Research On Early Warning And Location Of Burst In Water Supply Network Based On Deep Learning Method

The safety and stability of the water supply network is related to the water supply safety of the city,and the occurrence of burst events will destroy the stability of the water supply system,cause waste of water resources,and even cause secondary pollution of the network.Therefore,it is very important to carry out the early warning and location research of burst events in the water supply network to ensure the safety of water supply.In the previous research,the research on the placement method of the monitoring sensors and the research on the early warning and location of the burst have not been effectively unified.This paper unified them,proposed a pressure sensors layout method for burst early warning and location,and designed a model architecture for burst warning and location for deep learning methods.In the research of pressure sensor placement method,the Structured Deep Clustering Network based on graph neural network is used to fuse the topological features and multi-hydraulic features of the network and monitoring partitions are divided.In each monitoring partition,the most sensitive node is selected as the monitoring point using the indicator tensor defined according to the burst pipe characteristics.This method makes up for the lack of topological characteristics of the network in the previous research on the pressure sensor placement method and the shortcomings of the multi-hydraulic network that is not suitable.Then this method is compared with the traditional method in the experimental water supply network.In order to link with the burst location,the same clustering method is used to divide a larger number of location partitions,then,location partitions and the arranged pressure sensors are used as the monitoring system for the early warning and location research of the burst.In the research on early warning and location of burst,this paper designs a model architecture suitable for early warning and location of burst based on the deep learning,and the performance of three typical instantiation models are analyzed in the experimental water supply network.Considering flow monitoring data and the interconnection of monitoring sensors,this paper redesigns the model architecture with flow data and topology features of sensors,and compares the performance of the before and after improvement model in the experimental water supply network.The influence of the number of pressure sensors and the level of burst on the performance of the model is also discussed.The experimental results show that the Structured Deep Clustering Network can effectively integrate the topological features and multi-hydraulic features of the network.The distribution of pressure sensors arranged by the proposed method in this paper is more uniform and the coverage rate is higher.Three deep learning models can effectively carry out early warning and location of burst.Three models with the improved model architecture have a higher accuracy of burst warning and location,a lower false alarm rate of burst,and a shorter response time to burst events.The monitoring sensor placement method and deep learning model for burst early warning and location can meet the needs of real-time working.Them provide powerful tools and methods for the construction of smart water affairs for water companies.