Abstract: In this paper,a generation model evaluation method based on data analysis is proposed to study the PV generation model input. The method consists of three steps. Firstly,feature extraction based on signal analysis and feature engineering based on expert knowledge are combined to expand the data set,and outlier detection is performed to remove outlier samples. Secondly,the rationality of the input data is discussed through correlation analysis of the data set. Finally,the data set is modeled by artificial neural network,and principal component analysis is introduced into model training,and principal component analysis is introduced into the model training to analyze the performance of each model under three different meteorological conditions: sunny,rainy,and cloudy.Experiments show that the calculation result of the model trained by constructing the data set is more accurate than that trained by original data set,while the model introduced with principal component analysis is more efficient.