This paper studies the distributed flexible job shop scheduling problem incorporating just-in-time production consideration. Enterprises need to determine the number of factories to activate based on job delivery time and schedule jobs within activated factories to minimize the makespan, energy consumption, and production cost. A multi-objective mixed-integer linear programming model is constructed and solved by a memetic algorithm with reinforcement learning-driven evolutionary strategy. A heuristic initialization strategy is adopted to breed a high-quality population. During evolution, reinforcement learning treats evolved individuals as states, and the selection of partners as actions, deriving rewards from the quality of offspring. By recommending suitable partners for each parent, the blindness inherent in random matching is reduced. Finally, an adaptive local search strategy is applied for stagnant individuals to enhance exploitation performance. The effectiveness of the proposed algorithm is verified by simulation experiments conducted on two standard test suites and comparisons with five representative algorithms.