Application of machine learning techniques to predict anomalies in water supply networks

摘要: Methods to improve the operational efficiency of a water supply network by early detection anomalies are investigated making use data streams from multiple sensor locations within network. The is demonstration site Vitens, Dutch company that has several district metering areas where flow, pressure, electrical conductance and temperature measured logged online. Three different machine learning approaches tested for their feasibility detect anomalies. In first approach day-dependent support vector regression (SVR) models trained predicting measurement signals compared straightforward using mean median estimates, respectively. Using SVRs or averaged as real-time pattern recognizers on all available signals, large leakages can be detected. second utilizes adaptive orthogonal projections reports an event when number hidden variables required describe streaming user-defined degree (energy level threshold) increases. As third approach, (unsupervised) clustering techniques applied underlying patterns raw streams. Preliminary results indicate current set too limited in amount events harness potential these techniques.