Recurring automated model calibration for dynamically adaptive water distribution networks
File(s)
Author(s)
Waldron, Alexander Joseph
Type
Thesis
Abstract
The ability to build and maintain accurate hydraulic models in the water industry is a challenge of increasing importance as the applications of these models are used for short-term operational management, as well as strategic infrastructure investment and repair decisions.
However, uncertainties are introduced into hydraulic models as water networks evolve over time, both in terms of physical degradation and changing consumer demand within urban environments. The process of calibrating and validating the hydraulic model can bring back confidence to the end user, but it is traditionally only undertaken on an ad hoc basis as it is a labour intensive and expensive exercise with frequently questionable results.
This thesis utilises the ever-increasing quantity and quality of hydraulic data to develop robust and efficient model fitting methods for the purpose of hydraulic model calibration. Furthermore, data from long-term telemetry systems is used, together with advanced control, to study automatic and recurrent model validation for the continuous maintenance of dynamically adaptive networks.
Methods for model calibration have been extensively discussed in previously published literature. However, there have been key limitations, which were exacerbated by using theoretical water networks with fictitious data. These case studies often disregard the constraints of an operational model from the water industry, hence an underlying principle of this work is to implement advanced calibration procedures on operational networks and investigate the level of accuracy that can be achieved. Investigations into reducing the ill-posedness of the problem and its scalability are also undertaken.
Passive and active data sampling approaches are developed for continuously maintaining and improving the accuracy of the hydraulic model. Passive data sampling approaches use machine learning techniques to sample hydraulic data streams over an extended period, utilising natural variations and changes in control to improve the prediction accuracy of the hydraulic model. Active data sampling approaches involve optimally modifying hydraulic conditions via remotely actuated valves within dynamically adaptive networks for the purpose of improving the prediction accuracy of the hydraulic model.
Three large case studies that provide hydraulic data with high temporal and spatial resolution are used as unique test areas for implementing the methods presented in the thesis.
Using these methods, water companies can now recurrently validate and maintain their models, and as data and control become more ubiquitous, the process of automatic recurrent model validation will be further enhanced.
However, uncertainties are introduced into hydraulic models as water networks evolve over time, both in terms of physical degradation and changing consumer demand within urban environments. The process of calibrating and validating the hydraulic model can bring back confidence to the end user, but it is traditionally only undertaken on an ad hoc basis as it is a labour intensive and expensive exercise with frequently questionable results.
This thesis utilises the ever-increasing quantity and quality of hydraulic data to develop robust and efficient model fitting methods for the purpose of hydraulic model calibration. Furthermore, data from long-term telemetry systems is used, together with advanced control, to study automatic and recurrent model validation for the continuous maintenance of dynamically adaptive networks.
Methods for model calibration have been extensively discussed in previously published literature. However, there have been key limitations, which were exacerbated by using theoretical water networks with fictitious data. These case studies often disregard the constraints of an operational model from the water industry, hence an underlying principle of this work is to implement advanced calibration procedures on operational networks and investigate the level of accuracy that can be achieved. Investigations into reducing the ill-posedness of the problem and its scalability are also undertaken.
Passive and active data sampling approaches are developed for continuously maintaining and improving the accuracy of the hydraulic model. Passive data sampling approaches use machine learning techniques to sample hydraulic data streams over an extended period, utilising natural variations and changes in control to improve the prediction accuracy of the hydraulic model. Active data sampling approaches involve optimally modifying hydraulic conditions via remotely actuated valves within dynamically adaptive networks for the purpose of improving the prediction accuracy of the hydraulic model.
Three large case studies that provide hydraulic data with high temporal and spatial resolution are used as unique test areas for implementing the methods presented in the thesis.
Using these methods, water companies can now recurrently validate and maintain their models, and as data and control become more ubiquitous, the process of automatic recurrent model validation will be further enhanced.
Version
Open Access
Date Issued
2021-12
Date Awarded
2022-06
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Stoianov, Ivan
Sponsor
Engineering and Physical Sciences Research Council (EPSRC)
Anglian Water
Grant Number
EP/L016826/1
Publisher Department
Civil and Environmental Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)