The invention provides a reservoir stage power generation scheduling rule extraction method. Stage influence factors are screened based on grey relational analysis, a multivariate nonlinear regressionmodel, a support vector machine model and a BP artificial neural network model are adopted to fit and then obtain a stage reservoir power generation scheduling function, and a Bayesian model averaging method is applied to perform multi-model weighted averaging to obtain a final stage reservoir power generation scheduling function. According to the invention, GRA is introduced into the reservoir scheduling rule simulation for influence factor screening, so redundant attributes are removed, the model complexity is reduced, and the model simulation efficiency is improved; GRA and EVA are combined to realize the stage extraction of reservoir power generation rules, so the inconsistency of the correlation between decision variables and the influence factors for different periods of reservoir scheduling is balanced, the uncertainty of a single model which may be caused by the model structure is balanced, and the model simulation precision is improved; the requirement for reservoir power generation scheduling rule extraction is met, and the benefits of determinacy optimization scheduling in the power generation process are well inherited.