Purpose Port throughput forecasting is crucial for port planning and regional economic development, while traditional models suffer from poor adaptability to small-sample data, failure to capture periodic and time-varying features and low prediction accuracy. This study aims to develop a more accurate and robust model to realize precise forecasting of China's coastal port throughputs with different time scales and dimensions. Design/methodology/approach This study proposes a fractional-order time-varying grey Fourier model (FTGFM) by extending the traditional grey model framework. The proposed model integrates a fractional-order accumulation operator, a time-varying correction term and a truncated Fourier series to respectively characterize new-information priority, system dynamic evolution and periodic data fluctuations. The Whale Optimization Algorithm is employed to optimize nonlinear parameters. The effectiveness of the model is evaluated using quarterly data (2017–2025) and monthly data (2023–2025) for four core port throughput indicators, and its performance is compared with six comparative models based on mean absolute percentage error (MAPE), root mean squared error and Theil's U statistic. Findings Empirical results indicate that FTGFM(1,1) outperforms all benchmark models, with training and testing MAPE both below 3%. It effectively captures the periodic and time-varying trends in port throughput, and forecasts suggest China's port throughput will fluctuate steadily upward in 2026–2027. Originality/value The study structurally improves the grey model with a time-varying correction term and embeds Fourier series into grey modeling for port throughput forecasting for the first time, combined with the fractional-order accumulation operator to optimize information processing. The proposed model thus provides an effective new tool for forecasting small-sample time series with periodic and time-varying features.
Liang et al. (Thu,) studied this question.