Propagating Morphology Of Pressure Fluctuations And Detection Methods For Bursts In Water Distribution Networks

The pipe burst in the water distribution network is a compelling issue affecting the security of the water supply in China.A large amount of water is quickly lost when a pipe burst occurs,causing economic losses and resource waste.Some pipe bursts can even cause massive water cut-offs,traffic jams,and road collapses,which endanger public health and properties.Nowadays,water companies in China still rely mainly on user complaints and manual inspection,which are inefficient and have limited detection capacity for pipe bursts that occur at night,or in remote areas.When pipe bursts occur,the hydraulic state of the water distribution network is influenced and monitored data are changed.By analyzing the change of measured data,pipe bursts can be detected and localized timely,which enables water companies to repair broken pipes quickly and mitigate damages.At present,many scholars have carried out research on burst detection and localization.But the existing research still has some shortcomings,mainly including(1)a lack of in-depth understanding of measured data during bursts.Due to the lack of analysis of the propagation of pressure fluctuations during bursts,data extraction and identification methods are selected blindly in some studies;(2)the steady-state analysis and the transient analysis are separated.Pipe bursts can affect both transient and steady-state data.However,existing research independently analyzes one kind of data,causing limited utilization of information;(3)relying on flow data and ignoring pressure data.Most of the existing burst detection methods rely on the import and export flow data of the district metering area in the water distribution network.But flow sensors and the construction of district metering areas are relatively expensive.(4)limitations of application to large-scale and complex water distribution networks.In large-scale networks,the water consumption pattern is complex and pressure fluctuations caused by pipe bursts are difficult to identify.Few studies have been successfully verified in large-scale and complex water distribution networks.To solve the problems above,the characteristics of pressure fluctuations and methods for burst detection and localization in water distribution networks are studied in this paper.The propagation of pressure fluctuations during the pipe burst is analyzed and the temporal and spatial characteristics of pressure are analyzed.On this basis,a method that extracts pressure disturbances caused by bursts in water distribution networks is proposed.This method enables real-time and accurate burst detection.Then,the characteristics of the steady-state and transient pressure data were analyzed collaboratively.By combining dimension reduction and feature matching methods,bursts can be finally pinpointed.The changes of hydraulic state in pipe networks caused by bursts were divided into three stages,including sudden impact,oscillation,and continuous impact.The periodicity and frequency domain characteristics of the steady-state data and temporal and spatial features of transient-state data during bursts were studied in this study.The influence of factors(topology,burst size,etc.)on measured data was discussed.The propagation of pressure fluctuations during pipe bursts was comprehensively analyzed.To detect bursts accurately,a disturbance extraction method is proposed based on frequency domain characteristics of steady-state measured data.Through the time-frequency domain analysis of pressure data,disturbances in the steady-state data are extracted and identified.The method proposed has more robustness compared to existing prediction-classification methods that can easily be affected by errors in historical data and inaccurate prediction.The isolation forest algorithm is then used to analyze disturbances and identify whether a pipe burst occurs.The real-time burst detection method was verified in a large and complex real-life pipe network with a water supply area of 302km~2 and only 23 sensors.The proposed method can successfully detect 80%of bursts that occur in pipes with a diameter of more than800mm.Besides,there is a more than 20%probability of detecting small bursts with pressure drops of less than 1m at sensors.The method also detected a real pipe burst 7hours earlier than the water company found.In addition,this method only utilizes the pressure data and does not depend on district metering area and flow meters,and has strong applicability in practical engineering.To pinpoint pipe bursts,the K-means clustering is firstly used to divide the pipe network into small burst searching areas.Then the support vector classification is used to preliminarily locate the burst and narrow the search range down to a burst searching area.In this burst searching area,principal component analysis and support vector classification are used to extract and match transient data features,which reduces the difficulty to deal with a large amount of transient data and identify features.The precise location of bursts can be identified by the burst localization framework proposed in this study based on the collaborative analysis of steady-state and transient data.Case studies showed that 95.67%of the bursts can be successfully located within 500m in a burst searching area by using the burst localization framework.This study provides a complete solution system for burst detection and localization in water distribution networks.It also provides theoretical and technical support for burst management and intelligent control of the pipe network.It is of great significance to strengthen the network resilience and ensure the safety of the water supply in China.