Wind power probability prediction method for improving prediction precision of low-density sample region

The invention discloses a wind power probability prediction method for improving low-density sample region prediction precision. The method comprises the steps of collecting historical wind power and a numerical weather forecast data set of a certain wind power plant; constructing a deep belief mixed density network, extracting hidden features of input variables through a unique pre-training and fine tuning mechanism of the deep belief network, accurately representing probability distribution of wind power prediction power by using boundaries of Beta mixed probability distribution, and realizing nonlinear mapping between the hidden features and prediction power probability distribution parameters; a feature distribution smoothing technology is introduced to calibrate input features, and a label distribution smoothing technology is used for endowing each sample error with a differential weight, so that the adverse effect of a training sample distribution imbalance phenomenon on a prediction result is improved