FAG-scheduler: Privacy-Preserving Federated Reinforcement Learning with GRU for Production Scheduling on Automotive Manufacturing

The automotive manufacturing industry faces challenges in production planning, but current heuristic algorithms and solvers have limitations in scalability and local optima. Moreover, data security concerns are often overlooked. To address these issues, this paper introduces the FAG-Scheduler, a federated reinforcement learning approach integrating asynchronous advantage actor-critic, gated recurrent unit algorithms, and federated learning. By sharing model parameters instead of raw data, data security is ensured among participants. The FAG-Scheduler achieves optimal solutions in under 5 seconds and demonstrates high adaptability to other manufacturing contexts. It presents potential applications with significant improvements over conventional methods.