An islanded microgrid has a great effect on ocean islands and remote areas, but the uncertainty of source and load has a negative impact on the stable operation of an islanded microgrid. To reduce the pressure of intra-day dispatch, a two-stage day-ahead dispatch model of an islanded microgrid is constructed. By applying chaotic phase space reconstruction, multi-objective particle swarm, data-driven and linear programming methods, the negative effects of wind curtailment and loss load caused by uncertainties can be reduced through flexible resource regulation. This can reduce the operational cost while giving consideration to the efficiency and reliability of the system. The first stage takes the lowest integrated operation cost of microgrid, the highest utilization rate of renewable energy and the smallest load loss rate of system as objectives, and establishes a multi-objective microgrid economic dispatch model taking into account demand response resources. In the second stage, considering wind curtailment and load loss occurring after the first stage, the dispatch model of consuming wind curtailment and the frequency modulation power dispatch model are constructed by applying an extreme learning machine and XGBoost. Finally, through the comparison of simulation cases, the results show that the enhancement of demand response to system efficiency is at the cost of increasing dispatch cost and reducing load reliability. In contrast, the proposed two-stage dispatch method not only reduces dispatch cost, but also gives consideration to system efficiency and load reliability. This provides a reference for an islanded microgrid for rural areas, islands, etc. This work is supported by the National Natural Science Foundation of China (No. 52177110).