Integrated optimal pressure sensor placement and localisation of leak/burst events using interpolation and a genetic algorithm

Leak/burst events are a serious problem because they disturb customer supplies, lead to water loss and managing them consumes vast resources. Water companies are continually seeking solutions to improve the situation. Presented in this thesis is the development, verification and validation of an integrated framework of methods for determining the optimal configurations of pressure sensors in a DMA and for localising new leak/burst events using a data-driven leak/burst localisation technique. The integration of the leak/burst localisation technique with the sensor placement technique is a novel feature of this framework of methods. A data-driven leak/burst localisation technique, featuring a novel spatially constrained inverse-distance weighted interpolation technique, was developed which quantifies the change in pressure due to a new leak/burst event using pressure sensors deployed in a DMA, without using a hydraulic model. The leak/burst localisation technique combines data from multiple pressure sensors to localise a leak/burst event by interpolating using the distance travelled along pipes. The leak/burst localisation technique was combined with the GALAXY multi-objective evolutionary algorithm to identify the optimal sensor configurations and parameters for the leak/burst localisation technique efficiently. The sensor placement technique automatically determines the leak/burst event sizes for each DMA and groups them to minimise the number of leak/burst event scenarios which are considered. The framework of methods was developed and verified iteratively using data from hydraulic models and a real DMA and validated using data from 20 engineered events conducted in two real DMAs in the UK. During validation, the sensor placement technique identified the optimal sensor configurations from a constrained subset of hydrants in each DMA. The agreement between the leak/burst localisation performance for the real and modelled engineered events demonstrated that the sensor placement technique can accurately predict the expected level of performance which will be achieved in a real DMA, particularly as the number of optimal sensors increases. Engineered events as small as 3.5% of the peak daily flow (6% of the average daily flow) were correctly localised with search areas containing as few as 12% of the pipes in a DMA.