Drinking water distribution networks (WDN) carry the drinking water from source to consumers. As ageing infrastructures, they need to be monitored. Using pressure loggers is one possible way of monitoring inconsistencies and possible leakages in WDNs. However, installing pressure loggers is a resource-intensive work, which requires optimizing both the number and the location of pressure loggers. The goal of the presented work is to allocate a minimum number of loggers at most suitable locations, so that most information can be gained at lowest cost.

The optimization starts with allocating only one pressure logger, then gradually increasing the number of loggers. Information gain is continuously evaluated with a score called relative entropy. This score is obtained by Bayesian calibration of a WDN model with different numbers and locations of pressure loggers. The Bayesian calibration considers a Monte-Carlo ensemble of simulated pressures in the WDN in response to uncertain (randomized) pipe roughness coefficients. Locations in the network that show the highest information gain to changes in pipe roughness are considered as optimal for pressure logger placement.