We propose a memory-enhanced improvement search framework based on deep reinforcement learning to solve flexible job-shop scheduling problems.