Abstract: In the context of achieving the "four pres" (forecasting, early warning, drill and emergency plan) of intelligent water conservancy goals, in view of the fact that a single reservoir is unable to coordinate the multi-objective, multi-attribute, multi- level, and multi-stage flood control contradictions in the upper and lower reaches of the reservoir, an improved non-dominated sorting genetic algorithm-based reservoir group joint flood control optimal operation model is proposed. Taking the typhoon No. 2106 "In-Fa" and typhoon No. 2114 "Chanthu" experienced in Yuyao City as examples, the outflow joint operation schemes of a "one large and three medium" reservoir group are selected for simulation. The calculation results show that: compared with the regular operation, the highest water level at Yuyao Station during the typhoon "In-Fa" decreased by 0.17 m after being regulated by the improved genetic algorithm-based reservoir group joint optimal operation model, with a decrease amplitude reaching 4.82%; during the typhoon "Chanthu", the highest water level decreased by 0.14 m compared with the unregulated situation, with a decrease amplitude reaching 4.79%. This proves the effectiveness of this model in peak-shaving, in order to provide reference for the rational formulation of reservoir scheduling plans