For the rapid detection of burst pipes in water supply networks of district metering area, a new prediction-classification-correction three-stage Infomer-Z-score algorithm was proposed. The Infomer-Z-score algorithm solves the problem of low data processing efficiency and abnormally low water consumption in traditional methods. In the prediction stage, the deep learning Informer algorithm was used to predict long sequence time-series of pressure data for the pipe network so as to improve the accuracy of water pressure prediction and the efficiency of data processing. In the classification stage, the robustness of water pressure data over time was improved by using a multi-threshold classification method. In the pipe burst simulation test, the Infomer-Z-score algorithm achieved a 99.5% (DA) detection accuracy with a 90.9% true positive rate (TPR), and a 1.7% false positive rate (FPR). Long sequence time-series pressure forecasting can be used not only for burst detection, but also for effective pressure control in the network to reduce the risk of bursts.