Keywords
Data mining; Pressure sensor; Decision making; Resilience and preparedness; Water distribution networks
Start Date
5-7-2022 12:00 PM
End Date
8-7-2022 9:59 AM
Abstract
Extracting useful information from sensors, which record water distribution network (WDN) data, is essential to improve network performance, increasing the system preparedness and resilience, advancing towards network digitalisation. Due to the large volume of data generated, analysis of pressure head requires advanced techniques to reduce dimensionality. While previous works were typically based on comparing hydraulic simulations and observed data, there is a lack of study on pattern recognition, a helpful method for event detection, localisation, and prevention. Since the number of metering devices and their operativity has a crucial role in recognition of key patterns, a spatial evaluation of system behaviour (with the focus on resilience) is conducted in this study with the uses of pressure head data from sensors and complemented by simulated data where there is the absence of available sensor data. Comparing the heatmaps leads to extracting key patterns (i.e. landmarks), which will be helpful for decision-makers to increase the preparedness by making arrangements against critical events and allow classification and prediction of the system behaviour. This paper focuses on recognition of the possible landmarks in the network representing a key feature (particularly pressure) in the presence and absence of leakage through spatial analysis with the objective of dimensionality reduction. To do that, this paper explores a dataset of incidents, leakage/burst events, and ordinary network operations captured through sensors and expert knowledge in a WDN in Spain in order to obtain relevant information (in form of landmarks) from them. Hydraulic simulations for the whole network, have been also conducted to identify landmarks that allow both to complement and to validate their relevance obtained through the sensor data. Results were promising, recognising the patterns that allow to characterise the system behaviour when influenced by leakage/burst event.
Exploring a Spatial Dynamic Approach and Landmark Detection for Leakage/Burst Event Characterisation in Water Distribution Networks
Extracting useful information from sensors, which record water distribution network (WDN) data, is essential to improve network performance, increasing the system preparedness and resilience, advancing towards network digitalisation. Due to the large volume of data generated, analysis of pressure head requires advanced techniques to reduce dimensionality. While previous works were typically based on comparing hydraulic simulations and observed data, there is a lack of study on pattern recognition, a helpful method for event detection, localisation, and prevention. Since the number of metering devices and their operativity has a crucial role in recognition of key patterns, a spatial evaluation of system behaviour (with the focus on resilience) is conducted in this study with the uses of pressure head data from sensors and complemented by simulated data where there is the absence of available sensor data. Comparing the heatmaps leads to extracting key patterns (i.e. landmarks), which will be helpful for decision-makers to increase the preparedness by making arrangements against critical events and allow classification and prediction of the system behaviour. This paper focuses on recognition of the possible landmarks in the network representing a key feature (particularly pressure) in the presence and absence of leakage through spatial analysis with the objective of dimensionality reduction. To do that, this paper explores a dataset of incidents, leakage/burst events, and ordinary network operations captured through sensors and expert knowledge in a WDN in Spain in order to obtain relevant information (in form of landmarks) from them. Hydraulic simulations for the whole network, have been also conducted to identify landmarks that allow both to complement and to validate their relevance obtained through the sensor data. Results were promising, recognising the patterns that allow to characterise the system behaviour when influenced by leakage/burst event.
Stream and Session
false