Machine learning in real-time control of water systems

Abstract

In real-time control (RTC) of combined urban and rural water systems the so-called centralised control requires information from different locations in the water system and hence sensitive to the communication network breakdown during extreme storm runoff events. Optimisation algorithms used in advanced forms of centralised control require considerable computing times and thus may be impractical for RTC. To overcome these problems, the application of machine learning methods is proposed, using artificial neural networks and fuzzy adaptive systems. Results obtained in a realistic case study show that the trained controllers, can replicate centralised control behaviour quite accurately and rapidly, while using only local data sources.