To maximize the comprehensive benefits of multifunctional reservoirs, a multi-objective optimization operation model for power generation, water supply, flood control, ice prevention, and ecology of cascade reservoirs in the upper reaches of the Yellow River was constructed. Multiple linear regression model, random forest model, and LightGBM machine learning model were combined to extract multi-objective operation rules for cascade reservoirs, and the extraction effect was evaluated by combining improved Taylor plots and typical operation schemes. The results show that the two machine learning models have overall better performance in extracting operation rules than traditional multiple linear regression model. The newly introduced LightGBM model training results have more practical guidance significance, and can effectively guide the optimization of reservoir discharge under power generation, water supply, and balanced operation needs.