Abstract: Accurate daily water supply forecasting is crucial for achieving refined scheduling and cost reduction and efficiency improvement goals in urban water supply systems. This article systematically studies the variation patterns of daily water supply in Guangzhou based on multi-source heterogeneous data from 2017 to 2024, including meteorological characteristics, holiday information, and historical water volume of various water plants and zones. Through methods such as time series analysis, autocorrelation, and partial autocorrelation analysis, the seasonality of "high in summer and low in winter" in water supply, the periodicity of "high on weekdays and low on weekends", and the "valley effect" during the Spring Festival were revealed. And based on this, the performance of eight machine learning models, including linear regression, random forest, and LightGBM, was compared and analyzed. Among them, the LightGBM model optimized by hyperparameters performed the best, effectively integrating multidimensional feature factors in the daily water supply prediction task in Guangzhou, with an average absolute percentage error(MAPE) as low as 1.01%. This study provides a high-precision water volume prediction core algorithm for the development of smart water management in Guangzhou, which has important theoretical value and engineering application prospects.