Abstract: To better understand complicated nonlinear relationships between the input and output variables of optimal operation for cascade hydropower stations, this study develops a projection pursuit regression model using a ridge function of nonlinear Hermite polynomials and optimizes the operation with a real accelerating genetic algorithm. This model combines correlation analysis with multivariate stepwise regression analysis method to select the independent variables of dispatching function, obtaining a better way to determine the dispatching factors in formulating optimal operation rules for the cascade stations. Application of the model to one example shows that its fitting accuracy and robustness in solving dispatching function are superior to BP artificial neural network model.