In order to increase the revenue of the microgrid optimized dispatch and reduce its pollution cost, an improved Non-dominated Sorting Genetic Algorithm (NSGA-II) is proposed. The specific improvement method is to use Latin hypercube sampling to generate the initial population and incorporate taboo search into the elite strategy. Taking the maximum economic benefit of the microgrid system and the minimum pollutant emission cost as the optimization goal, the peak-valley time-of-use electricity price mechanism is adopted to construct a grid-connected microgrid model containing 5 dispatch variables. Comparing the fitness convergence curves of GNSGA-II, NSGA-II, and MOEA/D algorithms through MATLAB simulation, it is preliminarily proved that the improved algorithm converges faster and has higher convergence; compare the Pareto frontier solutions of GNSGA-II algorithm and NSGA-II algorithm Based on the analysis of the 24h micro-source output curve based on the optimal compromise solution, the results further confirm its feasibility and superiority.