Construction method, prediction method and system of neural network prediction model

The invention provides a neural network prediction model construction method, prediction method and system, and belongs to the field of wind power plant operation state prediction.The method comprises the steps that SCADA data of a wind turbine generator set and load data of key components of the wind turbine generator set are obtained; converting the wind turbine generator SCADA data and the wind turbine generator key component load data to obtain initial samples, and selecting samples from the initial samples by adopting a neighbor component analysis method; and training, testing and verifying the neural network based on the selected sample to obtain a neural network prediction model. According to the method, sample feature selection and dimension reduction fatigue equivalent load prediction are carried out by adopting a neighbor component analysis method, so that the calculation amount is reduced, and meanwhile, the accuracy and reliability of a prediction result of the constructed neural network predictio