A method and system for pipeline signal recognition based on VMD and multi-feature fusion

Abstract

A pipeline signal recognition method and system based on VMD and multi-feature fusion is disclosed, relating to the field of pipeline signal recognition technology, to address the problem of poor pipeline signal feature extraction affecting the accuracy of pipeline leak detection in existing technologies. The key technical points of this invention include: preprocessing the pipeline signal using a VMD algorithm, proposing a multi-angle distance measurement algorithm—WCC algorithm—and using weighted correlation coefficients and cosine values ​​to determine the VMD decomposition layer number K; determining the characteristic mode components based on the similarity between the IMFs components and the original signal, and extracting feature parameters of the characteristic modes, including entropy features, waveform parameters, and time-frequency domain features; constructing a feature vector group based on multi-feature fusion using the extracted feature parameters; and inputting the feature vector group into a pre-trained neural network for pipeline signal condition identification. This invention effectively improves the accuracy of signal feature extraction, thereby improving the accuracy of pipeline signal recognition.

Description

Technical Field

[0001] This invention relates to the field of pipeline signal recognition technology, specifically to a pipeline signal recognition method and system based on VMD and multi-feature fusion. Background Technology

[0002] Pipeline transportation, due to its unique advantages in transporting liquids, gases, and slurries, has been widely used in various industries. Pipeline transportation has become one of the five major modes of transportation in my country. However, pipeline leaks are difficult to prevent due to corrosion, aging, cathodic protection failure, natural disasters, construction work, and human-caused damage such as theft. Pipeline leaks not only affect the normal operation of pipeline transportation, causing environmental pollution and resource waste, but also seriously impact people's normal lives, threatening their lives and causing severe property damage. Therefore, employing appropriate pipeline leak detection technologies to monitor pipelines, prevent leaks, and promptly detect and accurately locate leaks can effectively reduce environmental pollution and economic losses.

[0003] Pipeline signal processing is a nonlinear signal processing problem. Commonly used nonlinear signal processing methods include wavelet transform, singular value denoising, EMD decomposition, and variational mode decomposition. Although EMD and VMD methods have been increasingly applied to pipeline leak detection in recent years, both have certain limitations. After adaptively decomposing the signal into a sum of several intrinsic mode functions, the EMD algorithm is prone to problems such as under-envelope, over-envelope, power-off effects, and modal chaos. VMD can avoid the problems of EMD decomposition, but it requires pre-setting some parameters before decomposition, and the values ​​of these parameters affect the effectiveness of VMD decomposition.

[0004] In the process of leak detection, extracting the feature information of the signal plays a key role in improving the accuracy of pipeline leak detection. Therefore, extracting effective features from pipeline signals is of great significance for reducing the false alarm rate. Summary of the Invention

[0005] In view of the above problems, this invention proposes a pipeline signal recognition method and system based on VMD and multi-feature fusion to solve the problem that the pipeline leakage detection accuracy is affected by the poor pipeline signal feature extraction effect in the prior art.

[0006] According to one aspect of the present invention, a pipeline signal recognition method based on VMD and multi-feature fusion is provided, the method comprising the following steps:

[0007] Step 1: Collect pipeline signals under different operating conditions and construct a dataset;

[0008] Step 2: Perform variational mode decomposition on the pipeline signal in the dataset to obtain multiple intrinsic mode components; wherein, in the variational mode decomposition, the optimal number of decomposition layers is determined by using a multi-angle measurement distance algorithm;

[0009] Step 3: Compare the similarity between each intrinsic mode component and the original pipeline signal to determine the characteristic components;

[0010] Step 4: Calculate the entropy features, waveform features, and time-frequency domain features of the feature components to obtain a feature vector set based on multi-feature fusion;

[0011] Step 5: Input the feature vector group into the probabilistic neural network for training to obtain a trained pipeline signal recognition model based on the probabilistic neural network.

[0012] Step 6: Input the pipeline signal to be identified into the trained pipeline signal recognition model based on probabilistic neural network for recognition and obtain the recognition result.

[0013] Furthermore, the operating condition types mentioned in step 1 include three types: normal, knocking, and leakage.

[0014] Furthermore, the specific process of determining the optimal number of decomposition levels in variational mode decomposition using the multi-angle distance measurement algorithm in step 2 includes: determining the number of decomposition levels K in variational mode decomposition using weighted correlation coefficients and cosine values, initializing the number of decomposition levels K, and calculating the WCC values ​​between adjacent modes corresponding to the number of decomposition levels K. The formula for calculating the WCC value is as follows:

[0015] WCC=λCC+ηCD

[0016] Where CC represents the correlation coefficient, CD represents the cosine distance, and λ and η are weighting coefficients;

[0017] When the WCC value is not greater than the preset maximum threshold, the number of decomposition layers K is increased by 1; the above process is repeated, that is, the WCC value between adjacent modes corresponding to the number of decomposition layers K+1 is calculated; until the WCC value is greater than the preset maximum threshold, let the number of decomposition layers at this time be K+N, then the optimal number of decomposition layers is K+N-1, where N represents a natural number.

[0018] Furthermore, the formulas for calculating the weighting coefficients λ and η in step 2 are as follows:

[0019]

[0020]

[0021] in, The coefficient of variation represents the correlation coefficient. S CC μ is the standard deviation of the correlation coefficient.CC The mean of the correlation coefficients; The coefficient of variation represents the cosine distance. S CD μ is the standard deviation of the cosine distance. CD The mean of the cosine distance.

[0022] Furthermore, in step 4, the entropy feature is the scatter entropy, the waveform feature is the margin factor, and the time-frequency domain feature is the standard deviation.

[0023] According to another aspect of the present invention, a pipeline signal recognition system based on VMD and multi-feature fusion is provided, the system comprising:

[0024] The signal acquisition module is configured to collect pipeline signals under different operating conditions and build a dataset.

[0025] The variational mode decomposition module is configured to perform variational mode decomposition on the pipeline signal in the dataset to obtain multiple intrinsic mode components; wherein, in the variational mode decomposition, a multi-angle measurement distance algorithm is used to determine the optimal number of decomposition layers in the variational mode decomposition;

[0026] The feature component determination module is configured to compare the similarity between each intrinsic mode component and the original pipeline signal to determine the feature components.

[0027] The feature vector determination module is configured to calculate the entropy features, waveform features, and time-frequency domain features of the feature components to obtain a feature vector set based on multi-feature fusion.

[0028] The recognition model training module is configured to input the feature vector group into a probabilistic neural network for training, thereby obtaining a trained pipeline signal recognition model based on the probabilistic neural network.

[0029] The signal recognition module is configured to input the pipeline signal to be recognized into a pre-trained pipeline signal recognition model based on a probabilistic neural network for recognition, and obtain the recognition result.

[0030] Furthermore, the operating condition types in the signal acquisition module include three operating conditions: normal, knocking, and leakage.

[0031] Furthermore, the specific process of determining the optimal number of decomposition layers in the variational mode decomposition module using the multi-angle distance measurement algorithm includes: determining the number of decomposition layers K in the variational mode decomposition using the weighted correlation coefficient and cosine value index, initializing the number of decomposition layers K, and calculating the WCC value between adjacent modes corresponding to the number of decomposition layers K. The WCC value calculation formula is as follows:

[0032] WCC=λCC+ηCD

[0033] Where CC represents the correlation coefficient, CD represents the cosine distance, and λ and η are weighting coefficients;

[0034] When the WCC value is not greater than the preset maximum threshold, the number of decomposition layers K is increased by 1; the above process is repeated, that is, the WCC value between adjacent modes corresponding to the number of decomposition layers K+1 is calculated; until the WCC value is greater than the preset maximum threshold, let the number of decomposition layers at this time be K+N, then the optimal number of decomposition layers is K+N-1, where N represents a natural number.

[0035] Furthermore, the calculation formulas for the weighting coefficients λ and η in the variational mode decomposition module are as follows:

[0036]

[0037]

[0038] in, The coefficient of variation represents the correlation coefficient. S CC μ is the standard deviation of the correlation coefficient. CC The mean of the correlation coefficients; The coefficient of variation represents the cosine distance. S CD μ is the standard deviation of the cosine distance. CD The mean of the cosine distance.

[0039] Furthermore, in the feature vector determination module, the entropy feature is the scatter entropy, the waveform feature is the margin factor, and the time-frequency domain feature is the standard deviation.

[0040] The beneficial technical effects of this invention are:

[0041] This invention proposes a pipeline signal recognition method and system based on VMD and multi-feature fusion. First, the pipeline signal is preprocessed using the VMD algorithm. During this process, a multi-angle distance measurement algorithm—WCC (Weighting of Correlation Coefficient and Cosine Distance)—is proposed. The optimal decomposition level K for VMD is determined using weighted correlation coefficients and cosine values. Then, characteristic mode components (IMFs) are determined based on their similarity to the original signal. Feature parameters of these characteristic modes are extracted, including entropy features, waveform parameters, and time-frequency domain features, with scatter entropy, margin factor, and standard deviation selected respectively. A high-dimensional feature vector set based on multi-feature fusion is constructed using the extracted feature parameters. Finally, the feature vector set is input into a PNN for pipeline signal condition recognition. This invention effectively improves the accuracy of signal feature extraction, thereby enhancing the accuracy of pipeline signal recognition. Attached Figure Description

[0042] The present invention can be better understood by referring to the description given below in conjunction with the accompanying drawings, which together with the following detailed description are included in and form part of this specification, and are used to further illustrate preferred embodiments of the invention and explain the principles and advantages of the invention.

[0043] Figure 1 This is a flowchart of a pipeline signal recognition method based on VMD and multi-feature fusion according to an embodiment of the present invention;

[0044] Figure 2 This is a flowchart illustrating the determination of the optimal decomposition level K based on the WCC value in an embodiment of the present invention;

[0045] Figure 3 This is a schematic diagram of the PNN network structure in an embodiment of the present invention;

[0046] Figure 4 This is a schematic diagram of the structure of a pipeline signal recognition system based on VMD and multi-feature fusion according to an embodiment of the present invention. Detailed Implementation

[0047] To enable those skilled in the art to better understand the present invention, exemplary embodiments or examples of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described embodiments or examples are merely some, not all, of the embodiments or examples of the present invention. All other embodiments or examples obtained by those skilled in the art based on the embodiments or examples of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0048] This invention proposes a pipeline signal recognition method based on VMD and multi-feature fusion, such as... Figure 1 As shown, the method includes the following steps:

[0049] Step 1: Collect pipeline signals under different operating conditions and construct a dataset;

[0050] Step 2: Perform variational mode decomposition on the pipeline signal in the dataset to obtain multiple intrinsic mode components; wherein, in the variational mode decomposition, the optimal number of decomposition layers is determined by using a multi-angle measurement distance algorithm;

[0051] Step 3: Compare the similarity between each intrinsic mode component and the original pipeline signal to determine the characteristic components;

[0052] Step 4: Calculate the entropy features, waveform features, and time-frequency domain features of the feature components to obtain a feature vector set based on multi-feature fusion;

[0053] Step 5: Input the feature vector group into the probabilistic neural network for training to obtain a trained pipeline signal recognition model based on the probabilistic neural network.

[0054] Step 6: Input the pipeline signal to be identified into the trained pipeline signal recognition model based on probabilistic neural network for recognition and obtain the recognition result.

[0055] In this embodiment, preferably, the operating condition types include three types: normal, knocking, and leakage.

[0056] In this embodiment, preferably, the specific process of determining the optimal number of decomposition layers in variational mode decomposition using the multi-angle distance measurement algorithm in step 2 includes: determining the number of decomposition layers K in variational mode decomposition using weighted correlation coefficients and cosine values, initializing the number of decomposition layers K, and calculating the WCC values ​​between adjacent modes corresponding to the number of decomposition layers K. The WCC value calculation formula is as follows:

[0057] WCC=λCC+ηCD

[0058] Where CC represents the correlation coefficient, CD represents the cosine distance, and λ and η are weighting coefficients;

[0059] When the WCC value is not greater than the preset maximum threshold, the number of decomposition layers K is incremented by 1, and the above process is repeated, that is, the WCC value between adjacent modes corresponding to the number of decomposition layers K+1 is calculated; until the WCC value is greater than the preset maximum threshold, let the number of decomposition layers at this time be K+N, then the optimal number of decomposition layers is K+N-1, where N represents a natural number.

[0060] In this embodiment, preferably, the calculation formulas for the weighting coefficients λ and η in step 2 are as follows:

[0061]

[0062]

[0063] in, The coefficient of variation represents the correlation coefficient. S CC μ is the standard deviation of the correlation coefficient. CC The mean of the correlation coefficients; The coefficient of variation represents the cosine distance. S CD μ is the standard deviation of the cosine distance. CD The mean of the cosine distance.

[0064] Another embodiment of the present invention proposes a pipeline signal recognition method based on VMD and multi-feature fusion, comprising the following steps:

[0065] Step 1: Collect three types of pipeline signals—normal, knocking, and leaking—using an acoustic sensor to construct a dataset.

[0066] Step 2: Use the WCC value to determine the number of decomposition layers K in the VMD algorithm.

[0067] According to an embodiment of the present invention, a WCC algorithm is proposed in step 2. This algorithm uses a weighted correlation coefficient and a cosine value index to determine the VMD decomposition layer number K. Different K values ​​are set for VMD decomposition, with the penalty factor α defaulting to 2000. The WCC values ​​between adjacent modes under different K values ​​are calculated. The specific formula is shown below.

[0068] WCC=λCC+ηCD

[0069] Where CC is the correlation coefficient, CD is the cosine distance, and λ and η are weighting coefficients, as shown in the following formula:

[0070]

[0071] in, The coefficient of variation of the correlation coefficient. S is the coefficient of variation of the cosine distance. CC μ is the standard deviation of the correlation coefficient. CC S is the mean of the correlation coefficients. CD μ is the standard deviation of the cosine distance. CD The mean of the cosine distance.

[0072] A smaller WCC value between adjacent modes indicates a lower similarity between the two modes; a larger WCC value indicates a higher similarity between the two modes, which may also indicate that VMD decomposition is over-decomposed. According to correlation coefficient theory, a correlation coefficient between 0.1 and 0.3 indicates a weak correlation, 0.3 to 0.8 indicates a moderate correlation, and 0.8 to 1 indicates a strong correlation. Based on correlation coefficient theory and extensive experimental evidence, when the WCC value is greater than 0.8, it indicates that VMD decomposition is over-decomposed, and the optimal K value is K-1. This allows us to determine the number of decomposition layers K in VMD decomposition. The specific flowchart is as follows... Figure 2 As shown.

[0073] Step 3: The pipeline signal is decomposed into several IMFs by the VMD algorithm. The IMFs components are more regular and the signal characteristics are more obvious than the original signal. By comparing the similarity between each modal component and the original signal, the characteristic components are determined.

[0074] Step 4: Calculate the entropy features, waveform features, and time-frequency domain features of the feature components to form a feature vector group based on multi-feature fusion.

[0075] According to an embodiment of the present invention, when analyzing the characteristic parameters of the pipeline signal in step 4, the commonly used characteristic parameters can be roughly divided into the following three categories: time-frequency domain characteristics, entropy characteristics, and waveform characteristics.

[0076] (1) Time-frequency domain characteristics: The waveform characteristics of the pipeline signal can be represented by analyzing the time-frequency domain characteristics of the pipeline signal. For example, the mean can measure the average distribution trend of the pipeline data; the peak value is obtained by using the difference between the maximum and minimum values ​​in the pipeline data, which can reflect the fluctuation of the signal; the root mean square is used to calculate the vibration energy of the pipeline data, and its value affects the magnitude of the energy; the standard deviation can measure the degree of dispersion between pipeline signals. In this embodiment of the invention, the standard deviation is selected as the characteristic parameter.

[0077] (2) Entropy characteristics: Entropy characteristic parameters can measure the complexity of pipeline data. Pipeline data has the characteristics of nonlinearity and complexity. Entropy characteristic parameters can be used to measure the randomness inside the entire system. Therefore, entropy characteristic parameters can be used as characteristic indicators to measure the complexity inside pipeline data. In this embodiment of the invention, scattered entropy is selected as the characteristic parameter.

[0078] (3) Waveform characteristics: When the three operating conditions of the pipeline signal are collected, the flow velocity inside the pipe is different, which will lead to a large difference in the waveform of the three operating conditions. Therefore, the waveform characteristics of the pipeline signal can effectively reflect different operating conditions, and the characteristics are dimensionless and will not change due to changes in external physical quantities. The margin factor can reflect the different impact characteristics of the signal due to drastic changes. In this embodiment of the invention, the margin factor is selected as the characteristic parameter.

[0079] Step 5: Build a model for PNN using the extracted feature vectors and perform recognition and classification.

[0080] According to embodiments of the present invention, probabilistic neural networks (PNNs) are developed based on the Bayes minimum risk criterion. They utilize the Parzen window method to estimate rate density, have a relatively simple architecture, and possess the energy efficiency for classification and feedback, thus being widely used. The powerful nonlinear classification capabilities of PNN networks enable them to play an important role in classification. In essence, they utilize sample mapping capabilities to form a network diagnostic system with strong adaptability and fault tolerance.

[0081] PNN (Programmable Neural Network) networks are a type of supervised learning network that includes a competitive layer and radial basis layers. A four-layer feedforward neural network composed of a certain number of Gaussian functions, this network is developed based on radial basis neural networks. It mainly consists of an input layer, a pattern layer, a summation layer, and an output layer. The first layer is the input layer, which receives the feature vectors of the training samples; the second layer is the pattern layer, which calculates the distance between the input vector and the training samples, and the distance result represents the proximity between the output vector and the training samples; the third layer is the summation layer, which connects the input vectors of the second layer and represents the output of the neural network as a probability vector; the fourth layer is the output layer, which outputs the signals corresponding to different states of each neuron. Its general framework is as follows: Figure 3 As shown.

[0082] Another embodiment of the present invention proposes a pipeline signal recognition system based on VMD and multi-feature fusion, such as... Figure 4 As shown, the system includes:

[0083] Signal acquisition module 10 is configured to collect pipeline signals under different operating conditions and construct a dataset;

[0084] Variational mode decomposition module 20 is configured to perform variational mode decomposition on the pipeline signal in the dataset to obtain multiple intrinsic mode components; wherein, in the variational mode decomposition, a multi-angle measurement distance algorithm is used to determine the optimal number of decomposition layers in the variational mode decomposition;

[0085] The feature component determination module 30 is configured to compare the similarity between each intrinsic mode component and the original pipeline signal to determine the feature components.

[0086] The feature vector determination module 40 is configured to calculate the entropy features, waveform features, and time-frequency domain features of the feature components to obtain a feature vector group based on multi-feature fusion.

[0087] The recognition model training module 50 is configured to input the feature vector group into a probabilistic neural network for training to obtain a trained pipeline signal recognition model based on the probabilistic neural network.

[0088] The signal recognition module 60 is configured to input the pipeline signal to be recognized into a pre-trained pipeline signal recognition model based on a probabilistic neural network for recognition, and obtain the recognition result.

[0089] In this embodiment, preferably, the operating condition types in the signal acquisition module 10 include three operating conditions: normal, knocking, and leakage.

[0090] In this embodiment, preferably, the specific process of determining the optimal number of decomposition layers in the variational mode decomposition using the multi-angle distance measurement algorithm in the variational mode decomposition module 20 includes: determining the number of decomposition layers K in the variational mode decomposition using the weighted correlation coefficient and cosine value index, initializing the number of decomposition layers K, and calculating the WCC value between adjacent modes corresponding to the number of decomposition layers K. The WCC value calculation formula is as follows:

[0091] WCC=λCC+ηCD

[0092] Where CC represents the correlation coefficient, CD represents the cosine distance, and λ and η are weighting coefficients;

[0093] When the WCC value is not greater than the preset maximum threshold, the number of decomposition layers K is incremented by 1, and the above process is repeated, that is, the WCC value between adjacent modes corresponding to the number of decomposition layers K+1 is calculated; until the WCC value is greater than the preset maximum threshold, let the number of decomposition layers at this time be K+N, then the optimal number of decomposition layers is K+N-1, where N represents a natural number.

[0094] In this embodiment, preferably, the calculation formulas for the weighting coefficients λ and η in the variational mode decomposition module 20 are as follows:

[0095]

[0096]

[0097] in, The coefficient of variation represents the correlation coefficient. S CC μ is the standard deviation of the correlation coefficient. CC The mean of the correlation coefficients; The coefficient of variation represents the cosine distance. S CD μ is the standard deviation of the cosine distance. CD The mean of the cosine distance.

[0098] In this embodiment, preferably, the entropy feature in the feature vector determination module 40 is the scatter entropy, the waveform feature is the margin factor, and the time-frequency domain feature is the standard deviation.

[0099] The functionality of the pipeline signal recognition system based on VMD and multi-feature fusion described in this embodiment can be explained by the aforementioned pipeline signal recognition method based on VMD and multi-feature fusion. Therefore, for the parts not detailed in this embodiment, please refer to the above method embodiments, and they will not be repeated here.

[0100] Although the invention has been described with respect to a limited number of embodiments, those skilled in the art will understand from the foregoing description that other embodiments are conceivable within the scope of the invention described herein. The disclosure of the invention is illustrative and not restrictive, and the scope of the invention is defined by the appended claims.

Claims

1. A pipeline signal recognition method based on VMD and multi-feature fusion, characterized in that, The method comprises the following steps: Step 1, collecting pipeline signals of different working condition types to construct a data set; Step 2, performing variational modal decomposition on the pipeline signals in the data set to obtain multiple intrinsic modal components; wherein, in the variational modal decomposition, a multi-angle measurement distance algorithm is used to determine the optimal decomposition layer number in the variational modal decomposition, comprising: using a weighted correlation coefficient and a cosine value index to determine the decomposition layer number K in the variational modal decomposition, initializing the decomposition layer number K, calculating the WCC value between adjacent modes corresponding to the decomposition layer number K, and the WCC value calculation formula is as follows: ； wherein CC denotes a correlation coefficient, CD denotes a cosine distance, and is a weighting coefficient; When the WCC value is not greater than a preset maximum threshold, the decomposition layer number K is increased by 1; the above process is repeated, that is, the WCC value between adjacent modes corresponding to the decomposition layer number K+1 is calculated; until the WCC value is greater than the preset maximum threshold, assuming that the decomposition layer number at this time is K+N, then the optimal decomposition layer number is K+N-1, and N represents a natural number; Step 3, comparing the similarity of each intrinsic modal component and the original pipeline signal to determine a feature component; Step 4, calculating the entropy feature, waveform feature and time-frequency domain feature of the feature component to obtain a feature vector group based on multi-feature fusion; Step 5, inputting the feature vector group into a probability neural network for training to obtain a trained pipeline signal recognition model based on the probability neural network; Step 6, inputting a pipeline signal to be identified into the trained pipeline signal recognition model based on the probability neural network for identification to obtain an identification result.

2. The pipeline signal recognition method based on VMD and multi-feature fusion according to claim 1, characterized in that, The working condition types in step 1 include normal, knocking and leakage.

3. The pipeline signal recognition method based on VMD and multi-feature fusion according to claim 2, characterized in that, The weighting factor in step 2 and The calculation formula is as follows: ； ； wherein, the coefficient of variation of the correlation coefficient, , the standard deviation of the correlation coefficient, the mean of the correlation coefficient; the coefficient of variation of the cosine distance, , the standard deviation of the cosine distance, the mean of the cosine distance.

4. The pipeline signal recognition method based on VMD and multi-feature fusion according to claim 3, characterized in that, The entropy feature in step 4 uses scatter entropy, the waveform feature uses a margin factor, and the time-frequency domain feature uses a standard deviation.

5. A pipeline signal recognition system based on VMD and multi-feature fusion, characterized in that, It comprises: A signal acquisition module configured to collect pipeline signals of different working condition types to construct a data set; A variational modal decomposition module configured to perform variational modal decomposition on the pipeline signals in the data set to obtain multiple intrinsic modal components; wherein, in the variational modal decomposition, a multi-angle measurement distance algorithm is used to determine the optimal decomposition layer number in the variational modal decomposition, comprising: using a weighted correlation coefficient and a cosine value index to determine the decomposition layer number K in the variational modal decomposition, initializing the decomposition layer number K, calculating the WCC value between adjacent modes corresponding to the decomposition layer number K, and the WCC value calculation formula is as follows: ； wherein CC denotes a correlation coefficient, CD denotes a cosine distance, and is a weighting coefficient; When the WCC value is not greater than a preset maximum threshold, the decomposition layer number K is increased by 1; the above process is repeated, that is, the WCC value between adjacent modes corresponding to the decomposition layer number K+1 is calculated; until the WCC value is greater than the preset maximum threshold, assuming that the decomposition layer number at this time is K+N, then the optimal decomposition layer number is K+N-1, and N represents a natural number; A feature component determination module configured to compare the similarity of each intrinsic modal component and the original pipeline signal to determine a feature component; A feature vector determination module configured to calculate the entropy feature, waveform feature and time-frequency domain feature of the feature component to obtain a feature vector group based on multi-feature fusion; An identification model training module configured to input the feature vector group into a probability neural network for training to obtain a trained pipeline signal recognition model based on the probability neural network; The signal recognition module is configured to input a pipe signal to be recognized into a trained pipe signal recognition model based on a probabilistic neural network for recognition to obtain a recognition result.

6. The pipeline signal identification system based on VMD and multi-feature fusion according to claim 5, characterized in that, The working condition types in the signal acquisition module include three working conditions of normal, knocking and leakage.

7. The pipeline signal identification system based on VMD and multi-feature fusion according to claim 6, characterized in that, The calculation formula of the weighting coefficient in the variational modality decomposition module and The calculation formula is as follows: ； ； wherein, the coefficient of variation of the correlation coefficient, , the standard deviation of the correlation coefficient, the mean of the correlation coefficient; the coefficient of variation of the cosine distance, , the standard deviation of the cosine distance, the mean of the cosine distance.

8. The pipeline signal identification system based on VMD and multi-feature fusion according to claim 7, characterized in that, In the feature vector determination module, the entropy feature adopts a scatter entropy, the waveform feature adopts a margin factor, and the time-frequency domain feature adopts a standard deviation.

Images