The discharge process at sluice gates in coastal tidal reaches is significantly influenced by tides, upstream inflows, and operational controls, making conventional weir formulas inadequate for accurate simulation of unsteady flow conditions. To improve calculation accuracy, this study proposes a three-layer BP neural network model based on trend variation features. Time-series indicators such as upstream and downstream water levels, their rates of change, water level difference change rate, and discharge change rate are used as model inputs to simulate sluice outflows. Using measured data from four major sluices in the Lixia River region as case studies, we compared the simulation results of the proposed method and conventional weir formulas. Results show that conventional methods yield large errors, while the BP neural network model achieves correlation coefficients above 0.75 and average errors below 3.00% during typical gate operations, indicating that the proposed model has higher accuracy and applicability. This method can effectively capture the dynamic relationship between discharge and tidal level during gate operations and offer a novel approach for accurate simulation under complex hydrological conditions.