struggle to model the nonlinear, nonstationary, and noise-laden characteristics of throughput data, creating a clear gap in the ability to provide reliable predictions. To address this, we propose a novel hybrid model, VMD-ISE-TCNT, designed to tackle these challenges. The model employs variational mode decomposition (VMD) to decompose time series into intrinsic modes, with an improved signal energy (ISE) criterion automating the selection of optimal mode numbers. These modes are categorized into low- and high-frequency components and forecasted separately using temporal convolutional networks (TCNs), leveraging their strength in capturing multiscale temporal dependencies. The Theil UII-S loss function is integrated to enhance model robustness by prioritizing proportional accuracy and reducing outlier sensitivity. Empirical evaluations using 24 years of data from China's two largest container ports—Shanghai and Shenzhen—demonstrate the superior performance of the VMD-ISE-TCNT model compared to traditional and hybrid benchmarks. By addressing frequency-specific patterns and automating key processes, this model provides a scalable and interpretable solution for advancing port operations and ensuring resilience in global trade.