Q-Learning-Based Predictive Control for Capacity Adjustment in Reconfigurable Job-Shops

Nowadays, companies are faced with challenges to produce a variety of products with a satisfactory productivity in the presence of unpredictable demand fluctuation that may seriously disturb the work plan and lead to performance deterioration. Capacity adjustment is an effective strategy which typically achieved by labour-oriented approaches. In this paper, we consider a machinery-based capacity adjustment via reconfig-urable machine tools (RMTs) to regulate work in process (WIP) for performance improvement. Yet, the referred optimization and control concerning RMTs allocation are intractable. To this end, we propose an intelligent predictive control scheme that combines genetic algorithm (GA) and Q-Iearning. In this case, the complex and possible frequent reconfiguration derived from RMTs can be firstly well predicted and accordingly decreased, while guaranteeing a desirable production control performance. Also, the computational complexity of the rolling reconfiguration optimization problems can be significantly reduced. The effectiveness of the proposed method is validated in a multi-product multi-workstation job shop system.