Abstract: As the penetration of distributed generation (DGs) in DC microgrids continues to increase, their output fluctuations and load forecast deviations pose significant challenges to the stability of system energy management. To address the limited adaptability of single-layer optimization methods in dealing with real-time dynamic disturbances, this paper proposes a two-layer optimization energy management framework. The upper layer utilizes an improved multi-objective particle swarm optimization algorithm to enhance population diversity through chaotic initialization, introduces time-varying inertia weights to balance global exploration and local exploitation, and combines a roulette wheel selection mechanism to suppress premature convergence, achieving multi-objective global optimization of economic and environmental costs over a 24-hour period. The lower layer, based on model predictive control, performs rolling optimization on a 15-minute cycle, provides feedback correction based on the power deviation between actual and planned output, and implements differentiated operation strategies based on time-of-use electricity prices. This strategy effectively enhances the adaptability and robustness of the microgrid in dynamic environments, achieving the coordinated optimization of safe operation and economic benefits.