To address the issues of high generation costs and insufficient economic benefits of microgrid clusters under complex constraints, an optimal scheduling model for microgrid clusters based on the dual-population golden jackal optimization (DGJO) algorithm is proposed. First, with the objective of minimizing the total cost, an optimal scheduling model for microgrid clusters is constructed, incorporating operation costs, energy storage costs, electricity trading costs, and environmental costs. Second, the DGJO algorithm is developed, in which Lévy flight is employed to achieve adaptive convergence, a dual-population strategy is adopted to balance exploration and exploitation, and Harris hawks encircling and cache-foraging operators are incorporated to improve optimization accuracy. Then, DGJO is applied to optimize the hyperparameters of the temporal convolutional network (TCN) and the bidirectional gated recurrent unit (BiGRU), thereby improving convergence speed and model generalization capability. Finally, case studies demonstrate that the proposed model exhibits strong robustness under complex constraints and disturbance scenarios and effectively reduces the overall system cost.