Wind energy is a pollution-free renewable energy, and the proportion of wind power is increasing year by year globally. In view of the large fluctuations in the output of wind power generation, which leads to the instability of the grid power, a short-term wind power prediction model based on integrated multi-scale long short-term memory (LSTM) is proposed. By using LSTM’s special processing capabilities for sequence data, combined with the information contained in different scales’ time data, to predict short-term wind power after integration. It is conducive to comprehensive control and dispatch of power resources. Experiments are conducted on the real data set of wind power generation in the northeast of our country, and the results prove that the method in this paper has high prediction accuracy and strong stability.