The invention discloses a wind field power predication method based on a deep neural network in the technical field of power system prediction and control. The method includes the steps that firstly, data of the wind speed, the wind direction, the temperature, the humidity and the atmospheric pressure which are actually measured are obtained; secondly, the projection pursuit is adopted for extracting main constituents, the neutral-position absolute deviation serves as a projection index in the projection pursuit, the interference of outliers irrelevant or little relevant to data structures and features can be effectively removed, and the main constituents can be extracted stably. A deep neural network model is adopted, and a predication model between five influence factors and the output power of a wind field is established, wherein the five influence factors include the wind speed, the wind direction, the humidity, the temperature and the atmospheric pressure. The power of the wind field is predicated, and the predicated power is obtained. According to the wind field power predication method, the precision of predicated power in future 72 hours of the wind electric field is improved, a basis is provided for reasonable dispatching of power grids, and the grid connection pressure is relieved.