Microgrid energy management plays a crucial role in ensuring the security and economy of microgrid. This paper proposes an energy management strategy of microgrid based on deep Q-network, which takes the renewable energy generation and load demand into comprehensive account. Regarding the stochastic characteristics of photovoltaic power in day-ahead scheduling, the stochastic programming is utilized to generate multiple photovoltaic output scenarios under various probabilities. It is combined with wind power, load power, market electricity price and the state of charge of battery to form the environmental information, and the operation index of microgrid is optimized through the consistent interaction of deep Q-network and environmental information. The simulation results signify that the proposed energy management strategy can reduce the safety index cost of equipment and improve the utilization rate of renewable energy. In the scenario with strong randomness of photovoltaic output, it is verified that the deep Q-network energy management strategy based on stochastic programming has excellent adaptability.