To address the uncertainty of intermittent energy sources and improve the economic efficiency and operation performance of microgrids, a novel three-layer coupled microgrid dispatch model is proposed based on model predictive control. The model is optimized and solved using an improved dung beetle algorithm. First, by comprehensively considering time-sharing tariffs and pollution control costs, the model mitigates the impact of day-ahead uncertainties and performs online correction of the optimal control variables, thereby forming a new three-layer scheduling structure. Second, improvements are made to the traditional dung beetle algorithm in aspects such as population initialization, rolling behavior, and foraging behavior. The feasibility of the algorithm is verified by simulation, showing improved accuracy and convergence speed. Subsequently, the improved dung beetle algorithm is used to optimize the economic performance of the scheduling layer and determine the optimal control actions in rolling control. Finally, comparative analyses in terms of economic performance, rolling scheduling, and control effect demonstrate that the proposed model and algorithm can effectively enhance the environmental and economic performance of microgrids while improving system controllability and stability.