Water conveying pipeline defect classification method, device and equipment and storage medium

Abstract

The invention relates to the technical field of defect classification, and discloses a water conveying pipeline defect classification method, device and equipment and a storage medium, and the method comprises the steps: firstly, through a plurality of sonic sensors which are installed in a water conveying pipeline at equal intervals, a plurality of sonic signals when the water conveying pipeline has no defect and has different types of defects can be comprehensively collected from different positions; secondly, the acquired initial sound wave signal set is processed by using an empirical mode decomposition method and a Hilbert transform method, so that the processed target signal sample set has representativeness and analyzability; and finally, performing defect classification by using a multi-classification Stacking integration model integrated by four target Stacking integration models for distinguishing different defects, so that unique features of the type of defects can be better learned and identified, the detection accuracy of specific defects is improved, and the accuracy and reliability of a water conveying pipeline defect classification result are further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of defect classification, and in particular to a method, device, equipment and storage medium for classifying defects in a water pipeline. Background Art

[0002] Pipeline transportation has the advantages of less space, low construction cost, fast construction speed, high reliability and strong continuity. It is currently the main mode of water transportation. Therefore, water pipelines have become an important part of urban construction and an important lifeline of the country. In order to extend the service life of the pipeline and reduce damage caused by human and non-human factors, water pipelines are usually built under the surface. However, as the construction time goes by, the pipeline will inevitably leak, rupture, corrode and other phenomena. While wasting water resources, it may also cause a certain degree of pollution to the water quality.

[0003] In order to ensure the safety, structural integrity and sustainability of water pipeline operation, it is necessary to conduct quality inspections on water pipelines regularly. Currently, there are problems such as fuzzy defect determination and great difficulty in defect detection of buried water pipelines. Summary of the invention

[0004] In view of this, the present invention provides a method, device, equipment and storage medium for classifying defects in water pipelines to solve the problem that defect determination in buried water pipelines is ambiguous and difficult.

[0005] In a first aspect, the present invention provides a method for classifying defects in a water pipeline, wherein a plurality of acoustic wave sensors are arranged in the water pipeline, and the distances between the acoustic wave sensors are equal; the method comprises:

[0006] Multiple acoustic wave sensors are used to collect an initial acoustic wave signal set of a water pipeline, wherein the initial acoustic wave signal set includes multiple acoustic wave signals when the water pipeline is defect-free and has different types of defects. Based on the initial acoustic wave signal set, an empirical mode decomposition method and a Hilbert transform method are used to process the target signal sample set to obtain a target signal sample set. A multi-classification Stacking integration model is obtained, wherein the multi-classification Stacking integration model is integrated by four target Stacking integration models for distinguishing different defects. The multi-classification Stacking integration model is used to classify the water pipeline defects corresponding to the target signal sample set to obtain a classification result of the water pipeline defects.

[0007] The water pipeline defect classification method provided by the present invention firstly uses a plurality of equally spaced acoustic wave sensors installed in the water pipeline to comprehensively collect a plurality of acoustic wave signals when the water pipeline is defect-free and has different types of defects from different positions, thereby improving the comprehensiveness and accuracy of the initial acoustic wave signal set. Secondly, the collected initial acoustic wave signal set is processed using the empirical mode decomposition method and the Hilbert transform method, so that the processed target signal sample set can be more representative and analyzable. Finally, the target Stacking integrated model for distinguishing different defects can better learn and identify the unique characteristics of this type of defect, thereby improving the detection accuracy of specific defects, and thus improving the accuracy and reliability of the water pipeline defect classification results when the water pipeline defects corresponding to the target signal sample set are classified using a multi-classification Stacking integrated model integrated by four target Stacking integrated models for distinguishing different defects. Therefore, by implementing the present invention, accurate classification of water pipeline defects is achieved, and the problems of fuzzy defect judgment and difficulty in the current buried water pipeline defect detection are solved.

[0008] In an optional implementation, based on the initial sound wave signal set, after processing by the empirical mode decomposition method and the Hilbert transform method, a target signal sample set is obtained, including:

[0009] The initial acoustic wave signal set is decomposed by using the empirical mode decomposition method to obtain a plurality of intrinsic mode functions; the plurality of intrinsic mode functions are processed by the Hilbert transform method to obtain a plurality of initial signal instantaneous frequencies and a plurality of initial instantaneous amplitudes; based on the plurality of initial signal instantaneous frequencies and the plurality of initial instantaneous amplitudes, a target signal sample set is determined.

[0010] The water pipeline defect classification method provided by the present invention decomposes the initial acoustic wave signal set through the empirical mode decomposition method, which can split the complex signal into multiple relatively simple components, and is helpful for more in-depth analysis of the intrinsic structure and characteristics of the signal. Furthermore, the multiple intrinsic mode functions are processed by the Hilbert transform method, and the signal can be understood from two important dimensions, frequency and amplitude, based on the multiple initial signal instantaneous frequencies and multiple initial instantaneous amplitudes obtained, which greatly enriches the grasp of signal characteristics and provides more information basis for further analysis and processing of signals. Finally, by combining multiple initial signal instantaneous frequencies and multiple initial instantaneous amplitudes to determine the target signal sample set, more representative and analyzable signal samples can be screened out more accurately.

[0011] In an optional implementation, obtaining a multi-classification Stacking integration model includes:

[0012] A historical acoustic wave signal set of a water pipeline is collected by using multiple acoustic wave sensors; based on the historical acoustic wave signal set, a historical signal sample set is obtained by processing it using an empirical mode decomposition method and a Hilbert transform method; a preset Stacking integration model is trained based on the historical signal sample set to obtain a first target Stacking integration model, which is used to distinguish pipeline leakage defects from other defects; the step of training the preset Stacking integration model based on the historical signal sample set is repeated to obtain a second target Stacking integration model, a third target Stacking integration model and a fourth target Stacking integration model, which are used to distinguish pipeline rupture defects from other defects, the third target Stacking integration model is used to distinguish pipeline corrosion defects from other defects, and the fourth target Stacking integration model is used to distinguish pipeline non-defective conditions from other defects; a multi-classification Stacking integration model is determined based on the first target Stacking integration model, the second target Stacking integration model, the third target Stacking integration model and the fourth target Stacking integration model.

[0013] The water pipeline defect classification method provided by the present invention repeats the steps of training the preset Stacking integration model through the historical signal sample set, and can obtain a target Stacking integration model that distinguishes different defect types. This targeted training method can enable each model to focus on a specific defect differentiation task, better learn and capture the characteristics of the corresponding defects, thereby improving the accuracy of the judgment of specific defects. Finally, the multi-classification Stacking integration model is determined by combining the obtained multiple target Stacking integration models, which can combine the advantages of each model to achieve multi-classification judgment of the pipeline status, thereby providing support for improving the accuracy and reliability of the water pipeline defect classification results.

[0014] In an optional implementation, a preset Stacking integration model is trained based on a historical signal sample set to obtain a first target Stacking integration model, including:

[0015] Based on the historical signal sample set, the first-layer classifier of the preset Stacking ensemble model is trained using the extreme gradient boosting tree algorithm, K nearest neighbor algorithm and decision tree algorithm to obtain the target feature data set; based on the target feature data set, the second-layer classifier of the preset Stacking ensemble model is trained using the logistic regression algorithm to obtain the first target Stacking ensemble model.

[0016] The water pipeline defect classification method provided by the present invention respectively uses the extreme gradient boosting tree algorithm, the K nearest neighbor algorithm and the decision tree algorithm to train the first layer classifier of the preset Stacking integrated model, which fully combines the advantages of different algorithms and can extract the characteristics of the historical signal sample set from different angles, thereby improving the accuracy and robustness of the first layer classifier, so that the obtained target feature data set can fuse the feature information extracted by different algorithms, and can more comprehensively and accurately reflect the essential characteristics of the historical signal samples. Furthermore, the second layer classifier of the preset Stacking integrated model is trained using a logistic regression algorithm, which can further optimize the classification effect of the model and improve the ability to distinguish different situations. Finally, after training two layers of classifiers, a first target Stacking integrated model with higher accuracy and reliability can be obtained.

[0017] In an optional implementation, based on the historical signal sample set, the first layer classifier of the preset Stacking ensemble model is trained using the extreme gradient boosting tree algorithm, the K nearest neighbor algorithm, and the decision tree algorithm to obtain a target feature data set, including:

[0018] Based on the historical signal sample set, the first layer classifier of the preset Stacking integrated model is trained using the extreme gradient boosting tree algorithm, the K nearest neighbor algorithm and the decision tree algorithm to obtain multiple initial feature data sets; multiple feature evaluation values ​​of each initial feature data set are calculated; based on the multiple feature evaluation values ​​of each initial feature data set, the target weight value of each initial feature data set is calculated; based on the multiple initial feature data sets and the target weight value of each initial feature data set, the target feature data set is determined.

[0019] The water pipeline defect classification method provided by the present invention inserts an evaluation of the classification effects of the three classifiers of the first layer between the first layer classifier and the second layer classifier, improves the accuracy of the evaluation by using a combination of multiple evaluation criteria, and uses the evaluation results to add weights to the data input by the first layer classifier into the second layer classifier to further improve the accuracy of the entire model.

[0020] In an optional implementation, a multi-classification Stacking integrated model is used to classify the water pipeline defects corresponding to the target signal sample set to obtain a water pipeline defect classification result, including:

[0021] The first target Stacking integration model, the second target Stacking integration model, the third target Stacking integration model and the fourth target Stacking integration model are respectively used to classify the water pipeline defects corresponding to the target signal sample set to obtain multiple initial classification results; based on the multiple initial classification results, the water pipeline defect classification results are determined.

[0022] The water pipeline defect classification method provided by the present invention utilizes four target Stacking integrated models for distinguishing different defects to classify the water pipeline defects corresponding to the target signal sample set, which can comprehensively consider various defect types that may occur in the water pipeline, ensure accurate judgment of the pipeline status, and improve the accuracy and reliability of the water pipeline defect classification results.

[0023] In a second aspect, the present invention provides a water pipeline defect classification device, wherein a plurality of acoustic wave sensors are arranged in the water pipeline, and the distances between the acoustic wave sensors are equal; the device comprises:

[0024] The acquisition module is used to use multiple acoustic wave sensors to collect the initial acoustic wave signal set of the water pipeline, and the initial acoustic wave signal set includes multiple acoustic wave signals when the water pipeline is defect-free and has different types of defects; the processing module is used to obtain the target signal sample set based on the initial acoustic wave signal set through the empirical mode decomposition method and the Hilbert transform method; the acquisition module is used to obtain a multi-classification Stacking integration model, and the multi-classification Stacking integration model is integrated by four target Stacking integration models for distinguishing different defects; the classification module is used to classify the water pipeline defects corresponding to the target signal sample set using the multi-classification Stacking integration model to obtain the water pipeline defect classification result.

[0025] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the water pipeline defect classification method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0026] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the water pipeline defect classification method of the first aspect or any corresponding embodiment thereof.

[0027] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the water pipeline defect classification method of the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0029] Figure 1 is a schematic flow chart of a method for classifying defects in a water pipeline according to an embodiment of the present invention;

[0030] Figure 2 is a flow chart of another method for classifying defects in a water pipeline according to an embodiment of the present invention;

[0031] Figure 3 is a flow chart of another method for classifying defects in a water pipeline according to an embodiment of the present invention;

[0032] Figure 4 is a structural block diagram of a water pipeline defect classification device according to an embodiment of the present invention;

[0033] Figure 5 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0035] At present, the acoustic wave detection method is a pipeline defect detection method developed based on the principle that objects vibrate and emit sound after collision, forming sound waves. Compared with image monitoring, optical fiber temperature sensor detection and other methods, the method of installing acoustic wave sensors for pipeline acoustic wave monitoring has the advantages of simple installation conditions, sensitive detection, and is particularly effective for small flow defect detection. However, due to the large noise interference of acoustic wave signals, it is difficult to extract defect signals.

[0036] As a new method, intelligent models such as machine learning and deep learning have great advantages in identification and classification and have been used in fault diagnosis and defect detection in many fields. This method can achieve defect classification by extracting appropriate underlying features from acoustic wave signals and optimizing classifiers. However, due to the small number of pipeline acoustic wave defect samples, data imbalance and low classification precision often occur in actual analysis.

[0037] According to an embodiment of the present invention, an embodiment of a method for classifying defects in a water pipeline is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0038] In this embodiment, a method for classifying defects in a water pipeline is provided, which can be used in electronic devices such as computers, mobile phones, tablet computers, etc. A plurality of acoustic wave sensors are arranged in the water pipeline, and the distances between the acoustic wave sensors are equal.

[0039] Figure 1 is a flow chart of a method for classifying defects in a water pipeline according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0040] Step S101: using a plurality of acoustic wave sensors to collect an initial acoustic wave signal set of a water pipeline.

[0041] The initial sound wave signal set may include multiple sound wave signals when the water pipeline is free of defects and when the water pipeline has defects of different types.

[0042] Specifically, when a water pipeline has a defect, the pressure inside and outside the pipeline around the defect changes, and the pressure at the defect point increases. The interaction between water and pipeline, the impact of water on the medium in which the pipeline is buried, and the friction between stones and gravel will all generate acoustic signals, which are collected by acoustic sensors. Acoustic sensors are installed on the pipeline at regular intervals. This evenly distributed installation method ensures that acoustic signals can be captured in a timely manner at different locations, regardless of whether there are no defects or different types of defects.

[0043] Furthermore, by using an acoustic wave sensor to collect acoustic wave signals when there are no defects and when there are different types of defects, and recording the defect types corresponding to the acoustic wave signals when defects occur, an initial acoustic wave signal set is obtained, which can provide clear labels for subsequent analysis and classification, and facilitate and accurately identify different types of defects.

[0044] Step S102: Based on the initial sound wave signal set, the target signal sample set is obtained by processing the initial sound wave signal set using the empirical mode decomposition method and the Hilbert transform method.

[0045] Among them, the empirical mode decomposition method (EMD) is a signal processing method, which is mainly used to decompose a complex signal into multiple intrinsic mode functions (IMFs) and a residual.

[0046] Hilbert transform is a signal processing method, which is mainly used to convert real signals into analytical signals to obtain information such as the instantaneous frequency and instantaneous amplitude of the signal.

[0047] Specifically, by combining the empirical mode decomposition method and the Hilbert transform method to analyze and process the initial sound wave signal set, the processed target signal sample set can be made more representative and analyzable.

[0048] Step S103, obtaining a multi-classification Stacking integration model.

[0049] Among them, the multi-classification Stacking ensemble model is integrated by four target Stacking ensemble models for distinguishing different defects.

[0050] Specifically, the multi-classification Stacking ensemble model is a model that classifies multiple different categories by integrating multiple sub-models.

[0051] Furthermore, the target Stacking integration model for distinguishing different defects can better learn and identify the unique features of this type of defects, thereby improving the detection accuracy of specific defects. Therefore, in this embodiment, by integrating four target Stacking integration models for distinguishing different defects, the characteristics of the corresponding defects can be better learned and identified.

[0052] Step S104, using a multi-classification Stacking integrated model to classify the water pipeline defects corresponding to the target signal sample set, to obtain a water pipeline defect classification result.

[0053] Specifically, the obtained target signal sample set is input into the multi-classification Stacking integration model for processing, and the corresponding water pipeline defect classification results are output.

[0054] The water pipeline defect classification method provided in this embodiment firstly uses a plurality of equally spaced acoustic wave sensors installed in the water pipeline to comprehensively collect multiple acoustic wave signals when the water pipeline is defect-free and has different types of defects from different positions, thereby improving the comprehensiveness and accuracy of the initial acoustic wave signal set. Secondly, the collected initial acoustic wave signal set is processed using the empirical mode decomposition method and the Hilbert transform method, so that the processed target signal sample set can be more representative and analyzable. Finally, the target Stacking integrated model for distinguishing different defects can better learn and identify the unique characteristics of this type of defect, thereby improving the detection accuracy of specific defects, and thus improving the accuracy and reliability of the water pipeline defect classification results when the multi-classification Stacking integrated model integrated by four target Stacking integrated models for distinguishing different defects is used to classify the water pipeline defects corresponding to the target signal sample set. Therefore, by implementing the present invention, accurate classification of water pipeline defects is achieved, and the problems of fuzzy defect judgment and difficulty in the current detection of buried water pipeline defects are solved.

[0055] In this embodiment, a method for classifying defects in a water pipeline is provided, which can be used in electronic devices such as computers, mobile phones, tablet computers, etc. A plurality of acoustic wave sensors are arranged in the water pipeline, and the distances between the acoustic wave sensors are equal. Figure 2 is a flow chart of a method for classifying defects in a water pipeline according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:

[0056] Step S201, using multiple acoustic wave sensors to collect an initial acoustic wave signal set of the water pipeline. Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0057] Step S202: Based on the initial sound wave signal set, the target signal sample set is obtained by processing the initial sound wave signal set using the empirical mode decomposition method and the Hilbert transform method.

[0058] Specifically, the above step S202 includes:

[0059] Step S2021, using an empirical mode decomposition method to decompose the initial sound wave signal set to obtain a plurality of intrinsic mode functions.

[0060] Specifically, the specific process of decomposing the initial sound wave signal set using the empirical mode decomposition method may include:

[0061] (1) Find all the extreme points in the original signal f(t), that is, the initial sound wave signal set.

[0062] (2) Use the cubic spline difference method to envelop the maximum point of the signal and obtain the upper envelope fmax (t), similarly, envelop the minimum value of the signal and obtain the lower envelope f min (t).

[0063] Among them, the cubic spline interpolation method represents a curve fitting method that can generate a smooth curve between a given set of data points. At the maximum point of the signal, a continuous upper envelope can be obtained through cubic spline interpolation, which can better fit the local maximum of the signal.

[0064] (3) Take the mean of the upper and lower envelopes to get a mean line g 1 (t):

[0065]

[0066] Furthermore, the mean line g 1 (t) can reflect the average trend of the signal within a certain time range. By calculating the mean of the upper and lower envelopes, the local fluctuations in the signal can be removed to obtain a relatively smooth trend line.

[0067] (4) Compare the original signal f(t) with the mean line g 1 (t) Take the difference and get h 1 (t):

[0068] h 1 (t) = f(t) - g 1 (t)

[0069] Specifically, h 1 (t) can reflect the difference between the original signal and the average trend.

[0070] (5) Test g 1 (t) Whether the two constraints of the intrinsic mode function (IMF) are met. If not, h 1 (t) is regarded as the original signal, and the above steps (1) to (4) are repeated until the condition is satisfied. In this embodiment, the SD threshold screening termination formula is used for judgment:

[0071]

[0072] Among them, the two constraints of the IMF can include:

[0073] 1) In the entire time range, the number of local extreme points and zero crossing points must be equal or differ by at most one;

[0074] 2) At any time, the mean of the upper envelope defined by the local maximum point and the lower envelope defined by the local minimum point is zero.

[0075] Furthermore, SD is an indicator that measures the difference between the signal and the IMF. When the value of SD is between 0.2 and 0.3, it can be considered that g 1 (t) Satisfy the two constraints of the IMF.

[0076] (6) When h 1 After satisfying the two constraints of IMF, the original signal f(t) and h 1 (t) is subtracted to obtain the residual signal r 1 (t):

[0077] r 1 (t) = f(t) - h 1 (t)

[0078] Specifically, the residual signal r 1 (t) may contain the remainder of the original signal after removing one IMF component.

[0079] (7) The remaining signal r 1 (t) is taken as the new original signal, and the above steps (1) to (6) are repeated to obtain the second component h 2 (t), the third component h 2 (t), up to the nth component h n (t), until the remaining signal r n (t) is a monotone sequence or a constant series. The decomposition is completed. Further, the original signal can be expressed as:

[0080]

[0081] Where: h i (t) represents the IMF components of different frequencies in the original signal, which are distributed from high frequency to low frequency; r n (t) represents the average change trend of the original signal.

[0082] The above-mentioned empirical mode decomposition method can decompose complex acoustic wave signals into multiple relatively simple IMF components and a residual signal. Each IMF component represents the characteristics of the signal at different frequency scales, which is helpful for more in-depth analysis and processing of acoustic wave signals.

[0083] Step S2022: Process the multiple intrinsic mode functions using the Hilbert transform method to obtain multiple initial signal instantaneous frequencies and multiple initial instantaneous amplitudes.

[0084] Specifically, the IMF components decomposed by the empirical mode decomposition method are then subjected to Hilbert transform to finally obtain multiple signal instantaneous frequencies and multiple instantaneous amplitudes, which may specifically include the following steps:

[0085] (1) Perform Hilbert transform on each IMF component to construct an analytical signal. The real part of the analytical signal is the original IMF component, and the imaginary part is the result after Hilbert transform.

[0086] (2) The instantaneous frequency and instantaneous amplitude can be easily calculated by analyzing the signal.

[0087] Step S2023: determining a target signal sample set based on a plurality of initial signal instantaneous frequencies and a plurality of initial instantaneous amplitudes.

[0088] Specifically, all the obtained multiple initial signal instantaneous frequencies and multiple initial instantaneous amplitudes are The form indicates, is a vector representing frequency, that is, represents the target signal sample set.

[0089] Through this It can represent the frequency changes of the signal at different time points, providing important information for further analysis of the signal characteristics.

[0090] Step S203, obtain a multi-classification Stacking ensemble model. Figure 1 Step S103 of the illustrated embodiment will not be described in detail here.

[0091] Step S204: Use the multi-classification Stacking integrated model to classify the water pipeline defects corresponding to the target signal sample set to obtain the water pipeline defect classification result. Figure 1 Step S104 of the illustrated embodiment will not be described in detail here.

[0092] The water pipeline defect classification method provided in this embodiment decomposes the initial acoustic wave signal set through the empirical mode decomposition method, which can split the complex signal into multiple relatively simple components, which is helpful for more in-depth analysis of the intrinsic structure and characteristics of the signal. Furthermore, by processing multiple intrinsic mode functions through the Hilbert transform method, the signal can be understood from two important dimensions, frequency and amplitude, based on the obtained multiple initial signal instantaneous frequencies and multiple initial instantaneous amplitudes, which greatly enriches the grasp of signal characteristics and provides more information basis for further analysis and processing of signals. Finally, by combining multiple initial signal instantaneous frequencies and multiple initial instantaneous amplitudes to determine the target signal sample set, more representative and analyzable signal samples can be screened out more accurately.

[0093] In this embodiment, a method for classifying defects in a water pipeline is provided, which can be used in electronic devices such as computers, mobile phones, tablet computers, etc. A plurality of acoustic wave sensors are arranged in the water pipeline, and the distances between the acoustic wave sensors are equal. Figure 3is a flow chart of a method for classifying defects in a water pipeline according to an embodiment of the present invention. Figure 3 As shown, the process includes the following steps:

[0094] Step S301, using multiple acoustic wave sensors to collect an initial acoustic wave signal set of the water pipeline. Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0095] Step S302: Based on the initial sound wave signal set, the target signal sample set is obtained by processing the initial sound wave signal set using the empirical mode decomposition method and the Hilbert transform method. Figure 2 Step S202 of the illustrated embodiment will not be described in detail here.

[0096] Step S303, obtaining a multi-classification Stacking integration model.

[0097] Specifically, the above step S303 includes:

[0098] Step S3031, using multiple acoustic wave sensors to collect a historical acoustic wave signal set of the water pipeline.

[0099] The specific process can be referred to the description of step S101 above, which will not be repeated here.

[0100] Step S3032: Based on the historical sound wave signal set, the historical signal sample set is obtained by processing using the empirical mode decomposition method and the Hilbert transform method.

[0101] The specific process can refer to the description of step S202 above, which will not be repeated here.

[0102] Step S3033: train the preset Stacking integration model based on the historical signal sample set to obtain a first target Stacking integration model.

[0103] Among them, the first target Stacking integrated model is used to distinguish pipeline leakage defects from other defects.

[0104] In some optional implementations, the above step S3033 includes:

[0105] Step a1: Based on the historical signal sample set, the first layer classifier of the preset Stacking ensemble model is trained using the extreme gradient boosting tree algorithm, the K nearest neighbor algorithm and the decision tree algorithm to obtain the target feature data set.

[0106] Among them, the extreme gradient boosting tree algorithm (eXtreme Gradient Boosting, XGBoost) represents an efficient gradient boosting tree algorithm implementation with high accuracy and efficiency; the K-Nearest Neighbors algorithm (KNN) represents a simple and effective machine learning algorithm, mainly used for classification and regression problems; the decision tree algorithm (Light Gradient Boosting Machine, LGBM) represents an efficient gradient boosting framework that improves the performance of the model by continuously adding decision trees.

[0107] Specifically, the preset Stacking integration model has two layers. The first layer is the learning layer, which includes three different strong learners. The model is trained using three strong learners: K-nearest neighbor (KNN), extreme gradient boosting (Xgboost) and decision tree (LGBM). The training results are synthesized into new feature data as the input of the second layer.

[0108] First, the generated historical signal sample set is divided into training set A and training set B according to the ratio of 7:3. train and test set A test To prevent overfitting, a 4-fold cross validation method was used to train the training set A. train Perform 4-fold split to form 4 sub-datasets {A train1 , A train2 , A train3 , A train4}. The i-th sub-dataset A traini As a new validation set. The other three sub-datasets are used as new training sets and input into the K-nearest neighbor (KNN) algorithm for training. The number of KNN neighbors is 6. After the training is completed, the validation set A traini Input into the trained K-nearest neighbor (KNN) algorithm to obtain the classification result P 1-i . The test set A test Input into the trained K-nearest neighbor (KNN) algorithm to obtain the classification result T 1-i Repeat the operation four times, with the values ​​of i being 1, 2, 3, and 4 respectively. 1-1 , P 1-2 , P 1-3 , P 1-4 With T 1-1 , T 1-2 , T 1-3 , T 1-4 .

[0109] Furthermore, P 1-1 , P 1-2 , P 1-3 , P 1-4 Integrate into a set P of 4 rows and 1 column1 , T 1-1 , T 1-2 , T 1-3 , T 1-4 Integrate into a set T of 4 rows and 1 column 1 , and complete the K-nearest neighbor (KNN) algorithm training.

[0110] Secondly, the generated historical signal sample set is divided into training set A according to 7:3 train and test set A test To prevent overfitting, a 4-fold cross validation method was used to train the training set A. train Perform 4-fold split to form 4 sub-datasets {A train1 , A train2 , A train3 , A train4}. The i-th sub-dataset A traini As a new validation set, the other three sub-datasets were used as new training sets and input into the extreme gradient boosting (Xgboost) algorithm for training.

[0111] The Xgboost model uses the L2 regularization model, namely Ridge regression, and adds a term λ‖w‖ after the loss function 2 2 After training, the validation set A traini Input into the trained extreme gradient boosting (Xgboost) algorithm to obtain the classification result P 2-i . The test set A test Input into the trained extreme gradient boosting (Xgboost) algorithm to obtain the classification result T 2-i Repeat the operation four times, with the values ​​of i being 1, 2, 3, and 4 respectively. 2-1 , P 2-2 , P 2-3 , P 2-4 With T 2-1 , T 2-2 , T 2-3 , T 2-4 .

[0112] Furthermore, P 2-1 , P 2-2 , P 2-3 , P 2-4 Integrate into a set P of 4 rows and 1 column 2 , T 2-1 , T 2-2 , T 2-3 , T 2-4 Integrate into a set T of 4 rows and 1 column 2 , and complete the extreme gradient boosting (Xgboost) algorithm training.

[0113] Finally, the generated historical signal sample set is divided into training set A and training set B according to the ratio of 7:3. train and test set A test To prevent overfitting, a 4-fold cross validation method was used to train the training set A. train Perform 4-fold split to form 4 sub-datasets {A train1 , A train2 , A train3 , A train4}. The i-th sub-dataset A traini As a new validation set, the other three sub-datasets were used as new training sets and input into the decision tree algorithm (LGBM) for training.

[0114] After training, the validation set A traini Input into the trained decision tree algorithm (LGBM) to obtain the classification result P 3-i . The test set A test Input into the trained decision tree algorithm (LGBM) to obtain the classification result T 3-i Repeat the operation four times, with the values ​​of i being 1, 2, 3, and 4 respectively. 3-1 , P 3-2 , P 3-3 , P 3-4 With T 3-1 , T 3-2 , T 3-3 , T 3-4 .

[0115] Furthermore, P 3-1 , P 3-2 , P 3-3 , P 3-4 Integrate into a set P of 4 rows and 1 column 3 , T 3-1 , T 3-2 , T 3-3 , T 3-4 Integrate into a set T of 4 rows and 1 column 3 , and complete the decision tree algorithm (LGBM) training.

[0116] In some optional implementations, the above step a1 includes:

[0117] Step a11, based on the historical signal sample set, the first layer classifier of the preset Stacking ensemble model is trained using the extreme gradient boosting tree algorithm, the K nearest neighbor algorithm and the decision tree algorithm to obtain multiple initial feature data sets.

[0118] Step a12, calculating multiple feature evaluation values ​​of each initial feature data set.

[0119] Step a13, calculating the target weight value of each initial feature data set based on the multiple feature evaluation values ​​of each initial feature data set.

[0120] Step a14, determining a target feature data set based on multiple initial feature data sets and target weight values ​​of each initial feature data set.

[0121] According to the description of step a1 above, the multiple initial feature data sets are P 1 , P 2 , P 3 , T 1 , T 2 , T 3 .

[0122] Specifically, since the performance of the models in the first-layer classifiers varies, in order to better obtain effective information in the first-layer classifiers, the output results of the three classifiers in the first layer are weighted averaged based on the performance of the first-layer classifiers. In order to more accurately judge the performance of the first-layer classifiers, three evaluation methods, accuracy, F1 value and AUC value, are integrated, that is, the multiple feature evaluation values ​​corresponding to each initial feature data set are accuracy, F1 value and AUC value.

[0123] Each evaluation method measures the performance of the model from different perspectives:

[0124] (1) Accuracy: It directly reflects the proportion of correct predictions made by the model and can be calculated using the following relationship:

[0125]

[0126] Where TP represents a true positive example (actually a positive example and predicted as a positive example); TN represents a true negative example (actually a negative example and predicted as a negative example); FP represents a false positive example (actually a negative example but predicted as a positive example); and FN represents a false negative example (actually a positive example but predicted as a negative example).

[0127] (2) F1 value: It takes into account both precision and recall and can be calculated using the following relationship:

[0128]

[0129] Where: Indicates accuracy; represents the recall rate;

[0130] (3) AUC value: The area under the ROC curve is used to measure the classification performance of the model. The closer the AUC value is to 1, the better the model prediction effect is; the closer it is to 0.5, the worse the model classification effect is.

[0131] Furthermore, the values ​​of the three evaluation methods are calculated for the three first-layer classifiers respectively. Then the results are added together and normalized to serve as the weights W of the three first-layer classifiers. 1 , W 2 , W 3 .

[0132] Finally, W 1 P 1 , W 2 P 2 , W 3 P 3 , W 1 T 1 , W 2 T 2 , W 3 T 3 As the target feature dataset B.

[0133] Step a2: Based on the target feature data set, the second layer classifier of the preset Stacking integration model is trained using a logistic regression algorithm to obtain a first target Stacking integration model.

[0134] Specifically, the second layer of the preset Stacking ensemble model uses a logistic regression classifier to complete the classification of the data: W 1 P 1 , W 2 P 2 , W 3 P 3 As the feature of the new training set input to the second layer classifier, W 1 T 1 , W 2 T 2 , W 3 T 3 As the features of the new test set input into the second-layer classifier, the new training set is used to train the logistic regression algorithm for classification, and the new test set is used to test the training results.

[0135] As the input target feature dataset B of the second-layer classifier, the target feature dataset B is split into 4 groups, forming 4 groups of datasets, of which 3 groups are used as training sets and 1 group is used as validation set. The training set of each group is used for logistic regression training, and then the validation set is predicted to obtain the output of the corresponding validation set. The four groups of data are averaged to obtain dataset B 1 , while B 1 It is the final output result of the first target Stacking integration model.

[0136] Finally, after training the two-layer classifier, the first target Stacking ensemble model with higher accuracy and reliability can be obtained.

[0137] Step S3034, repeat the step of training the preset Stacking integration model based on the historical signal sample set, and obtain the second target Stacking integration model, the third target Stacking integration model and the fourth target Stacking integration model.

[0138] Among them, the second target Stacking integrated model is used to distinguish pipeline rupture defects from other defects, the third target Stacking integrated model is used to distinguish pipeline corrosion defects from other defects, and the fourth target Stacking integrated model is used to distinguish pipeline defect-free conditions from other defects.

[0139] Specifically, the above step S3033 is repeatedly performed, and target stacking integrated models for different purposes are trained. Each target model is distinguished for a specific pipeline defect type to meet the accurate analysis requirements for different defect conditions of the water pipeline.

[0140] Step S3035, determining a multi-classification Stacking integration model based on the first target Stacking integration model, the second target Stacking integration model, the third target Stacking integration model and the fourth target Stacking integration model.

[0141] Specifically, the trained first target Stacking integration model, second target Stacking integration model, third target Stacking integration model and fourth target Stacking integration model can be integrated into a multi-classification Stacking integration model, which can combine the advantages of each model to realize multi-classification judgment of pipeline status, thereby providing support for improving the accuracy and reliability of water pipeline defect classification results.

[0142] Step S304: classify the water pipeline defects corresponding to the target signal sample set using a multi-classification Stacking integrated model to obtain a water pipeline defect classification result.

[0143] Specifically, the above step S304 includes:

[0144] Step S3041, respectively classify the water pipeline defects corresponding to the target signal sample set using the first target Stacking integration model, the second target Stacking integration model, the third target Stacking integration model and the fourth target Stacking integration model to obtain multiple initial classification results.

[0145] Specifically, based on the first target Stacking integration model, the second target Stacking integration model, the third target Stacking integration model and the fourth target Stacking integration model, a one-to-many combination can be used for classification.

[0146] One-to-many combination classification is a multi-classification strategy. In this problem, each defect type (leakage, rupture, corrosion, no defect) is distinguished from other types, thereby achieving multi-classification of water pipeline defects. Specifically, it can include:

[0147] (1) For the first target Stacking ensemble model, it is trained to distinguish pipeline leakage defects from other defects. When classifying the target signal sample set, the model will output a result indicating that the sample belongs to a leakage defect or does not belong to a leakage defect (i.e., one of the other types).

[0148] (2) The second objective Stacking ensemble model is used to distinguish pipeline rupture defects from other defects and to determine whether a sample is a rupture defect when classifying the target signal sample set.

[0149] (3) The third objective is that the Stacking integrated model distinguishes pipeline corrosion defects from other defects and determines whether the sample is a corrosion defect.

[0150] (4) The fourth goal is that the Stacking integrated model distinguishes between defect-free conditions and other defects and determines whether the sample is in a defect-free state.

[0151] Furthermore, by respectively classifying the water pipeline defects corresponding to the target signal sample set through the above-mentioned first target Stacking integration model, second target Stacking integration model, third target Stacking integration model and fourth target Stacking integration model, multiple corresponding initial classification results can be obtained. For example, the first target model may output that the probability of the sample belonging to a leakage defect is 0.7, and the probability of not belonging to a leakage defect is 0.3; the second target model may output that the probability of the sample belonging to a rupture defect is 0.2, and the probability of not belonging to a rupture defect is 0.8. By analogy, each target Stacking integration model will give a judgment result on a specific defect type.

[0152] Step S3042, determining the water pipeline defect classification result based on multiple initial classification results.

[0153] Specifically, by combining the results of four Stacking ensemble models, four situations, namely pipeline leakage, pipeline rupture, pipeline corrosion, and pipeline integrity, are distinguished to complete the defect classification task.

[0154] The water pipeline defect classification method provided in this embodiment uses the extreme gradient boosting tree algorithm, the K nearest neighbor algorithm and the decision tree algorithm to train the first layer classifier of the preset Stacking integrated model, which fully combines the advantages of different algorithms, can extract the features of the historical signal sample set from different angles, and improve the accuracy and robustness of the first layer classifier, so that the obtained target feature data set can integrate the feature information extracted by different algorithms, and can more comprehensively and accurately reflect the essential characteristics of the historical signal samples. Further, the second layer classifier of the preset Stacking integrated model is trained using the logistic regression algorithm, which can further optimize the classification effect of the model and improve the ability to distinguish different situations. Finally, after training the two-layer classifier, the first target Stacking integrated model with higher accuracy and reliability can be obtained. At the same time, the evaluation of the classification effect of the three classifiers of the first layer is inserted between the first layer classifier and the second layer classifier, and the evaluation method of combining multiple evaluation criteria is used to improve the accuracy of the evaluation. The method of adding weights to the data input by the first layer classifier into the second layer classifier using the evaluation results further improves the accuracy of the entire model. Finally, four target Stacking ensemble models for distinguishing different defects are used to classify the water pipeline defects corresponding to the target signal sample set. This can comprehensively consider various defect types that may occur in the water pipeline, ensure accurate judgment of the pipeline status, and improve the accuracy and reliability of the water pipeline defect classification results.

[0155] In this embodiment, a water pipeline defect classification device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0156] This embodiment provides a water pipeline defect classification device, wherein a plurality of acoustic wave sensors are arranged in the water pipeline, and the distances between the acoustic wave sensors are equal; Figure 4 As shown, the device comprises:

[0157] The acquisition module 401 is used to use multiple acoustic wave sensors to acquire an initial acoustic wave signal set of the water pipeline, wherein the initial acoustic wave signal set includes multiple acoustic wave signals when the water pipeline is free of defects and when the water pipeline has defects of different types.

[0158] The processing module 402 is used to obtain a target signal sample set based on the initial sound wave signal set by processing the initial sound wave signal set using an empirical mode decomposition method and a Hilbert transform method.

[0159] The acquisition module 403 is used to acquire a multi-classification Stacking integration model, where the multi-classification Stacking integration model is integrated by four target Stacking integration models for distinguishing different defects.

[0160] The classification module 404 is used to classify the water pipeline defects corresponding to the target signal sample set using a multi-classification Stacking integrated model to obtain a water pipeline defect classification result.

[0161] In some optional implementations, the processing module 402 includes:

[0162] The decomposition submodule is used to decompose the initial sound wave signal set using the empirical mode decomposition method to obtain multiple intrinsic mode functions.

[0163] The processing submodule is used to process multiple intrinsic mode functions through the Hilbert transform method to obtain multiple initial signal instantaneous frequencies and multiple initial instantaneous amplitudes.

[0164] The first determination submodule is used to determine a target signal sample set based on a plurality of initial signal instantaneous frequencies and a plurality of initial instantaneous amplitudes.

[0165] In some optional implementations, the acquisition module 403 includes:

[0166] The acquisition submodule is used to collect the historical acoustic wave signal set of the water pipeline using multiple acoustic wave sensors.

[0167] The processing submodule is used to obtain a historical signal sample set based on the historical sound wave signal set through processing using the empirical mode decomposition method and the Hilbert transform method.

[0168] The training submodule is used to train the preset Stacking integration model based on the historical signal sample set to obtain the first target Stacking integration model, which is used to distinguish pipeline leakage defects from other defects.

[0169] The repeating submodule is used to repeat the steps of training the preset Stacking integration model based on the historical signal sample set, and obtain the second target Stacking integration model, the third target Stacking integration model and the fourth target Stacking integration model. The second target Stacking integration model is used to distinguish the pipeline rupture defect from other defects, the third target Stacking integration model is used to distinguish the pipeline corrosion defect from other defects, and the fourth target Stacking integration model is used to distinguish the pipeline defect-free situation from other defects.

[0170] The second determination submodule is used to determine a multi-classification Stacking integration model based on the first target Stacking integration model, the second target Stacking integration model, the third target Stacking integration model and the fourth target Stacking integration model.

[0171] In some optional embodiments, the training submodule includes:

[0172] The first training unit is used to train the first layer classifier of the preset Stacking integration model based on the historical signal sample set using the extreme gradient boosting tree algorithm, the K nearest neighbor algorithm and the decision tree algorithm to obtain the target feature data set.

[0173] The second training unit is used to train the second layer classifier of the preset Stacking integration model based on the target feature data set by using a logistic regression algorithm to obtain a first target Stacking integration model.

[0174] In some optional implementations, the first training unit includes:

[0175] The training subunit is used to train the first layer classifier of the preset Stacking ensemble model based on the historical signal sample set using the extreme gradient boosting tree algorithm, the K nearest neighbor algorithm and the decision tree algorithm to obtain multiple initial feature data sets.

[0176] The first calculation subunit is used to calculate multiple feature evaluation values ​​of each initial feature data set.

[0177] The second calculation subunit is used to calculate the target weight value of each initial feature data set based on the multiple feature evaluation values ​​of each initial feature data set.

[0178] The determination subunit is used to determine a target feature data set based on a plurality of initial feature data sets and a target weight value of each initial feature data set.

[0179] In some optional implementations, the classification module 404 includes:

[0180] The classification submodule is used to classify the water pipeline defects corresponding to the target signal sample set using the first target Stacking integration model, the second target Stacking integration model, the third target Stacking integration model and the fourth target Stacking integration model respectively, and obtain multiple initial classification results.

[0181] The third determination submodule is used to determine the water pipeline defect classification result based on multiple initial classification results.

[0182] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0183] The water pipeline defect classification device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0184] The embodiment of the present invention also provides a computer device having the above Figure 4 The water pipeline defect classification device shown.

[0185] See also Figure 5 , Figure 5 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 5 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 5 A processor 10 is taken as an example.

[0186] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0187] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0188] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0189] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0190] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0191] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0192] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0193] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for classifying defects in water pipelines, characterized in that: A plurality of acoustic wave sensors are arranged in the water delivery pipeline, and the distances between the acoustic wave sensors are equal; the method comprises: Using the multiple acoustic wave sensors to collect an initial acoustic wave signal set of the water pipeline, the initial acoustic wave signal set includes multiple acoustic wave signals when the water pipeline is free of defects and has different types of defects; Based on the initial sound wave signal set, a target signal sample set is obtained by processing with an empirical mode decomposition method and a Hilbert transform method; Obtain a multi-classification Stacking ensemble model, where the multi-classification Stacking ensemble model is integrated by four target Stacking ensemble models for distinguishing different defects; The multi-classification Stacking integrated model is used to classify the water pipeline defects corresponding to the target signal sample set to obtain a water pipeline defect classification result.

2. The method according to claim 1, characterized in that Based on the initial sound wave signal set, the target signal sample set is obtained by processing with the empirical mode decomposition method and the Hilbert transform method, including: Decomposing the initial acoustic wave signal set using the empirical mode decomposition method to obtain a plurality of intrinsic mode functions; Processing the plurality of intrinsic mode functions by the Hilbert transform method to obtain a plurality of initial signal instantaneous frequencies and a plurality of initial instantaneous amplitudes; The target signal sample set is determined based on the multiple initial signal instantaneous frequencies and the multiple initial instantaneous amplitudes.

3. The method according to claim 1, characterized in that Get the multi-classification Stacking ensemble model, including: Using the plurality of acoustic wave sensors to collect a historical acoustic wave signal set of the water pipeline; Based on the historical sound wave signal set, a historical signal sample set is obtained by processing the historical mode decomposition method and the Hilbert transform method; Training a preset Stacking integration model based on the historical signal sample set to obtain a first target Stacking integration model, where the first target Stacking integration model is used to distinguish pipeline leakage defects from other defects; Repeat the step of training the preset Stacking integration model based on the historical signal sample set, and obtain a second target Stacking integration model, a third target Stacking integration model, and a fourth target Stacking integration model, wherein the second target Stacking integration model is used to distinguish between pipeline rupture defects and the other defects, the third target Stacking integration model is used to distinguish between pipeline corrosion defects and the other defects, and the fourth target Stacking integration model is used to distinguish between a pipeline defect-free situation and the other defects; The multi-classification Stacking integration model is determined based on the first target Stacking integration model, the second target Stacking integration model, the third target Stacking integration model and the fourth target Stacking integration model.

4. The method according to claim 3, characterized in that The preset Stacking integration model is trained based on the historical signal sample set to obtain a first target Stacking integration model, including: Based on the historical signal sample set, the first layer classifier of the preset Stacking ensemble model is trained using the extreme gradient boosting tree algorithm, the K nearest neighbor algorithm and the decision tree algorithm to obtain a target feature data set; Based on the target feature data set, the second layer classifier of the preset Stacking integration model is trained using a logistic regression algorithm to obtain the first target Stacking integration model.

5. The method according to claim 4, characterized in that Based on the historical signal sample set, the first layer classifier of the preset Stacking ensemble model is trained using the extreme gradient boosting tree algorithm, the K nearest neighbor algorithm and the decision tree algorithm to obtain the target feature data set, including: Based on the historical signal sample set, the first layer classifier of the preset Stacking ensemble model is trained using the extreme gradient boosting tree algorithm, the K nearest neighbor algorithm and the decision tree algorithm to obtain multiple initial feature data sets; Calculating multiple feature evaluation values ​​for each initial feature data set; Calculating target weight values ​​of each initial feature data set based on the multiple feature evaluation values ​​of each initial feature data set; The target feature data set is determined based on the multiple initial feature data sets and the target weight value of each initial feature data set.

6. The method according to claim 3, characterized in that The multi-classification Stacking integrated model is used to classify the water pipeline defects corresponding to the target signal sample set to obtain the water pipeline defect classification results, including: Respectively using the first target Stacking integration model, the second target Stacking integration model, the third target Stacking integration model and the fourth target Stacking integration model to classify the water pipeline defects corresponding to the target signal sample set to obtain multiple initial classification results; Based on the multiple initial classification results, the water pipeline defect classification result is determined.

7. A water pipeline defect classification device, characterized in that: A plurality of acoustic wave sensors are arranged in the water delivery pipeline, and the distances between the acoustic wave sensors are equal; the device comprises: A collection module, used for collecting an initial acoustic wave signal set of the water pipeline using the multiple acoustic wave sensors, wherein the initial acoustic wave signal set includes multiple acoustic wave signals when the water pipeline is free of defects and has different types of defects; A processing module, used for obtaining a target signal sample set based on the initial sound wave signal set by processing the initial sound wave signal set using an empirical mode decomposition method and a Hilbert transform method; An acquisition module, used for acquiring a multi-classification Stacking integration model, where the multi-classification Stacking integration model is integrated by four target Stacking integration models for distinguishing different defects; The classification module is used to classify the water pipeline defects corresponding to the target signal sample set using the multi-classification Stacking integrated model to obtain the water pipeline defect classification result.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the water pipeline defect classification method according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the water pipeline defect classification method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the water pipeline defect classification method according to any one of claims 1 to 6.

Images