Research And Implementation On Burst Diagnosis Method Of Water Supply Network With Machine Learning

With the speeding up of urbanization,the demand for water resources is booming,people at the same time in the development process of water industry will inevitably encounter the pipe explosion because of pipeline aging,corrosion and third-party man-made destruction,resulting in waste of water resources and environmental damage,so it is particularly important to carry out the research on tube detection and diagnosis of water supply pipe network.In this study,the machine learning method is used as a means to study the following three aspects,centering on the real-time detection of water supply pipe network burst.(1)Optimal deployment method of water supply pipeline pressure gaugeAt present,the deployment of pressure gauge nodes at the bottom of water supply pipeline network is mainly based on the manual experience and the demand of big customers,and the lack of systematic scientific demonstration easily leads to the situation of node redundancy and insufficient coverage.To solve this problem,the method of clustering and then screening was adopted in this paper.At first,the K-means method is used to cluster the manometer.Whereafter,the index of node burst sensitivity is used as the screening basis,and the nodes with high tube burst sensitivity are selected as the deployment points of the manometer.The optimized pressure gauge node is used to monitor the overall running state of the water supply pipe network and provide data support for the diagnosis of pipe burst.(2)An evaluation method for the overall condition of water supply pipeline network based on the most unfavorable point of water pressure.It is the most common practice in water sector to speculate the overall situation of water supply of water supply pipe based on hydraulic pressure readings,but the current speculation is often according to subjective judgment of front-line workers,easily affected by many uncertain factors,so we need an objective mathematical model to estimate the degree of anomaly in the reading of the most unfavorable pressure point.In this study,a transfer learning framework was introduced,and the historical pressure data of different periods were used to form multiple source domains,and then the pressure indicator prediction model of the most unfavorable point of water pressure was established.In addition,the mathematical model proposed in this paper also considers the correlation between the number of different pressure gauge nodes in the whole network and carries on the modeling expression to enhance the accuracy of the model.Finally,by the comparison between the predicted value and the real value of the pressure at the most unfavorable point of water pressure,it can clearly indicate whether there is abnormal situation in the current water supply pipe network,which avoid the uncertainty caused by manual judgment.(3)Label-free online tube explosion positioning methodIn a real working scenario,the data of water supply network obtained based on the pressure meter and flow meter is difficult to contain the necessary data labels,which make regular supervision and learning methods cannot be directly used for tube positioning.To resolve the problem of such as number of wrong,first of all,this study based on the shortest path algorithm for each section to match the corresponding node pressure gauge,and pressure gauge in pipe section of pipeline network.Then,the local outlier factor algorithm is used to estimate the outlier degree of each pipe segment in real time through the matched manometer.Finally,the outlier degree of the pipe segment is converted to the probability of pipe burst,and the pipe segment with higher probability of pipe burst is identified as the pipe burst,which realize the real-time location of pipe burst in the water supply pipe network.