An Edge Cloud Collaborative Leakage Detection Approach for Urban Water Distribution Systems

Leakage detection for urban water distribution systems is important. Traditional cloud-based urban water leakage detection methods are not applicable due to excessive bandwidth consumption. Meanwhile, simple edge-based leakage detection methods suffer from poor computing resources for complex algorithms and updates. Therefore, an edge cloud collaborative urban water pipeline leakage detection method (ECCLD) is proposed. The training of the prediction models and the calculation of thresholds are carried out in the cloud. Online leakage detections, however, are performed at the edges. In addition, the cloud regularly updates and deploys the parameters of the prediction models and the thresholds to the edges in case of changes caused by, for example, temperature, economy, or estate constructions to guarantee accuracy (ACC). The neural network (NN) model used in this method is the long short-term memory (LSTM) network and is combined with transfer learning (TL) for the construction and reconstruction of the pipeline flow prediction models at different edges. To validate the proposed method, the BattLeDIM dataset is employed. The results show that the presented method has high prediction ACC, high detection ACC, and low bandwidth consumption.