Efficient Multi-Objective Optimization on Dynamic Flexible Job Shop Scheduling Using Deep Reinforcement Learning Approach

Previous research focuses on approaches of deep reinforcement learning (DRL) to optimize diverse types the single-objective dynamic flexible job shop scheduling problem (DFJSP), e.g., energy consumption, earliness and tardiness penalty machine utilization rate, which gain many improvements in terms objective metrics comparison with metaheuristic algorithms such as GA (genetic algorithm) dispatching rules MRT (most remaining time first). However, optimization floor cannot satisfy requirements modern smart manufacturing systems, multiple-objective DFJSP has become mainstream core intelligent workshops. A complex production environment a real-world factory causes entities have sophisticated characteristics, job’s non-uniform processing time, uncertainty operation number restraint due avoidance single machine’s prolonged slack well overweight load, make method combination DRL brought up adapt at different rescheduling points accumulate maximum rewards for global optimum. In our work, we apply structure dual layer DDQN (DLDDQN) solve real new arrivals, two objectives are optimized simultaneously, i.e., minimization delay sum makespan. The framework includes layers (agents): higher one is named goal selector, utilizes function approximator selecting reward form from six proposed ones that embody objectives, while lower one, called an actuator, decide optimal rule Q value. generated benchmark instances trained converged perfectly, comparative experiments validated superiority generality DLDDQN.