Abstract
In addressing the inherent inefficiencies of centralized scheduling systems in automated container terminals, this study proposes a Cloud-Edge-Fog Distributed Collaborative Intelligence (CEF-DCI) framework integrated with a hierarchical game-theoretic model to optimize resource allocation and coordination among heterogeneous equipment clusters. The framework combines cooperative game theory at the upper layer for global benefit allocation using Shapley values, and non-cooperative game theory at the lower layer for autonomous equipment decision-making under quota constraints. Through high-fidelity digital twin simulations based on Yangshan Port’s 2023 operational data, the model demonstrates significant performance improvements: an 18.7% increase in daily throughput, a 31.7% reduction in AGV idle time, a 22.3% shortening of container handling cycles, and a 28.5% enhancement in equipment utilization, compared to traditional centralized approaches. The study further validates the system’s robustness under throughput fluctuations and equipment failures, highlighting its practical viability for real-world port operations. These findings provide a scalable, efficient, and resilient solution for next-generation intelligent port systems.
Introduction: The Scheduling Dilemma in Automated Terminals
Automated Container Terminals (ACTs) represent the pinnacle of modern logistics, yet they face a persistent "Goldilocks" problem in scheduling. Operators must manage a complex choreography of quayside cranes, yard cranes, and fleets of Automated Guided Vehicles (AGVs).
Traditional centralized systems, while theoretically optimal, often suffer from communication latencies exceeding 500ms. In a high-speed port environment, such delays lead to traffic congestion and idle equipment. Conversely, fully distributed systems offer rapid response times but frequently fall victim to the "tragedy of the commons," where individual robots compete for the same paths or tasks without a global coordinator, leading to systemic inefficiency. This research introduces the Cloud-Edge-Fog Distributed Collaborative Intelligence (CEF-DCI) framework to bridge this gap, balancing global strategy with local autonomy.
Methodology: The CEF-DCI Framework
The proposed architecture distributes the computational workload across three tiers: the Cloud handles long-term strategy, Fog nodes (regional servers) manage area coordination, and Edge nodes (the robots themselves) execute real-time tasks. The core of this system is a two-layer hierarchical game-theoretic model:
- Upper Layer (Cloud-Fog): This layer utilizes Cooperative Game Theory. Equipment clusters are treated as "super-agents." By employing Shapley values, the system ensures fair and efficient task allocation across the port, effectively preventing any single area from becoming a bottleneck.
- Lower Layer (Fog-Edge): This layer operates on Non-Cooperative Game Theory. Individual AGVs act as independent agents seeking to maximize their own efficiency within the quotas established by the upper layer. Modeled as a Generalized Nash Equilibrium (GNE), this allows robots to make split-second decisions regarding pathing and collision avoidance without waiting for centralized instructions.
This dual-layer approach effectively cures the "herd effect" common in centralized systems, where multiple robots rush toward the same high-priority task while others remain underutilized.
Experimental Validation: Digital Twin of Yangshan Port
The framework was validated using a high-fidelity digital twin built in AnyLogic Professional, utilizing 2023 operational data from Shanghai Yangshan Port Phase IV—the world’s largest automated terminal. The simulation environment included:
- Scale: 2.8 million container movements and historical AGV trajectories sampled at 10Hz.
- Hardware: Modeling of 5 quay cranes, 8 rail cranes, and 20 AGVs across a 120,000 square meter yard.
The study found that while the Cloud-Edge-Fog architecture itself accounted for 65% of performance gains, the hierarchical game mechanism provided the critical 35% marginal value required for elite performance levels.
Key Results and Performance Gains
Compared to traditional Centralized Scheduling (CS), the CEF-DCI framework demonstrated superior performance across all critical metrics:
- Throughput: An 18.7% increase in daily containers handled (approximately 323 additional containers per day).
- Operational Efficiency: A 31.7% reduction in AGV idle time and a 22.3% reduction in the overall handling cycle.
- Resource Utilization: Equipment utilization rose from 62.3% to 80.1%, approaching the theoretical limits of advanced terminals.
- Latency: Communication latency was reduced from 500ms to 50ms, enabling safer and more precise coordination.
Furthermore, the system exhibited remarkable resilience. In the event of a Cloud failure, Fog nodes were capable of maintaining port operations for up to two hours with only a minor dip in efficiency.
Conclusion and Future Directions
The CEF-DCI framework provides a scalable and resilient solution for the complexities of modern port automation. While the current model assumes a stable 5G network and treats equipment failures as independent events, future research will address cascading failures and integrate auxiliary tasks such as customs inspections. Additionally, the researchers aim to explore Multi-Agent Deep Reinforcement Learning (MADDPG) to further enhance the speed and adaptivity of the game-theoretic solvers.
Summary updated