Abstract: Accurate prediction of water supply is one of the key factors for the efficient and stable operation of urban water supply systems. In particular,the accurate forecast of short-term precipitation will help narrow the gap between production and sales and significantly reduce energy consumption. This paper proposes an artificial bee colony optimization multi-variable gated recurrent unit model（WD-ABC-MGRU）based on wavelet decomposition,which can effectively capture multi-variable and multi-scale time series characteristics,thereby significantly improving the model’s ability to handle complex patterns and achieve accurate water supply prediction. WD-ABC-MGRU selects the strongly correlated factors affecting water supply,combines the components of water supply obtained by wavelet decomposition and coefficient reconstruction and performs phase space reconstruction to form the model input,builds the model separately for prediction,and sums the predicted values of each model to obtain the final water supply prediction result. The experimental results show that the WD-ABCMGRU model provides more accurate water supply forecasts than the GRU,MGRU,WD-MGRU and WD-ABC-BP models,with an R2 of0.87 and a MAPE of 1.75%. It can be seen that this model can provide a practical solution for the accurate prediction of urban water supply,thereby contributing to the efficient and stable operation of the urban water supply system.