CN-121993748-A - Gas pipeline leakage identification and positioning method and system based on machine learning

CN121993748ACN 121993748 ACN121993748 ACN 121993748ACN-121993748-A

Abstract

The invention relates to the technical field of gas pipeline leakage identification, and discloses a gas pipeline leakage identification and positioning method and system based on machine learning, wherein the method comprises the steps of acquiring pressure, flow and acoustic vibration time sequence data; generating a time-scale map, calculating topological structure characteristics, taking the topological structure characteristics, pressure time sequence characteristics and flow time sequence characteristics as multi-mode input, processing through a space-time diagram convolution network, outputting pipeline leakage identification probability and preliminary leakage position probability distribution, determining confidence coefficient of a likelihood function, updating through Bayesian reasoning to obtain posterior probability distribution of a leakage position, determining a fuzzy threshold, calculating and judging the asymmetric expected risk of a leakage state and a normal state, judging the current state as to-be-confirmed if the absolute value of the asymmetric expected risk difference value of the two states is lower than the fuzzy threshold, otherwise judging the current state as the leakage state or the normal state according to the expected risk minimum principle. The invention can improve the gas pipeline leakage recognition precision.

Inventors

ZHANG YINDI
LV YIBING
LI YIMO
WANG XIAOLONG
Lin Zhuocheng
Hao Haiyu

Assignees

长江大学

Dates

Publication Date: 20260508
Application Date: 20260323

Claims (10)

1. The gas pipeline leakage identification and positioning method based on machine learning is characterized by comprising the following steps of acquiring pressure, flow and acoustic vibration time sequence data of a plurality of sensor nodes distributed along a gas pipeline, performing continuous wavelet transformation on the acoustic vibration time sequence data to generate a time-scale map, and calculating a continuous coherent Betti number sequence of the time-scale map to obtain topological structure characteristics; Constructing a pipeline network as a space-time diagram model, taking topological structure features, pressure time sequence features and flow time sequence features as multi-mode input, processing through a space-time diagram convolution network, and outputting pipeline leakage identification probability and preliminary leakage position probability distribution; Determining the confidence coefficient of a likelihood function of a negative pressure wave propagation physical model based on the shannon entropy, and combining the prior probability and the likelihood function, and obtaining posterior probability distribution of the leakage position through Bayesian reasoning update; And if the absolute value of the asymmetric expected risk difference value of the two states is lower than the fuzzy threshold, judging the current state as to be confirmed, otherwise, judging the current state as the leakage state or the normal state according to the expected risk minimum principle.
2. The machine learning based gas pipeline leakage identification and localization method of claim 1, wherein the performing continuous wavelet transform on the acoustic vibration time series data generates a time-scale map, and calculates a continuous coherent Betti number sequence of the time-scale map to obtain a topological structure feature, comprising: Performing continuous wavelet transformation on acoustic vibration time sequence data of each sensor node to obtain a time-scale map, performing graying and binarization processing on the time-scale map to generate a binary image, converting the binary image into a cubic complex shape, and calculating the number Betti of 0 th order by using a continuous coherent algorithm Sequence and number Betti of 1 th order The sequence is characteristic of the topology.
3. The machine learning based gas pipeline leakage identification and localization method of claim 1, wherein the constructing the pipeline network as a space-time diagram model and using the topology feature, the pressure time sequence feature and the flow time sequence feature as multi-modal inputs, the processing by the space-time diagram convolution network comprises: The method comprises the steps of defining each sensor node as a node of a space-time diagram model, defining a physical pipeline section connected with adjacent nodes as an edge of the space-time diagram model, constructing the space-time diagram model, splicing topological structure features, pressure time sequence features and flow time sequence features of each node in feature dimensions to form a multi-mode input feature matrix of the nodes, wherein the space-time diagram convolution network is formed by stacking space-time convolution blocks, and each space-time convolution block comprises a layer of diagram convolution network for extracting space features and a layer of gating circulation unit for extracting time features.
4. The machine learning based gas pipeline leakage identification and localization method of claim 1, wherein the shannon entropy is calculated by the following formula: wherein, the method comprises the steps of, Is the entropy of shannon, Is the probability of leakage occurring at the ith pipe segment.
5. The machine learning based gas pipeline leak identification and localization method of claim 4, wherein the determining the confidence of the likelihood function of the negative pressure wave propagation physical model based on shannon entropy comprises: Confidence of likelihood function Calculated by the following formula: ; Wherein, the For the average leakage information entropy reference value obtained based on historical data statistics, k is a preset positive parameter, Is shannon entropy.
6. The machine learning based gas pipeline leakage identification and localization method of claim 5, wherein the combining the prior probability and the likelihood function to obtain the posterior probability distribution of the leakage position through bayesian inference update comprises: For each possible leak location Based on the physical model of negative pressure wave propagation, calculating the theoretical time of reaching each sensor by the negative pressure wave, and comparing with the actual monitored reaching time to obtain likelihood function value ; Posterior probability distribution of leak location Calculated by the following formula: ; Wherein, the For the preliminary leak location probability distribution, Is the confidence of the likelihood function.
7. The machine learning based gas pipeline leak identification and localization method of claim 1, wherein the determining a fuzzy threshold based on the pipeline leak identification probability comprises: Fuzzy threshold Calculated by the following formula: ; Wherein, the Is a preset reference risk difference threshold value, Is a preset positive parameter, and the preset positive parameter is a preset positive parameter, A probability is identified for a pipe leak.
8. The machine learning based gas pipeline leak identification and localization method of claim 1, wherein the calculating the asymmetric expected risk determined as a leak state and a normal state, respectively, comprises: Setting the cost of missing report risk as The cost of false alarm risk is And (2) and ; Calculating an asymmetric expected risk of determining a leakage state by the following formula : ; Calculating the asymmetric expected risk determined to be in a normal state by the following formula : ; Wherein, the A probability is identified for a pipe leak.
9. The machine learning based gas pipeline leak identification and localization method of claim 2, wherein a time-scale map is obtained by performing a continuous wavelet transform on acoustic vibration timing data of each sensor node using Morlet's mother wavelet.
10. The gas pipeline leakage identifying and positioning system based on machine learning is characterized by comprising a memory and a processor, wherein the memory stores computer instructions, and the processor realizes the gas pipeline leakage identifying and positioning method based on machine learning according to any one of claims 1-9 when executing the computer instructions.

Description

Gas pipeline leakage identification and positioning method and system based on machine learning Technical Field The invention relates to the technical field of gas pipeline leakage identification, in particular to a gas pipeline leakage identification and positioning method and system based on machine learning. Background Gas is a clean energy source whose long distance transport relies on a vast network of pipes. However, gas leaks occur due to factors such as pipe corrosion, third party damage, or material aging. The leakage not only causes obvious economic loss and energy waste, but also is more likely to cause serious safety accidents such as fire and explosion, and forms serious threat to the lives, properties and ecological environment of the public. The existing pipeline leakage detection methods are mainly divided into a direct detection method and an indirect detection method. Although the direct detection method (such as manual inspection) has intuitive results, the method has the inherent defects of long inspection period, poor real-time performance, high labor cost, difficult realization of global coverage and the like. In contrast, indirect detection methods infer leakage by analyzing changes in fluid dynamic parameters such as pressure, flow, acoustic waves, etc. inside the pipeline, with better real-time and coverage potential. In an indirect detection method based on data analysis, a physical model method identifies a leakage-induced characteristic wave by establishing a hydrodynamic equation. However, the method has extremely high precision requirements on the pipeline model, is easily interfered by background noise generated by normal operations such as valve opening and closing, compressor starting and stopping, and the like, so that the method has weaker recognition capability on tiny leakage and relatively higher false alarm rate. The machine learning method improves the accuracy of recognition to a certain extent by learning the leakage pattern in the history data. Nevertheless, the existing machine learning methods still have a plurality of limitations that firstly, the existing machine learning methods depend on single signals such as pressure or flow, the complementary advantages of multi-mode information cannot be fully utilized, secondly, most models ignore the time-space correlation of signal propagation in a pipe network topological structure, global evolution process of leakage events is difficult to capture, furthermore, in a positioning link, the effect of a data driving model is easily influenced by unbalance of training samples, and finally, the traditional fixed threshold alarming strategy cannot adapt to variable operation conditions, and lacks assessment of asymmetric risks caused by false alarm and false alarm, and the intelligent level of decision is to be improved. Disclosure of Invention The invention provides a gas pipeline leakage identification and positioning method and system based on machine learning to solve the problems that in the prior art, the gas pipeline leakage identification precision is insufficient, and false alarm and missing alarm are easy to cause. In a first aspect, the machine learning-based gas pipeline leakage identification and positioning method of the present invention includes the following steps: The method comprises the steps of obtaining pressure, flow and acoustic vibration time sequence data of a plurality of sensor nodes distributed along a gas pipeline, carrying out continuous wavelet transformation on the acoustic vibration time sequence data to generate a time-scale map, and calculating a continuous coherent Betti number sequence of the time-scale map to obtain topological structure characteristics; Constructing a pipeline network as a space-time diagram model, taking topological structure features, pressure time sequence features and flow time sequence features as multi-mode input, processing through a space-time diagram convolution network, and outputting pipeline leakage identification probability and preliminary leakage position probability distribution; Determining the confidence coefficient of a likelihood function of a negative pressure wave propagation physical model based on the shannon entropy, and combining the prior probability and the likelihood function, and obtaining posterior probability distribution of the leakage position through Bayesian reasoning update; And if the absolute value of the asymmetric expected risk difference value of the two states is lower than the fuzzy threshold, judging the current state as to be confirmed, otherwise, judging the current state as the leakage state or the normal state according to the expected risk minimum principle. Preferably, the generating a time-scale map by performing continuous wavelet transformation on the acoustic vibration time sequence data, and calculating a continuous coherent Betti number sequence of the time-scale map, to obtain a topological structure feature, i