Abstract: The microgrid system normally consists of a variety of distributed power sources. To cut the operating cost of the microgrid, intelligent algorithms are often employed to dispatch the microgrid. Intelligent algorithms are prone to fall into local optimal solutions when solving microgrid scheduling models, resulting in poor accuracy. Therefore, based on the Northern Goshawk algorithm, this paper proposes a hybrid strategy improved Northern Goshawk algorithm (HNGO), which uses reverse learning, Metropolies criterion and adaptive T-distribution variation to enhance its accuracy. Meanwhile, a demand response model considering the output characteristics of renewable energy is built, so that the load curve is closer to the output curve of renewable energy. Then, a microgrid optimization scheduling model with the lowest daily operating cost is established, and HNGO is used to find the solution. Our simulation results show the proposed algorithm achieves accuracy, and our proposed demand response model significantly reduces fuel costs.