Study On The Prediction And Location Model Of Urban Water Supply Network Burst Based On Deep Learning

Bursting of water supply pipe network is a frequent phenomenon in the water supply industry today,characterized by strong suddenness and great harm,a wide range of impacts,and easy to cause adverse impacts on urban production,life,and environmental sanitation.With the continuous development of social economy and science and technology,there has been a significant breakthrough in the information technology of urban water supply networks.Using scientific methods to conduct realtime detection of leakage and explosion in the network is of great significance for saving water resources,ensuring the safe operation of urban water supply networks,and the development of social economy.This article obtains the basic data required for urban water supply network modeling through field research and measurement,builds a hydraulic model of urban water supply network based on EPANET technology,and checks the model parameters to ensure the accuracy of the model,laying a foundation for subsequent research such as pressure monitoring point optimization and pipe burst leakage location.To optimize the layout of pressure monitoring points,first calculate the impact coefficient of each node when the pipe bursts,then use fuzzy C-means clustering to partition the pipe network.After partitioning,select the points with the greatest impact in each small area as monitoring point groups,then use correlation coefficients to divide nodes with high similarity into the same area,and finally select the highest impact coefficient from the divided areas as pressure monitoring points.The remaining nodes in the area are monitored.Select 3-4 pressure measuring points in each zone,and add pressure measuring points at the end of the pipe network,the most unfavorable points,and other areas.Finally,arrange 21 pressure measuring points throughout the pipe network to achieve the optimal layout of pressure measuring points.The BP neural network and LSTM are used to perform a fitting prediction of water pressure fluctuations over a period of time.The prediction results show that the BP neural network is prone to large fluctuations in the prediction during the training set and is susceptible to sudden low pressure data,which affects the prediction of subsequent water pressures.However,the mean square error(RMSE)by the time of the test set is 0.0072714.Although the overall trend can be predicted correctly,there are still deviations for each sample point,The overall prediction is conservative,and the prediction effect for sudden rise in water pressure is not good.The mean square error(RMSE)of LSTM is 0.0068492,and the overall relative error is very small,mostly within 0.1.Moreover,the historical water pressure fluctuations during tube burst can be recorded through a memory gate,which can be used to optimize subsequent prediction parameters.The overall stability is very good,very close to the actual value,which allows for timely warning and control of abnormal conditions,reducing or avoiding pipe bursts in the water supply network due to sudden changes in water pressure from the source,thereby reducing the impact on the urban water supply network.The one-dimensional convolutional neural network is used to predict the pipe burst point in the pipe network,and the node multiple method is used to simulate the pipe burst flow.After the pipe network adjustment,the influence coefficient of water pressure change at each node is obtained.The influence coefficients of 21 pressure monitoring points arranged in Chapter 3 are used as the characteristic values of the pipe burst.The accuracy of the pipe burst positioning under different activation functions and different pooling functions is compared.The results show that when the activation function is Relu,when the pool layer is at its maximum,the maximum prediction accuracy of pipe burst is 49.5%,with an average error distance of 202.4 m.The prediction accuracy of the model for the positioning of pipe bursts in the entire pipeline network is less than that for small areas.The second area of the pipeline network has increased from 4 pressure measuring points to 7 pressure measuring points,and the prediction accuracy has increased from 51.2% to 64.2%,an increase of 13%.From 7pressure measuring points to 10 pressure measuring points,the prediction accuracy increased from 64.2% to 75.8%,an increase of 11.6%.The average error distance is201.6 meters,172.8 meters,and 138.6 meters,respectively.Finally,the relationship between the distance between the blasting point and the monitoring point,the diameter of the blasting pipe,and the blasting accuracy is analyzed.It is found that the closer the blasting point is to the pressure measuring point,the more accurate the prediction of the blasting position is,indicating that the pressure measuring point set on the basis of the blasting monitoring is more sensitive.When the pipe diameter at the pipe burst location rises from below 300 mm to 300-500 mm and above 500 mm,the prediction accuracy of the entire pipe network increases from 18.3%to 23.5% and 35.7%,and the prediction accuracy of the second zone of the pipe network increases from 58.7% to 63.8% and 79.2%.It can be seen that in pipe burst prediction,the larger the pipe diameter,the higher the average prediction accuracy.In summary,the in-depth learning water supply pipe network burst prediction and positioning model established in this article can be applied to daily pipe network burst management,with strong operability,which can help managers narrow the detection range of pipe burst accident points and reduce the detection difficulty.Figure [41] table [26] reference [96]...