Leakage zone identification for water distribution networks based on the alarm levels of pressure sensors

When a leakage event occurs, the values of pressure sensors in the water distribution network will drop, and leakage alarms will be triggered if the drop in pressure values exceed the alarm thresholds. Due to the similarity of the leakage characteristics between the adjacent nodes, it is difficult to identify the exact leakage node. Therefore, in this paper, a leakage zone identification method based on alarm levels and pattern identification is proposed. At first, leakage residual samples for each node are generated within the range of the leakage amount. To reduce the influence of nodes with similar leakage residual characteristics on the identification results, the residual values of each sample are converted into alarm levels. For the training samples, nodes with the same alarm level sample are merged into a node group and used as a label. Then, the Euclidean distance method is used to test the identification effect of the model. The enumeration method is adopted to optimize the sample interval, the alarm level interval and the feature dimension to enable the model to achieve an appropriate identification result. A life-sized network is presented in this paper to demonstrate the effectiveness of the proposed method. The results show that compared with the previous leakage zone identification method based on alarm characteristics, the proposed method can effectively reduce the size of the candidate leakage zone.