Fault detection in water supply systems using hybrid (theory and data-driven) modelling

摘要: In this paper we present a complex hybrid model in the water management field based on synergetic combination of deterministic and machine learning components. The objective Water Supply System (WSS) is to convey treated consumers through pressurized network pipes. A number meters gauges are used take continuous or periodic measurements that sent via telemetry system control operation center monitor network. Using typically limited measures together with demand predictions state must be assessed. Suitable estimation paramount importance diagnosing leaks other faults anomalies WSS. But task can really cumbersome, if not unachievable, for human operators. aim explore possibility technique borrowed from learning, specifically neuro-fuzzy approach, perform such task. For one thing, performed by using optimization techniques minimize discrepancies between taken values produced mathematical network, which tries reconcile all available information. But, another, although completely accurate, data containing non-negligible levels uncertainty, definitely influences precision estimated states. quantification uncertainty input (telemetry predictions) achieved means robust estate estimation. By making use states levels, say, fuzzy states, different anomalous obtained. These two steps rely theory-driven model. final train neural (using description associated anomaly) capable assessing WSS particular sets received predictions. This data-driven counterpart