Online Application of ANN and Fuzzy Logic System for Burst Detection

Minimising the loss of treated water from water supply systems due to burst and leakage is an ongoing issue for water service providers around the world. Flow monitoring techniques are currently used by the water industry to monitor leakage, generally offline through the application of mass balance type calculations or through observations of change in night line values. The data for such analysis has, until recently, been at best collected 24 hourly via SMS technology. The objective of the study reported here was to assess the online application of an AI system to a real distribution system and the potential benefits of so doing. Specifically the application of Artificial Neural Networks (ANNs) and Fuzzy Inference Systems (FIS), which are computational techniques in the field of Artificial Intelligence (AI). The online hybrid ANN/FIS system developed uniquely uses DMA (District Meter Areas) level flow data for the detection of leaks/bursts as they occur. The ANN model (a Mixture Density Network) was trained using a continually updated historic database that constructed a probability density model of the future flow profile. A FIS, used for classification, compared observed flows with the probability density function of predicted flows over time windows such that confidence intervals could be assigned to alerts and further, an accurate estimate of likely burst size provided. A Water Supply System in the UK was used for the case study. The case study pilot area has near real-time flow data provided by General Packet Radio Service (GPRS). The online AI leak/burst detection system was constructed to operate along side an existing flat line alarm system, and continuously analyse every twelve hours a set of 50 DMAs of various size, complexity and connectivity within the case study area. Results are presented from a six month period. The new system identified a number of events and alerts were raised prior to their notification in the control room; either through flat line alarms or customer contacts. Examples are given of their correlation with burst reports and subsequent mains repairs. 56% of AI alerts were found to correspond to bursts confirmed by repair data or customer contacts reporting bursts. The study shows that the integration of the AI system with near real time communications can facilitate rapid determination (i.e. before customers are impact) of abnormal flow patterns. It is concluded from the study that the system is an effective and viable tool for online burst detection in water distribution systems.