CN-121977177-A - Pipe network leakage three-dimensional positioning method and system based on physical constraint loss

CN121977177ACN 121977177 ACN121977177 ACN 121977177ACN-121977177-A

Abstract

The invention discloses a three-dimensional positioning method and a three-dimensional positioning system for pipe network leakage based on physical constraint loss, which relate to the technical field of pipe network leakage monitoring and comprise the steps of obtaining multi-source heterogeneous data and historical tag data; the method comprises the steps of encoding multi-source heterogeneous data, extracting features, obtaining multi-source fusion feature vectors, inputting the multi-source fusion feature vectors into a multi-source fusion prediction model, outputting pipe network prediction results, constructing physical constraint losses including topological physical constraints, hydraulic physical constraints and buried physical constraints, constructing a total loss function based on data driving losses and physical constraint losses, performing model training until the model converges, and predicting based on the multi-source fusion prediction model by utilizing real-time fusion feature vectors.

Inventors

LIU HUIZHI
KANG QIAN
PENG FAN
WANG MENGKANG
LUO XIAOLAN
LONG JIAO
Huang Yuzhuo
WANG ZILONG
ZOU YE
HUANG CHANG

Assignees

湖南省交通科学研究院有限公司

Dates

Publication Date: 20260505
Application Date: 20260403

Claims (9)

1. The three-dimensional pipe network leakage positioning method based on physical constraint loss is characterized by comprising the following steps of: S10, constructing a multisource fusion prediction model based on a training set, wherein the method specifically comprises the following steps: s11, acquiring multi-source heterogeneous data and historical tag data, wherein the multi-source heterogeneous data comprises pipe network static data and pipe network dynamic data, and the pipe network static data comprises pipe network topological structure data; S12, coding and feature extraction are carried out on the multi-source heterogeneous data based on a preset short-time window, a dynamic feature coding vector, a static feature coding vector, a topological feature coding vector and a leakage feature coding vector are obtained, and the four types of feature coding vectors are spliced into a multi-source fusion feature vector; S13, inputting the multi-source fusion feature vector into a multi-source fusion prediction model, outputting a pipe network prediction result, wherein the pipe network prediction result comprises a prediction leakage result, a prediction leakage grade, a prediction leakage pipeline ID and a prediction leakage three-dimensional position, the multi-source fusion prediction model comprises an SE module, a CBAM module and a transform module, the SE module carries out channel dimension attention weighting on the multi-source fusion feature vector and outputs a weighted optimization feature vector, the CBAM module takes the weighted optimization feature vector as input, carries out dual enhancement of space and channels, outputs a space enhancement feature vector, and the transform module outputs the pipe network prediction result according to the space enhancement feature vector; S14, constructing physical constraint loss including topological physical constraint, hydraulic physical constraint and burial depth physical constraint; S15, constructing a total loss function based on data driving loss and the physical constraint loss, wherein the data driving loss is acquired based on supervision tag data and the pipe network prediction result, and back propagation training SE weight parameters, CBAM weight parameters and transducer weight parameters are carried out based on the total loss function until a model converges; S20, acquiring pipeline real-time data with preset duration, performing data coding and standardization processing on the pipeline real-time data based on the preset short-time window to acquire a real-time fusion feature vector, and predicting based on the multi-source fusion prediction model by utilizing the real-time fusion feature vector to acquire the current leakage state, the current leakage level, the leakage pipeline ID and the current leakage three-dimensional position of the pipeline network.
2. The three-dimensional positioning method for pipe network leakage loss based on physical constraint loss is characterized in that step S12 specifically comprises the steps of carrying out time sequence slicing on pipe network dynamic data based on a preset short-time window to obtain a short-time window physical quantity sequence, carrying out numerical coding processing on the short-time window physical quantity sequence and pipe network static data to respectively obtain a dynamic feature coding vector, a static feature coding vector, a topological feature coding vector and a leakage feature coding vector, carrying out dimension alignment on the four types of feature coding vectors, and then splicing to obtain a multi-source fusion feature vector corresponding to the preset short-time window.
3. The three-dimensional positioning method for pipe network leakage loss based on physical constraint loss according to claim 2, wherein the pipe network dynamic data comprises historical pressure data, historical flow data, differential pressure data, flow fluctuation data and pressure gradient data, and the pipe network dynamic data is segmented and processed by the short-time window; the pipe network static data comprise pipe network topological structures, node coordinates, pipe diameters, materials, pipeline burial depths, laying years and connection modes.
4. A three-dimensional positioning method for pipe network leakage based on physical constraint loss according to any one of claims 1 to 3, wherein the topological physical constraint is used for constraining the distance between the predicted leakage three-dimensional position and the three-dimensional point position of the pipe network topology to be no more than a planned deviation distance; The hydraulic physical constraint is used for constraining the hydraulic pressure change to conform to Bernoulli equation and the flow to conform to continuity equation; and the burial depth physical constraint is used for constraining the leakage depth in the predicted leakage three-dimensional position to be in an actual burial depth section of the pipe network.
5. The three-dimensional positioning method for pipe network leakage loss based on physical constraint loss according to claim 4, wherein the topological violation distance loss in topological physical constraint The calculation mode of (a) is that Where N is the number of predicted samples in the training set, For the shortest Euclidean distance between the predicted leakage three-dimensional position output by the model in the ith sample and the center line of the pipe section corresponding to the predicted leakage pipeline ID, The deviation distance is set for the allowed.
6. The three-dimensional positioning method for pipe network leakage loss based on physical constraint loss according to claim 4, wherein the hydraulic conservation deviation loss in hydraulic physical constraint is The calculation mode of (a) is that Where N is the number of predicted samples in the training set, 、 The pressure value and the flow value at the leakage point predicted by the model in the ith sample are respectively, 、 The theoretical pressure value and the theoretical flow value are respectively calculated for the leakage points based on the Bernoulli equation and the fluid continuity equation.
7. The three-dimensional positioning method for pipe network leakage loss based on physical constraint loss according to claim 4, wherein the penalty loss of the burial depth interval in burial depth physical constraint is The calculation mode of (a) is that Where N is the number of predicted samples in the training set, For the leak depth value in the predicted leak three-dimensional position output by the model in the ith sample, 、 The i-th sample corresponds to the actual minimum and maximum burial depths of the pipeline, Is the Z-axis coordinate of the center of the cross section of the pipeline, For the bottom leak preference weight, The value range is 0.3 to 0.75.
8. The three-dimensional positioning method for pipe network leakage loss based on physical constraint loss according to claim 4, wherein, , wherein, For the topology violation distance loss corresponding to the topology physical constraint, For the hydraulic conservation deviation loss corresponding to the hydraulic physical constraint, Penalty for the burial depth interval corresponding to the burial depth physical constraint, Respectively, are the self-adaptive weight coefficients, + + =1; The total loss function Wherein And driving the loss for the data.
9. Pipe network leakage accurate positioning system based on physical constraint loss, its characterized in that: Comprising a data acquisition module and a data acquisition module, the method comprises the steps of acquiring pipe network static data, pipe network dynamic data and historical tag data; the data coding and feature extraction module is used for carrying out time sequence slicing and numerical coding on the data based on a preset short-time window, extracting four types of feature coding vectors and splicing the four types of feature coding vectors into a multi-source fusion feature vector; The multisource fusion prediction model module comprises an SE module, a CBAM module and a transducer module, and is used for outputting leakage results, leakage grades, leakage pipeline IDs and three-dimensional positions; the physical constraint loss construction module is used for constructing topological physical constraint, hydraulic physical constraint and burial depth physical constraint and calculating physical constraint loss; and the model training module is used for training model weight parameters based on the total loss function until the model converges.

Description

Pipe network leakage three-dimensional positioning method and system based on physical constraint loss Technical Field The invention relates to the technical field of pipe network leakage monitoring, in particular to a pipe network leakage three-dimensional positioning method and system based on physical constraint loss. Background With the acceleration of the urban process and the aging of the water supply network system, the problems of network leakage of the urban network due to corrosion, damage, loose interfaces and the like are increasingly prominent, so that water resource waste is caused, secondary disasters can be possibly caused, and the accurate and efficient positioning of the network leakage points is a key link for the maintenance of the water supply network. At present, the pipe network leakage positioning technology mainly comprises a plurality of technical routes based on acoustic signal detection, hydraulic model analysis, artificial intelligence methods and the like. In the field of water supply network leakage positioning, the traditional method mainly relies on an acoustic signal detection technology to position, the method relies on acoustic signals generated by leakage to be easily influenced by factors such as pipe network burial depth, pipeline materials and the like, acoustic signals are easy to attenuate and distort in the propagation process, so that positioning accuracy is low, and anti-interference capability is weak, CN115127037B discloses a water supply network leakage positioning method and system. With the development of artificial intelligence technology, a deep learning method is widely applied to pipe network leakage detection, CN111881999A discloses a water service pipeline leakage detection method and system based on a deep convolutional neural network, the method collects pipeline data through intelligent pipes, establishes and trains a convolutional neural network model, acquires the pipeline data in real time and carries out pretreatment, finally judges the position of a leakage point through the index comparison of a model prediction result and a leakage data label, the method improves the accuracy of leakage detection to a certain extent, but has strong dependence on acoustic signals, the positioning accuracy can be obviously reduced under the condition of large pipeline burial depth or serious environmental noise interference, and the problem of fitting and generalized failure easily occurs in a model design driven by pure data. Most of the methods depend on acoustic signal detection excessively, acoustic signals are easy to attenuate and distort due to factors such as pipe network burial depth and pipeline materials, the positioning accuracy is low due to weak anti-interference capability, the pure data driving model lacks physical rationality constraint, the output leakage points are easy to appear outside the pipe network, jump in burial depth and pressure flow violate the hydraulic rule, and the method cannot be directly applied to engineering practice. In view of this, there is a need for a three-dimensional localization method and system for pipe network leakage that fuses physical priors and does not rely on acoustic signals. Disclosure of Invention The invention mainly aims to provide a three-dimensional positioning method and system for pipe network leakage based on physical constraint loss, and aims to solve the technical problems that in the prior art, the pipe network leakage is monitored accurately due to the fact that acoustic signals are relied on for pure data fitting in pipe network leakage monitoring, the acoustic signals are prone to environmental interference, and the pure data fitting model is difficult to converge and reasonably and physically explain. In order to achieve the above purpose, the invention provides a three-dimensional positioning method for pipe network leakage loss based on physical constraint loss, comprising the following steps: S10, constructing a multisource fusion prediction model based on a training set, wherein the method specifically comprises the following steps: S11, acquiring multi-source heterogeneous data and historical tag data, wherein the multi-source heterogeneous data comprises pipe network static data and pipe network dynamic data, and the pipe network static data comprises pipe network topological structure data; S12, coding and feature extraction are carried out on multi-source heterogeneous data based on a preset short-time window, a dynamic feature coding vector, a static feature coding vector, a topological feature coding vector and a leakage feature coding vector are obtained, and the four types of feature coding vectors are spliced into a multi-source fusion feature vector; S13, inputting the multisource fusion feature vector into a multisource fusion prediction model, outputting a pipe network prediction result, wherein the pipe network prediction result comprises a prediction leakage result, a pred