The University of Sheffield
F94.pdf (468.11 kB)

CCWI2017: F94 'Valve Status Verification and Sensor Error Detection via Causal Inference from Sensor Data'

Download (468.11 kB)
journal contribution
posted on 2017-09-01, 15:05 authored by Dirk Vries, Joost van Summeren
Recent developments in (near) real-time sensor applications have the potential to provide operators and managers with useful information on drinking water distribution supply and need of its maintenance. A systematic methodology based on causal inference from observational data is proposed to increase knowledge of water supply distribution systems equipped with sensor networks. This methodology can be used to help identify deviations from expected operation of water supply and sensor infrastructure, using only observational data. We outline the first steps of two distinct procedures that use data from a sensor network, to infer a map of a causal dependence structure. These procedures are applied to scenario studies where an unexpected change in operation occurs, i.e. a valve status is different and a sensor bias is introduced. A draft outline of future steps is given that could improve and validate the methodology.