Variable selection is the basis of GM(1,N) model, and also one of the key factors that affect grey model performance. Based on the adaptability between actual data and theoretical model, this paper recognizes the parameter estimation of grey model as a general linear model and discusses the effect of variable selection on parameter estimation and prediction performance. It is proved that discarding those relevant variables which have little influence is helpful to improve the modeling accuracy. Then, a variable selection method for GM(1,N) model under RMS, R2 and AIC criteria is proposed, and the calculation order and calculation method are given. Finally, the method is applied to predict the agricultural drought vulnerability in Xinyang City, Henan Province. The results show that the method not only identifies the main influencing factors of drought vulnerability but also further improves the prediction accuracy