Green drainage facilities, such as bio-retention ponds and artificial wetlands, are critical in mitigating waterlogging disasters in urban areas. However, their storage capacities are often underutilised due to challenges such as initial storage volume constraints and inefficient flow path designs. This study presents an integrated approach that leverages advanced algorithms, including the Ensemble Kalman Filter (EnKF), Long Short-Term Memory (LSTM) neural networks, and dynamic programming, to address these challenges. A hydrological simulation model specifically tailored for wetlands and bioretention facilities was developed and experimentally validated, demonstrating high accuracy, with results aligning with observed measurements within 20 min of rainfall onset. Additionally, a novel method for runoff prediction was established, combining EnKF for accurate runoff area inversion and LSTM networks for short-term rainfall forecasting. This approach achieved a maximum error of less than 7 mm during moderate rain and 10 mm during heavy downpours. A control scheme utilising dynamic programming was also designed to optimise the operational management of green drainage facilities, focusing on minimising peak overflow and enhancing coordination within the urban infrastructure. Simulation results indicated that the proposed control strategies effectively reduced overflow peaks. Static predictions yielded reductions ranging from 51.3% to 60.0% during heavy rainfall events, while dynamic predictions achieved reductions between 39.0% and 75.4%. These findings underscore the potential of the proposed methodology to enhance the performance of green drainage facilities and strengthen urban resilience against waterlogging.