The effective forecasting of urban water supply network leakage is the premise and basis for network leakage control. In response to the problem that Long Short-Term Memory (LSTM) network tends to ignore the importance of data information, an urban water supply network leakage forecasting model is proposed based on the combination of LSTM and Attention Mechanism (AM). In this paper, the original leakage data of the District Metering Area in Zhuzhou City are used as the research object and pre-processed by Savitzky-Golay filter smoothing. Then, the LSTM network is used to extract the trajectory features of the leakage data. Finally, the AM network is introduced to extract the key feature information and automatically fit the weights to strengthen the ability of the model to obtain the key feature information of the leakage data. The experimental results show that the proposed LSTM-AM model could better capture correlations and has higher accuracy in pipe network leakage forecasting than the typical LSTM model and CNN-LSTM model.