Abstract:To achieve refined optimal operation of drainage systems, a daily inflow predication method for pumping stations based on water balance and Stacking fusion is proposed for the practical problem that some drainage pumping stations lack reliable real-time inflow monitoring data. Firstly, the simulation model of the front tank of the pumping station and the upstream pipeline network is constructed based on InfoWorks. A liquid level-volume functional relationship is derived. Combined with measured effluent flow data and the water balance principle, the daily inflow volume of the pumping station is back-calculated to construct the modeling dataset. Secondly, Random Forest (RF), XGBoost and LightGBM are adopted as base model, and an integrated prediction model is built with the Stacking fusion strategy to effectively integrate the prediction results of multiple models. Validation is carried out by using field operational data of an actual pumping station in East China. The results demonstrate that the ERMS of the fused model on the test set reaches 3, 523.63, 14.2% lower than the best single model. The mean absolute percentage error (EMAP) is 14.22%, a decrease of 22.2%. The coefficient of determination (R2) reaches 0.841. The study result verifies that the proposed approach can significantly improve the accuracy of daily-scale inflow predication under insufficient monitoring data, offering solid data support for refined optimal operation of drainage systems.