Hybrid GRA-GRM Model for Spatiotemporal Forecasting in Port Logistics Using Grey System Theory and Nonlinear Feature Selection

Abstract

Accurate and timely port logistics demand forecast is essential for increasing operational efficiency, cutting costs, and raising customer satisfaction in logistics and transportation networks in the age of artificial intelligence and quickly changing supply chain dynamics. This study introduces a new hybrid model, called the GRA-GRM combination model, that combines Grey Relational Analysis (GRA) and Grey Relational Modelling (GRM) to successfully handle the difficulties of feature selection and nonlinear pattern extraction that are present in port logistics demand forecasting. The algorithm outperforms traditional techniques in terms of predicted accuracy by using a comprehensive real- world dataset from five major Chinese cities, which includes over 10 million package transactions and comprehensive courier trajectories. The model's GRA component reduces dimensionality and improves interpretability by thoroughly identifying the key variables influencing port logistics demand. The GRM then models intricate nonlinear interactions using these chosen characteristics, improving prediction accuracy. Performance evaluation shows that the suggested GRA-GRM model outperforms baseline models like standalone GRM (MAE 8.43, RMSE 10.25, MAPE 12.3%, R² 0.841) and Long Short-Term Memory networks (LSTM) (MAE 7.92, RMSE 9.38, MAPE 10.7%, R² 0.865) in terms of predictive metrics, with Mean Absolute Error (MAE) of 5.91 minutes, Root Mean Square Error (RMSE) of 7.42 minutes, Mean Absolute Percentage Error (MAPE) of 7.6%, and coefficient of determination (R²) of 0.918. Additional studies include error distribution histograms that demonstrate the resilience and dependability of the model, time-series comparisons of the projected and real logistical demand, and geographical heatmaps that show the accuracy of regional demand forecast across many Areas of Interest (AOIs). Logistics managers and policymakers may take action based on feature significance rankings, which show that port container throughput and courier delivery frequency are important factors influencing demand swings. Even while the model performs well, issues like its susceptibility to input noise and real-time flexibility point to areas for further study, such as integrating with IoT data streams and sophisticated deep learning architectures. All things considered, this research offers an AI-driven forecasting framework that is scalable, interpretable, and efficient, facilitating more intelligent resource allocation and operational decision-making in intricate port logistics settings.