[Objective] Model Predictive Control （MPC） has been increasingly applied to the coordinated regulation of sluices and pumps in urban drainage systems to maximize the utilization of existing storage capacity and mitigate waterlogging risks. However，research on leveraging observational data to improve model prediction performance remains limited. This study focuses on measurement uncertainty in real-time control of urban drainage systems and proposes an Unscented Kalman Filter-based Model Predictive Control （UKF-MPC） method to enhance system adaptability under complex conditions. [Methods] By integrating model predictions with sensor observations，the proposed method incorporates data assimilation into the error feedback strategy and state update of the MPC prediction model，establishing a real-time data-driven MPC framework. The approach was validated in the Doumen area of Fuzhou City. [Results] （1） UKF-MPC significantly improved flow prediction accuracy by dynamically fusing observational information with model outputs，increasing the Nash-Sutcliffe Efficiency （NSE） from 0.39-0.58 to 0.56-0.79，with a mean Relative Improvement Percentage （RIP） of 36.4%. （2） Through data assimilation，the method upgraded traditional error feedback to state-space correction. Under 50-year and 100-year return period storm scenarios，UKF-MPC consistently generated more robust control strategies，enhancing system adaptability in complex environments and providing a framework capable of real-time system state perception and dynamic control optimization. [Conclusions] Compared with conventional MPC，the control strategies derived from UKF-MPC exhibit superior applicability and robustness under uncertain conditions. Future research will quantitatively analyze the influence of UKF parameter settings on correction performance and conduct systematic validation across multiple watersheds and diverse rainfall events using richer observational data，to further clarify the applicability boundaries and parameter sensitivity mechanisms of UKF-MPC. On this basis，further exploration will focus on incorporating state covariance information directly into the optimization problem to develop a risk-aware model predictive control method.