A Deep Reinforcement Learning Method For Job Shop Scheduling Problems in Traffic Management Using Graph Neural Networks

Abstract

A crucial component of urban infrastructure is traffic management, which encompasses a variety of challenges such as reducing congestion, minimizing costs, and optimizing vehicle movements. In this context, this study presents an innovative framework for solving job shop scheduling problems (JSSPs) by leveraging deep Q-Learning and graph neural networks (GNNs). These challenges arise across various traffic management scenarios, including air traffic control, train schedule management, and urban traffic optimization.The framework employs a single-policy model, similar to a constructive heuristic algorithm that builds solutions incrementally, ensuring that decisions are made based on real-time data and dynamic conditions. It is trained on viable rules and reward signals that guide the Q-learning agent in making optimal scheduling decisions. The GNN component processes the partial solution at each step, capturing complex relationships and dependencies within the traffic environment. This allows the agent to adapt its actions based on the evolving state of the system, ensuring more accurate and efficient management.Extensive testing across different-sized JSSP instances highlights the framework’s competitiveness in optimizing traffic management processes. The integration of GNNs provides a deeper understanding of the interconnections between various tasks and machines, while deep Q-Learning offers adaptability to dynamic and unpredictable traffic situations. By continuously learning from these interactions, the framework demonstrates significant potential for improving traffic flow, reducing delays, and enhancing the overall efficiency of transportation systems.Moreover, the study acknowledges the need for further validation in real-world scenarios, as current testing focuses primarily on controlled environments. Ensuring the framework’s adaptability to complex, real-time traffic situations is essential for practical implementation. By refining the model through real-world data and continuous evaluation, the framework can effectively address the challenges of urban traffic management, providing scalable and sustainable solutions for future smart cities.