One of the key areas in which production systems researchers are working these days is to find advanced optimization algorithms to efficiently schedule activities in manufacturing systems, which requires more sophisticated models with increased computational complexity. Therefore, there has been growing interest in this subject to improve the performance of meta-heuristics by incorporating reinforcement learning approaches. This paper deals with a dual-resource flexible job shop scheduling (DRFJSS) problem, in which each operation requires two resources (i.e., reconfigurable machine tool (RMT) and worker) to be processed. A mixed-integer linear programming (MILP) model is formulated to minimize the makespan. Since the proposed model cannot optimally solve most medium-sized instances, a self-learning whale optimization algorithm (SLWOA) is developed to deal efficiently with such a difficult problem. In the proposed SLWOA, an agent is trained by the state–action–reward–state–action (SARSA) algorithm to balance exploration and exploitation. The results show that the SLWOA has a stronger global search ability and faster convergence speed than the original whale optimization algorithm.