Research on artificial intelligence-driven container relocation problem for green ports

Abstract

Introduction: Container relocation in port yards represents a canonical NP-hard problem, characterized by high-dimensional nonlinear constraints and stringent real-time decision-making requirements. Methods: This study proposes a unified framework integrating an Intelligent Decision-Driven Model (IDDM), an Adaptive Data Generator (ADG), and an Optimization–Learning Closed-Loop Framework (OLCF). Results: The IDDM leverages heuristic search and machine learning within a multi-stage decision mechanism to mitigate the curse of dimensionality; in two-dimensional scenarios involving 50–100 containers, the model achieves an average response time of 9.83 ± 0.12 µs and reduces relocation operations by 61.68%. In three-dimensional experiments at the scale of 104 containers, total computation time remains consistently below 60s, satisfying real-time scheduling requirements for automated guided vehicles (AGVs). Additionally, the ADG integrates physical constraints and spatial autocorrelation (Moran’s I = 0.3064) to generate high-fidelity, three-dimensional yard configurations at a rate of 105 instances per cycle. Predictive models trained on this dataset achieve coefficient-of-determination values of R2 ≥ 0.85 (peaking at 0.882) across large-scale fully automated, medium-scale semi-automated, and small-scale conventional yard typologies. The OLCF methodology extracts and quantifies 17 key performance indicators. A multi-layer stacked ensemble predicts relocation counts with 90.76% accuracy (R2 = 0.9139), while a dynamic constraint-weighting mechanism balances movement frequency and energy consumption, thereby enhancing green operational efficiency in high-density container yards. Discussion: From both theoretical and practical perspectives, this work establishes a multi-stage collaborative optimization pathway by systematically integrating data-driven and model-driven approaches, limits strategy-generation time for 105-container-scale yards to under 60s, and provides a scalable technological paradigm for smart-port development, sustainable logistics, and the attainment of dual-carbon objectives.