CN113111923A - Water supply network leakage detection and positioning method based on one-dimensional migration learning convolutional neural network integrated model - Google Patents

Water supply network leakage detection and positioning method based on one-dimensional migration learning convolutional neural network integrated model Download PDF

Info

Publication number
CN113111923A
CN113111923A CN202110308422.5A CN202110308422A CN113111923A CN 113111923 A CN113111923 A CN 113111923A CN 202110308422 A CN202110308422 A CN 202110308422A CN 113111923 A CN113111923 A CN 113111923A
Authority
CN
China
Prior art keywords
model
data
neural network
target domain
dimensional
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110308422.5A
Other languages
Chinese (zh)
Inventor
周猛飞
徐银泽
杨彦辉
胡寅朝
郭添
蔡亦军
潘海天
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN202110308422.5A priority Critical patent/CN113111923A/en
Publication of CN113111923A publication Critical patent/CN113111923A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply

Abstract

The invention discloses a water supply network leakage detection and positioning method based on a one-dimensional migration learning convolutional neural network integrated model, which comprises the steps of firstly, pre-training a plurality of source tasks 1DCNN with different parameters by using source domain data, and enabling the source tasks 1DCNN to achieve excellent classification performance on the source domain data. Secondly, the migration source domain pre-trained 1DCNN achieves model training of the target domain by fine tuning parameters with a small learning rate, and a plurality of migration-learned one-dimensional convolutional neural networks (TL1DCNN) are obtained. And then, optimizing and obtaining the weight coefficient of each TL1DCNN by using a particle swarm optimization algorithm, and finally obtaining a one-dimensional migration learning convolution neural network integrated model. The method provided by the invention has better performance in indexes such as classification accuracy, recall rate, F _ score, confusion matrix and the like. The comparison result shows that the method provided by the invention has better performance in the aspects of accuracy and robustness.

Description

Water supply network leakage detection and positioning method based on one-dimensional migration learning convolutional neural network integrated model
Technical Field
The invention relates to the technical field of water supply network leakage detection and positioning, in particular to a water supply network leakage detection method based on an integrated one-dimensional migration learning convolutional neural network.
Background
The pipeline transportation is widely applied to occasions such as oil and gas transportation, urban tap water networks and the like due to the characteristics of large transportation capacity, safety, reliability, strong continuity and the like. Due to long-term use, pipelines can leak due to corrosion, aging, construction damage and other reasons, and huge environmental and economic losses are caused. Therefore, it is important to detect and identify potential anomalies and failures of the pipeline as early as possible and implement fault tolerant operations to minimize performance degradation and avoid hazardous situations.
Pipeline leak detection techniques have been widely studied, and the associated algorithms are numerous. There can be generally divided into hardware detection and software detection methods. The hardware detection method usually depends on hardware equipment, such as a hydrophone, an optical fiber and an acoustic method, and the hardware detection method generally needs to install expensive equipment, so that the hardware detection method is very expensive when being used in a large scale. The software detection method relies on pressure, flow and other measurement variables, and leakage detection and positioning are carried out by constructing a mathematical mechanism model or a machine learning method.
Although conventional software-based detection methods can address the problem of large numbers of pipe leaks and locations, conventional methods have many limitations. First, manual feature extraction requires strong expertise and experience. Second, only shallow features can be extracted, often hard to work with complex systems. These deficiencies limit the classification or regression performance of the model. In recent years, deep learning techniques have made breakthroughs in image processing, natural language processing, fault diagnosis, and the like. Deep learning can automatically extract deep features from raw data without relying on prior knowledge, the basic components of which are typically convolutional layers, pooling layers, and fully-connected layers. The early deep learning technology applied to fault detection is mostly two-dimensional convolutional neural network (2DCNN), the method needs to convert original one-dimensional signal data into a two-dimensional picture form, and compared with the one-dimensional convolutional neural network (1DCNN), some useful information may be lost and calculation consumption is increased in the conversion process. The 1DCNN model using raw data directly reduces complexity and saves computational overhead. The deep learning algorithm is based on the assumption that the training data and the data to be processed are in the same feature space and have the same distribution and a large amount of labeled data is required, but in practical applications, these assumptions often cannot be satisfied. To solve this problem, the migration learning is beginning to be applied to the field of failure diagnosis where it is difficult to obtain a large amount of data.
For a task with a small amount of data and a large number of feature dimensions, even transfer learning may not achieve a prominent effect. The integrated learning combined with a plurality of learners can generally reduce the risk of poor generalization performance and reduce the risk of trapping the learning algorithm into local minimum points. Therefore, the integration method has a satisfactory effect in many fields, and the combination of a plurality of individual learners with data, parameters and structure differences can improve the performance of the model.
Disclosure of Invention
In order to realize the purpose of more accurately extracting the characteristics of pipeline leakage from a small amount of data to realize accurate diagnosis, the invention provides a pipeline leakage detection and positioning method which is based on an integrated one-dimensional migration convolutional neural network as a base learner and integrates the results of a plurality of base learners for classification. The method effectively relieves the limitation that fault data are difficult to obtain, and improves the accuracy of the model.
A water supply network leakage detection and positioning method based on a one-dimensional migration learning convolutional neural network integrated model comprises the following steps:
step 1, acquiring N groups of pressure signal data sets of a source domain water supply network, wherein part of data is used as a source domain training set, and the other part of data is used as a source domain verification set;
acquiring M groups of target domain water supply network pressure signal data sets as target domain data sets, wherein M is smaller than M
Figure BDA0002988556160000021
A first part of data in the target domain data set is used as a target domain training set, a second part of data is used as a target domain verification set, and a third part of data is used as a target domain test set;
in the step 1, N groups of pressure signal data sets of a source area water supply network are acquired, wherein 65% -85% of data serve as a source area training set, and 15% -35% of data serve as a source area verification set; preferably, N groups of pressure signal data sets of the source-domain water supply network are acquired, wherein 75% of data are used as a source-domain training set, and 25% of data are used as a source-domain verification set;
taking 54% -74% of data in the target domain data set as a target domain training set, taking 6% -17% of data as a target domain verification set, and taking 15% -35% of data as a target domain test set; most preferably, 64% of the data in the target domain data set is used as the target domain training set, 11% of the data is used as the target domain validation set, and 25% of the data is used as the target domain test set.
M is less than
Figure BDA0002988556160000022
Step 2: inputting the source domain training set data of the step 1 into a one-dimensional convolutional neural network for source domain convolutional neural network model training, wherein the source domain verification set data of the step 1 is used for verifying the quality of the model; the model with the minimum cross entropy is used as a source domain one-dimensional convolution neural network model;
in step 2, the definition of cross entropy is expressed as:
Figure BDA0002988556160000023
wherein E represents a crossCross entropy, n represents a category index, m represents a sample index, y represents a sample label,
Figure BDA0002988556160000031
represents a predictive label;
and step 3: loading the source domain one-dimensional convolution neural network model in the step (2), setting the learning rate of the target domain one-dimensional convolution neural network as one tenth of the learning rate of the source domain convolution neural network model, using the parameters of the target domain training set data training model in the step (1), and using the target domain verification set data to verify the model;
and 4, step 4: repeating the step 2 and the step 3 for Y times to obtain Y migration learning one-dimensional convolution neural network models (TL1DCNN), calculating the similarity among the models and obtaining a similarity matrix, wherein all combination strategies of the submodels share the same
Figure BDA0002988556160000032
Respectively from
Figure BDA0002988556160000033
Obtaining a Y-1 combination strategy according to the principle that the sum of the similarity is minimum, wherein C represents a combination symbol; the calculation of the similarity S is expressed as:
Figure BDA0002988556160000034
Figure BDA0002988556160000035
wherein a and b represent the a and b elements of the probability vector P, I represents the prediction label of TL1DCNN, Si,jRepresenting the similarity of ith and jth TL1DCNN predicted tags;
and 5: the misclassification number in the target domain verification set in the step 1 is taken as an optimization target, and the weight of the integrated sub-model in the Y-1 combination strategy in the step 4 is optimized by using a particle swarm optimization algorithm, as shown in a formula (4); evaluating the performance of each combination strategy model by using the data of the target domain test set in the step 1, wherein the model with the highest accuracy is the finally obtained one-dimensional migration learning convolution neural network integrated model, as shown in a formula (5):
Figure BDA0002988556160000036
Result＝argmax(V) (5)
wherein k represents the number of lattices of the integrated TL1DCNN model, wiAnd PiRespectively representing a weight coefficient corresponding to the ith TL1DCNN model and an output probability vector, V representing a probability vector obtained after TL1DCNN integration, and Result representing a prediction label obtained by a final one-dimensional migration learning convolutional neural network integration model.
Obtaining a plurality of TL1 DCNNs in the step 4; and 5, obtaining an integrated model of the TL1DCNN based on a particle swarm optimization algorithm, and realizing the leakage detection and positioning of the water supply network.
Compared with the prior art, the invention has the following advantages:
the invention discloses a water supply network leakage detection and positioning method based on a one-dimensional migration learning convolutional neural network integrated model. The method has the main advantages that the migration learning is used for the one-dimensional convolutional neural network, and the fault diagnosis of pipeline leakage detection and positioning is carried out by integrating a plurality of TL1DCNN base learners by using the weight optimized by the PSO. The method provided by the method is tested by using pipeline leakage data, and the accuracy, the recall rate, the F _ score and the like respectively reach 90.5%, 91.8% and 90.4%, which are superior to other methods. The result shows that the method has certain prospect in the field of leakage detection and positioning of the water supply pipe network
Drawings
FIG. 1 is a schematic diagram of a modeling method.
FIG. 2 is a layout of the experimental setup.
Fig. 3 is the signal collected (tag value 0 indicates no leakage, odd number is large leakage, even number is small leakage).
FIG. 4a is a confusion matrix for the integrated TL1 DCNN.
FIG. 4b is the confusion matrix for the same weight integrated TL1 DCNN.
FIG. 4c is the confusion matrix for TL1 DCNN.
Detailed Description
The following detailed description of the embodiments of the present invention will be provided with reference to the drawings and examples, so that how to apply the technical means to solve the technical problems and achieve the technical effects can be fully understood and implemented.
The schematic diagram of the model method of the present invention is shown in fig. 1. The experimental setup of the present invention is shown in FIG. 2. The collected signals (label value 0 indicates no leakage, odd number is large leakage, even number is small leakage) are shown in fig. 3. The confusion matrix for TL1DCNN integration is shown in FIG. 4 a. The confusion matrix for the same weight integrated TL1DCNN is shown in FIG. 4 b. The confusion matrix for TL1DCNN is shown in FIG. 4 c.
A water supply network leakage detection and positioning method based on a one-dimensional migration learning convolutional neural network integrated model comprises the following steps:
step 1, acquiring N groups of pressure signal data sets of a source domain water supply network, wherein 75% of data is used as a source domain training set, and 25% of data is used as a source domain verification set; acquiring M groups of target domain water supply network pressure signal data sets, wherein M is far smaller than N, 64% of data in the target domain data sets are used as target domain training sets, 11% of data are used as target domain verification sets, and 25% of data are used as target domain test sets;
step 2, inputting the source domain training set data of the step 1 into a one-dimensional convolutional neural network for source domain convolutional neural network model training, wherein the source domain verification set data of the step 1 is used for verifying the quality of the model; the model with the minimum cross entropy is used as a source domain one-dimensional convolution neural network model; the definition of cross entropy can be expressed as:
Figure BDA0002988556160000051
wherein E represents cross entropy, n represents class index, m represents sample index, y represents sample label,
Figure BDA0002988556160000052
represents a predictive label;
step 3, loading the source domain one-dimensional convolutional neural network model in the step 2, setting the learning rate of the target domain one-dimensional convolutional neural network as one tenth of the learning rate of the source domain convolutional neural network model, using the parameters of the target domain training set data training model in the step 1, and using the verification data set data of the target domain to verify the model;
step 4, repeating the step 2 and the step 3 for Y times to obtain Y migration learning one-dimensional convolution neural network models (TL1DCNN), calculating the similarity among the models and obtaining a similarity matrix, wherein all combination strategies of the submodels are shared
Figure BDA0002988556160000053
Respectively from
Figure BDA0002988556160000054
Obtaining a Y-1 combination strategy according to the principle of minimum sum of similarity; the calculation of the similarity S can be expressed as:
Figure BDA0002988556160000055
Figure BDA0002988556160000056
where a and b represent the a and b elements of the probability vector P, I represents the predictive label of TL1DCNN, Si,jrepresenting the similarity of ith and jth TL1DCNN predicted tags;
step 5, with the number of misclassifications in the target domain verification set in the step 1 as an optimization target, optimizing the weight of the integrated sub-models in the Y-1 combination strategy in the step 4 by using a particle swarm optimization algorithm, as shown in a formula (4); evaluating the performance of each combination strategy model by using the data of the target domain test set in the step 1, wherein the model with the highest accuracy rate is the finally obtained one-dimensional migration learning convolution neural network integrated model, as shown in a formula (5);
Figure BDA0002988556160000057
result ═ argmax (v) (5) where k denotes the number of bins of the integrated TL1DCNN model, wiAnd PiRespectively representing a weight coefficient corresponding to the ith TL1DCNN model and an output probability vector, V representing a probability vector obtained after TL1DCNN integration, and Result representing a prediction label obtained by a final one-dimensional migration learning convolutional neural network integration model.
The invention discloses a method for detecting and positioning pipeline leakage by integrating and optimizing results of a plurality of base learners by taking a one-dimensional migration convolutional neural network (1DCNN) as the base learners. First, a plurality of source tasks 1DCNN of different parameters are pre-trained using source domain data, so that these 1DCNN achieve excellent classification performance on the source domain data. Secondly, the migration source domain pre-trained 1DCNN achieves model training of the target domain by fine tuning parameters with a small learning rate, and a plurality of migration-learned one-dimensional convolutional neural networks (TL1DCNN) are obtained. And then, optimizing and obtaining the weight coefficient of each TL1DCNN by using a particle swarm optimization algorithm, and finally obtaining a one-dimensional migration learning convolution neural network integrated model. The method provided by the invention has better performance in indexes such as classification accuracy, recall rate, F _ score, confusion matrix and the like. The comparison result shows that the method provided by the invention has better performance in the aspects of accuracy and robustness.
Negative pressure waves generated by leaks with different sizes at different positions are collected, and the specific operating conditions are as follows: the medium is water, the pipeline system comprises a main pipeline and two branch pipelines, the inner diameter of the main pipeline is 400mm, the length of the main pipeline is 1km, and the inner diameters and the lengths of the two branch pipelines are respectively 200mm, 250mm, 400m and 500 m. The roughness of the inner wall of the tube is 0.07mm, the propagation speed of negative pressure wave is 1000m/s, and the temperature is 20 ℃. To simulate leakage at different locations and different apertures, flow ball valves were randomly installed within a selected range of 900m upstream, with the aperture of the ball valve being between 3mm and 12 mm. The upstream and downstream pressure measuring points are PI101-PI104, the sampling frequency of the pressure gauge is 50HZ, and the time length is 10 s. The pressure signals collected from the 4 pressure measurement points are connected end to end as a set of samples.
Due to the practical scenario, it is not possible to obtain a large number of failure samples. Therefore, leakage faults can only be artificially produced within a small range in an actual pipe network, and a large amount of field data can be obtained with low damage. Thus, a large amount of data was collected for the range of tag values 1-6, totaling 3044 groups; a small amount of data was collected for the range of tag values 7-18, totaling 704 sets. The partitioning of the data set is shown in table 1.
TABLE 1 data set partitioning
Figure BDA0002988556160000061
First, 2358 sets of training set data of the source domain dataset were used to train the performance of 5 different parameters of the 1DCNN model, 786 sets of validation set data validation models. Second, the 717 sets of training set data for the target domain data set are used to fine tune the parameters of the source domain model to a higher degree of accuracy over the entire pipe region under study. And performing weighted summation on the classification probabilities output by the 5 TL1DCNN models to obtain the output of the integrated TL1DCNN model finally, namely the result of the method, wherein the weight value is determined by a particle swarm algorithm which sets the number of the samples of the verification set subjected to misclassification as an optimization objective function, the number of particles is 50, and the number of iteration rounds is 40.
Compared with TL1DCNN and an integrated TL1DCNN model with equal weight, the method has better performance, and detailed indexes are shown in a table 2.
The embodiments described in this specification are merely illustrative of implementations of the inventive concept and the scope of the present invention should not be considered limited to the specific forms set forth in the embodiments but rather by the equivalents thereof as may occur to those skilled in the art upon consideration of the present inventive concept.
TABLE 2 TL1DCNN, TL1DCNN and TL1DCNN effect table of the same weight integrated
Figure BDA0002988556160000071

Claims (5)

1. A water supply network leakage detection and positioning method based on a one-dimensional migration learning convolutional neural network integrated model comprises the following steps:
step 1: acquiring N groups of source domain water supply network pressure signal data sets, wherein part of data is used as a source domain training set, and the other part of data is used as a source domain verification set;
acquiring M groups of target domain water supply network pressure signal data sets as target domain data sets, wherein M is smaller than N, a first part of data in the target domain data sets is used as a target domain training set, a second part of data is used as a target domain verification set, and a third part of data is used as a target domain test set;
step 2: inputting the source domain training set data of the step 1 into a one-dimensional convolutional neural network for source domain convolutional neural network model training, wherein the source domain verification set data of the step 1 is used for verifying the quality of the model; the model with the minimum cross entropy is used as a source domain one-dimensional convolution neural network model;
and step 3: loading the source domain one-dimensional convolution neural network model in the step (2), setting the learning rate of the target domain one-dimensional convolution neural network as one tenth of the learning rate of the source domain convolution neural network model, using the parameters of the target domain training set data training model in the step (1), and using the target domain verification set data to verify the model;
and 4, step 4: repeating the step 2 and the step 3 for Y times to obtain Y migration learning one-dimensional convolution neural network models, namely TL1DCNN, calculating the similarity among the models and obtaining a similarity matrix, wherein all combination strategies of the submodels are common
Figure FDA0002988556150000011
Respectively from
Figure FDA0002988556150000012
Obtaining a Y-1 combination strategy according to the principle that the sum of the similarity is minimum, wherein C represents a combination symbol; the calculation of the similarity S is expressed as:
Figure FDA0002988556150000013
Figure FDA0002988556150000014
wherein a and b represent the a and b elements of the probability vector P, I represents the prediction label of TL1DCNN, Si，jRepresenting the similarity of ith and jth TL1DCNN predicted tags;
and 5: the misclassification number in the target domain verification set in the step 1 is taken as an optimization target, and the weight of the integrated sub-model in the Y-1 combination strategy in the step 4 is optimized by using a particle swarm optimization algorithm, as shown in a formula (4); evaluating the performance of each combination strategy model by using the data of the target domain test set in the step 1, wherein the model with the highest accuracy is the finally obtained one-dimensional migration learning convolution neural network integrated model, as shown in a formula (5):
Figure FDA0002988556150000021
Result＝argmax(V) (5)
wherein k represents the number of lattices of the integrated TL1DCNN model, wiAnd PiRespectively representing a weight coefficient corresponding to the ith TL1DCNN model and an output probability vector, V representing a probability vector obtained after TL1DCNN integration, and Result representing a prediction label obtained by a final one-dimensional migration learning convolutional neural network integration model.
2. The method for detecting and locating the water supply network leakage based on the one-dimensional migration learning convolutional neural network integrated model as claimed in claim 1, wherein in step 1, N sets of source domain water supply network pressure signal data sets are acquired, wherein 65% -85% of the data are used as a source domain training set, and 15% -35% of the data are used as a source domain verification set.
3. The method for detecting and locating the leakage of the water supply network based on the one-dimensional migration learning convolutional neural network integrated model as claimed in claim 1, wherein in step 1, M is less than
Figure FDA0002988556150000022
4. The method for detecting and positioning the leakage of the water supply network based on the one-dimensional migration learning convolutional neural network integrated model as claimed in claim 1, wherein in step 1, 54% -74% of data in the target domain data set is used as a target domain training set, 6% -17% of data is used as a target domain verification set, and 15% -35% of data is used as a target domain test set.
5. The method for detecting and positioning the leakage of the water supply network based on the one-dimensional migration learning convolutional neural network integrated model as claimed in claim 1, wherein in the step 2, the definition of the cross entropy is expressed as:
Figure FDA0002988556150000023
wherein E represents cross entropy, n represents class index, m represents sample index, y represents sample label,
Figure FDA0002988556150000024
representing a predictive label.
CN202110308422.5A 2021-03-23 2021-03-23 Water supply network leakage detection and positioning method based on one-dimensional migration learning convolutional neural network integrated model Pending CN113111923A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110308422.5A CN113111923A (en) 2021-03-23 2021-03-23 Water supply network leakage detection and positioning method based on one-dimensional migration learning convolutional neural network integrated model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110308422.5A CN113111923A (en) 2021-03-23 2021-03-23 Water supply network leakage detection and positioning method based on one-dimensional migration learning convolutional neural network integrated model

Publications (1)

Publication Number Publication Date
CN113111923A true CN113111923A (en) 2021-07-13

Family

ID=76710378

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110308422.5A Pending CN113111923A (en) 2021-03-23 2021-03-23 Water supply network leakage detection and positioning method based on one-dimensional migration learning convolutional neural network integrated model

Country Status (1)

Country Link
CN (1) CN113111923A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113919395A (en) * 2021-10-12 2022-01-11 大连理工大学 Water supply pipe network leakage accident diagnosis method based on one-dimensional convolutional neural network
CN114152442A (en) * 2021-12-24 2022-03-08 杭州电子科技大学 Rolling bearing cross-working condition fault detection method based on migration convolutional neural network

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113919395A (en) * 2021-10-12 2022-01-11 大连理工大学 Water supply pipe network leakage accident diagnosis method based on one-dimensional convolutional neural network
CN114152442A (en) * 2021-12-24 2022-03-08 杭州电子科技大学 Rolling bearing cross-working condition fault detection method based on migration convolutional neural network

Similar Documents

Publication Publication Date Title
Hu et al. Novel leakage detection and water loss management of urban water supply network using multiscale neural networks
CN109783906B (en) Intelligent analysis system and method for detecting magnetic flux leakage data in pipeline
CN111853555B (en) Water supply pipe network blind leakage identification method based on dynamic process
Lee et al. Oil and gas pipeline failure prediction system using long range ultrasonic transducers and Euclidean-Support Vector Machines classification approach
CN112258496A (en) Underground drainage pipeline disease segmentation method based on full convolution neural network
CN111664365B (en) Oil and gas pipeline leakage detection method based on improved VMD and 1DCNN
Zhou et al. A pipeline leak detection and localization approach based on ensemble TL1DCNN
CN113111923A (en) Water supply network leakage detection and positioning method based on one-dimensional migration learning convolutional neural network integrated model
CN116258086B (en) Gas pipeline risk assessment method and system
Bohorquez et al. Stochastic resonance enhancement for leak detection in pipelines using fluid transients and convolutional neural networks
Wang et al. Real-time identification of natural gas pipeline leakage apertures based on lightweight residual convolutional neural network
CN113253248B (en) Small sample vertical array target distance estimation method based on transfer learning
CN116704137B (en) Reverse modeling method for point cloud deep learning of offshore oil drilling platform
CN109632942B (en) Inversion method of pipeline defect size based on ensemble learning
CN117332324A (en) Pipeline leakage detection method and device, electronic equipment and storage medium
CN113935377A (en) Pipeline leakage aperture identification method combining feature migration with time-frequency diagram
Sabu et al. Water leakage detection using neural networks
Parjane et al. Convolutional Neural Network Approach to Detect Underwater Pipeline Degradation using IOT dataset
Wang et al. Automatic Damage Segmentation Framework for Buried Sewer Pipes Based on Machine Vision: Case Study of Sewer Pipes in Zhengzhou, China
Huang et al. Residual Swin transformer-based weld crack leakage monitoring of pressure pipeline
CN113970073B (en) ResNet-based water supply network leakage accurate positioning method
Zhang et al. Dy-GNet: A lightweight and efficient 1DCNN-based network for leakage aperture identification
CN115452957B (en) Small sample metal damage identification method based on attention prototype network
KR102600664B1 (en) Multiple leakage detection method and system thereof
CN113743363B (en) Shielded target identification method based on small sample of unmanned aerial vehicle system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination