AGVs have gained significant popularity in various industries. However, the existing literature rarely considers the integrated scheduling of production and logistics on the workshop due to the NP-hard property of both machine scheduling and AGV scheduling. The energy-efficient job shop scheduling problem with limited AGVs is investigated in this paper. A multi-objective memetic algorithm with deep Q-network (DQNMMA) is proposed to minimize the makespan and total energy consumption. In DQNMMA, ten features are selected to describe the current state of the population. This enables the DQN to dynamically adjust the crossover probability according to the population evolution. Formulas for calculating the head length and tail length of each node in the disjunctive graph model are presented for the first time to enable fast and accurate access to the critical paths. Building upon the analysis of critical paths, four problem properties are developed as the foundation for designing six neighborhood operators. Then, a property-based variable neighborhood search strategy is proposed to enhance the exploration capability of the algorithm. Numerous experimental results demonstrate that the proposed approaches can effectively enhance the performance of the algorithm, especially in solving large-scale problems. The comparative analysis with three other state-of-the-art multi-objective algorithms confirms the superiority and effectiveness of the proposed DQNMMA.