Abstract: Accurate prediction of port throughput is critical for enhancing the precision of port construction decisions. However, identifying the factors influencing port throughput and constructing a robust predictive model remain significant challenges. To address this issue, this paper develops a modeling framework based on grey systems theory. First, grey relational analysis is employed to identify the relevant factors impacting changes in port throughput. These influencing factors are then utilized to construct a multivariate weighted fractional grey model for throughput prediction. To enhance the predictive capability of the grey model, this study establishes an error-based objective function to determine the model's fractional order, followed by corrections to the residual sequence using Long Short-Term Memory (LSTM) neural networks to improve prediction accuracy. To validate the effectiveness of the proposed method, empirical analysis is conducted using port throughput data from Guangxi Province, with comparisons made against four commonly used models. The experimental results demonstrate that the grey modeling framework proposed in this paper achieves high accuracy, with a 0.67% reduction in mean absolute percentage error (MAPE) outperforming alternative models. Furthermore, the model predicts a rapid growth trend in port throughput in Guangxi over the next three years, suggesting that expanding the utilization rate of existing port infrastructure and accelerating investments in capacity expansion are critical for meeting future demand. These results highlight the model's potential as a decision-support tool for strategic planning in the port and logistics sectors. 

Key words: Port throughput prediction, Fractional grey prediction model, Grey relational analysis, LSTM, Particle swarm algorithm