Abstract: Aimed at the scheduling problem of urban water distribution system, a water supply pressure prediction model based on LightGBM (Light Gradient Boosting Machine) is proposed. The time characteristics of the historical data on pressure monitoring points were extracted. The monitoring points were sorted according to the feature importance. The control points were selected in two ways: one was according to feature weight, and the other one was combined feature weight and experience. Taking a water supply system in southern China as a research case, the results show it has high and stable prediction accuracy that the control points are selected by combining feature weight analysis and scheduling experience.