Recurring Automated Model Calibration for Dynamically Adaptive Water Distribution Networks
Started: October 2017
Supervisor: Dr Ivan Stoianov
Industry Partner: Anglian Water
Description of Research
Technological advances have made it possible and affordable to use significant numbers of near real-time pressure and flow monitoring devices in large scale Water Distribution Networks (WDNs). In addition, self-powered automatic control valves can be used in conjunction with pressure monitoring to create highly controllable and dynamic water networks. These networks have improved resilience with the ability to respond to ‘events’ and change of demand within the network. There is also the capability for leakage detection and quantifying accurate water consumption.
Building and maintaining a model of a WDN is difficult as they are continuously being stressed in a variety ways. An accurate model requires constant updating and adjustment in order for it to be adequate for use in industry.
Calibration of WDNs has large areas of ambiguity. The requirements for when a WDN model is calibrated are not well defined and don’t differentiate much for varying applications (e.g. for control purposes). On top of this, WDN model calibration is done on an ad hoc basis at highly irregular periods by expensive external consultants. This means WDN models are not maintained sufficiently to be useful for operational purposes.
With the aforementioned technologies in place, there is a significant amount of data for accurate initial calibration of WDN models. Automatic recurring model calibration can then be performed periodically to ensure accurate near real-time hydraulic simulations for control purposes.
Anglian Water’s WDNs offer a good opportunity to develop some of these issues in real world case studies. Using the advances in WDN control, formulation of calibration performance requirements for operational purposes can be established. Pragmatic solutions for initial and recurring calibration can also be matured, and ultimately the research will show whether WDN models can be reliable for continuous long-term use.