An Intelligent Scheduling Method for Heterogeneous Equipment Maintenance Tasks Based on Deep Reinforcement Learning and Genetic Algorithms

In the context of smart manufacturing, equipment maintenance task allocation faces multiple challenges, including heterogeneous workstation efficiency, spare parts inventory constraints, and dynamic resource fluctuations. To address the high computational complexity and poor robustness of traditional optimization methods in large-scale scheduling scenarios, this paper proposes a hybrid optimization framework integrating deep reinforcement learning (Advantage Actor-Critic, A2C) with genetic algorithms (GA). By constructing a workstation-task efficiency matrix and a composite state space, a multi-objective reward function is designed to guide the agent in achieving dynamic task scheduling within a Markov decision process. To overcome the local optima trap in policy learning, genetic algorithms periodically evolve the agent's hyperparameters and network weights, enhancing global search capabilities. Simulation results demonstrate that the proposed method outperforms traditional heuristic approaches and standalone reinforcement learning algorithms in minimizing makespan and achieving load balancing, validating its effectiveness and scalability in complex manufacturing system scheduling optimization.