A dual-population memetic algorithm based on a producer-consumer paradigm for energy-aware dynamic flexible job shop scheduling with AGVs

The flexible job shop scheduling problem with automated guided vehicles (FJSP-AGVs) is a key Industry 5.0 challenge that requires both high energy efficiency and strong adaptability. This work investigates an energy-aware dynamic flexible job shop scheduling problem with AGVs (EDFJSP-AGVs), which aims to jointly minimize the makespan and the total energy consumption. A mixed-integer programming (MIP) model is established for the static problem and its correctness is validated using a solver. To enhance system robustness against AGV failures, a dynamic decoding-based rescheduling mechanism is designed to enable rapid task reallocation. To solve EDFJSP-AGVs, we propose a dual-population memetic algorithm (DPMA) inspired by the producer-9consumer paradigm. It combines a global search population guided by deep Q-network (DQN)-based adaptive crossover selection with a local refinement population using variable neighborhood search (VNS). Comprehensive experiments on 20 benchmark instances demonstrate the superiority of DPMA over five state-of-the-art algorithms. Specifically, DPMA attains the best average rank in hypervolume (1.40), inverted generational distance (1.40), and entropy (1.90) metrics with statistical significance (p < 0.05).