Abstract: Plain polder areas are characterized by low-lying terrain and are highly susceptible to the combined impacts of external flooding and internal waterlogging. Insufficient hydrodynamic conditions are a key limiting factor for improving water environment in these regions. To address the subjectivity and difficulty of achieving global optimization associated with traditional regulation methods, this study takes the Dadang polder in the Taihu Basin as the study area. Joint inner-outer polder hydrological scenarios were constructed using a Copula function. The InfoWorks ICM model was employed to simulate the hydrodynamic processes within the polder area, and a surrogate model integrating CNN-LSTM-SelfAttention architecture was coupled with a genetic algorithm to perform multi-objective coordinated optimization of hydrodynamic regulation. The results show that the addition of a single diversion sluice increased the water exchange rate from 1.72% to 36.69%, which was further improved to 49.82% after coordinated optimization of gate–pump operation. Under the scenario of external flooding and heavy rainfall, the channel water level could be stably maintained below the flood warning level while remaining above the minimum control water level after water withdrawal. Furthermore, the duration guarantee rates of flow velocity and discharge in the river network were significantly improved. These findings can provide theoretical and technical support for flood control and water environment improvement in plain polder areas.