Pipeline leakage early warning method and device for multi-source heterogeneous gas pipe network data fusion

Through the support vector machine and deep convolutional neural network model, the multi-source heterogeneous gas pipeline data is preprocessed and feature extracted, and the D-S evidence theory is used to fuse, which solves the problem of leakage risk monitoring in gas pipeline data monitoring, and realizes the Effective early warning and safety improvement of gas pipeline leakage risk.

CN119084836APending Publication Date: 2024-12-06CHINA CERTIFICATION & INSPECTION GRP BEIJING CO LTD +1

Patent Information

Application Number
CN202410853691.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-28
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The prior art is difficult to effectively use multi-source heterogeneous gas pipeline data for leakage risk monitoring, resulting in the impact of the safety of the gas pipeline network.

Method used

The support vector machine model and deep convolutional neural network model are used to preprocess and feature extraction for multi-source heterogeneous data, and data fusion is combined with D-S evidence theory to achieve comprehensive decision-making on gas pipeline leakage.

Benefits of technology

Through precise feature extraction and data fusion, the uncertainty of data sources is eliminated, and an effective warning of the risk of leakage in the gas pipeline network is achieved, improving data throughput efficiency and real-time performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119084836A_ABST
    Figure CN119084836A_ABST
Patent Text Reader

Abstract

The invention provides a multi-source heterogeneous gas pipe network data fusion pipeline leakage early warning method and device, and the method comprises the steps: carrying out the preprocessing of multi-source heterogeneous data at a data level, employing different preprocessing methods for numerical data and image data, and achieving the preprocessing of the physical attribute data and image data of a gas pipe network; the feature level constructs a support vector machine model and a deep convolutional neural network model to respectively extract features of numerical data and image data, and outputs a classification probability to provide an evidence basis for a decision-making layer; and the decision-making layer carries out fusion decision-making on judgment results of the model based on a D-S evidence theory to obtain a final decision-making judgment on whether the gas pipeline leaks or not. According to the method, different feature extraction schemes are adopted for the heterogeneous data, accurate feature extraction of the multi-source heterogeneous data is achieved, the features of the heterogeneous data are fully fused through the evidence theory, uncertainty of data sources is eliminated, and finally effective early warning of potential leakage risks of the gas pipe network is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of gas pipeline safety early warning, and in particular to a pipeline leakage early warning method and device for multi-source heterogeneous gas pipeline network data fusion. Background Art

[0002] The gas pipeline network is a pipeline system used to transport gas. It can safely and efficiently transport gas from the production or storage site to the user to meet the user's gas demand. The gas pipeline network is one of the important public infrastructures in the city and an important part of the national energy construction. It is used to provide reliable, clean and safe energy supply to support economic development and energy transformation. It plays an important role in the urbanization process and energy supply, and has a positive impact on the sustainable development of society and economy. At present, China has built hundreds of thousands of kilometers of gas pipelines, covering most cities and rural areas. However, this also puts strict requirements on the safety of gas pipelines. Old gas pipelines in some areas have problems such as corrosion and leakage, which increases safety risks.

[0003] Through real-time analysis of gas pipeline network data, it is possible to effectively detect whether there is a potential leakage risk in the pipeline network. However, due to the large number of gas pipeline network equipment, a large amount of operating data is generated every moment. There are various types of pipeline sensors and multiple gas business scenarios. Multiple types of data come from multiple links and different departments, including structured data and unstructured data. How to use these multi-source heterogeneous data to achieve pipeline network safety monitoring is an urgent problem that needs to be solved. Summary of the invention

[0004] The present invention aims to provide a pipeline leakage early warning method and device for multi-source heterogeneous gas pipeline network data fusion that overcomes the above-mentioned problems or at least partially solves the above-mentioned problems.

[0005] To achieve the above object, the technical solution of the present invention is specifically implemented as follows:

[0006] One aspect of the present invention provides a pipeline leakage early warning method for multi-source heterogeneous gas pipeline network data fusion, comprising:

[0007] Acquire multi-source heterogeneous data from a gas network data center, and divide the multi-source heterogeneous data into numerical data and image data according to structural types;

[0008] Performing preprocessing operations on the numerical data and the image data respectively;

[0009] Build and train support vector machine models and deep convolutional neural network models;

[0010] Input the preprocessed numerical data into the support vector machine model to obtain the numerical data classification and discrimination results, and input the preprocessed image data into the deep convolutional neural network model to obtain the image data classification and discrimination results;

[0011] The numerical data classification and discrimination results and the image data classification and discrimination results are integrated according to the DS evidence theory to output a comprehensive decision result.

[0012] Optionally, the acquiring of multi-source heterogeneous data from a gas pipeline network data center includes:

[0013] Acquire the multi-source heterogeneous data from the gas pipeline network data center, wherein the multi-source heterogeneous data includes: gas pipeline network pressure, flow, temperature, humidity, sound, and image data;

[0014] The acquired multi-source heterogeneous data is format converted and standardized.

[0015] Optionally, dividing the multi-source heterogeneous data into numerical data and image data according to the structural type includes:

[0016] Performing feature extraction on the multi-source heterogeneous data to extract numerical features and image features;

[0017] The numerical features are combined into the numerical data, and the image features are combined into the image data.

[0018] Optionally, the preprocessing operation on the numerical data includes:

[0019] Check whether the numerical data has missing values, and if so, fill the missing values ​​with the median;

[0020] Use data normalization methods to detect whether there are outliers, and if so, handle them;

[0021] Performing standardization processing on the numerical data;

[0022] The preprocessing operation on the image data comprises:

[0023] Check the image data for damage, duplication, or incompleteness and delete low-quality or unqualified images;

[0024] Crop the image;

[0025] Data normalization converts pixel values ​​to the range [0,1].

[0026] Optionally, the constructing and training a support vector machine model includes:

[0027] Using known gas network leakage and normal numerical data as training sets and test sets, constructing and using the training sets to train the support vector machine model, using the test sets to evaluate the trained support vector machine model, and tuning the support vector machine model parameters;

[0028] The construction and training of the deep convolutional neural network model includes:

[0029] The hyperparameters of the deep convolutional neural network model are set, the deep convolutional neural network model is trained using known gas pipeline leakage and image data under normal conditions as a training set, and the weights and biases of the deep convolutional neural network model are optimized using an optimizer.

[0030] Optionally, the data classification and discrimination result and the image data classification and discrimination result, outputting a comprehensive decision result comprises:

[0031] Converting the probability output by the support vector machine model into first evidence;

[0032] Converting the probability output by the deep convolutional neural network model into a second piece of evidence;

[0033] Use Dempster combination rule to fuse classification results:

[0034] Bel(A∩B)=∑Bel(A i )*Bel(B j );.

[0035] Among them: Bel(A∩B) is the evidence of the merger, Bel(A i ) is the first evidence from the output of the support vector machine model, Bel(B j ) is the second evidence output from the deep convolutional neural network model;

[0036] Bel′(leakage) and Bel′(no leakage) are calculated by the above formula, and the result is the final probability of leakage and no leakage;

[0037] Compare Bel′(leakage) and Bel′(no leakage). If Bel′(leakage)-Bel′(no leakage)>0, make a comprehensive decision that the pipeline is leaking; if Bel′(leakage)-Bel′(no leakage)<0, make a comprehensive decision that the pipeline is not leaking; if Bel′(leakage)-Bel′(no leakage)=0, make a comprehensive decision that “it is impossible to determine whether there is a leak”.

[0038] Another aspect of the present invention provides a pipeline leakage early warning device for multi-source heterogeneous gas pipeline network data fusion, comprising:

[0039] An acquisition module, used for acquiring multi-source heterogeneous data from a gas network data center, and dividing the multi-source heterogeneous data into numerical data and image data according to structural types;

[0040] A preprocessing module, used for performing preprocessing operations on the numerical data and the image data respectively;

[0041] Training module, used to build and train support vector machine models and deep convolutional neural network models;

[0042] A discrimination module, used for inputting the preprocessed numerical data into the support vector machine model to obtain a classification discrimination result of the numerical data, and inputting the preprocessed image data into the deep convolutional neural network model to obtain a classification discrimination result of the image data;

[0043] The fusion module is used to fuse the numerical data classification and discrimination results and the image data classification and discrimination results according to the DS evidence theory, and output a comprehensive decision result.

[0044] Optionally, the acquisition module acquires multi-source heterogeneous data of the gas pipeline network data center in the following manner:

[0045] Acquire the multi-source heterogeneous data from the gas pipeline network data center, wherein the multi-source heterogeneous data includes: gas pipeline network pressure, flow, temperature, humidity, sound, and image data;

[0046] The acquired multi-source heterogeneous data is format converted and standardized.

[0047] Optionally, the acquisition module divides the multi-source heterogeneous data into numerical data and image data according to the structural type in the following manner:

[0048] Performing feature extraction on the multi-source heterogeneous data to extract numerical features and image features;

[0049] The numerical features are combined into the numerical data, and the image features are combined into the image data.

[0050] Optionally, the preprocessing module performs a preprocessing operation on the numerical data in the following manner:

[0051] Check whether the numerical data has missing values, and if so, fill the missing values ​​with the median;

[0052] Use data normalization methods to detect whether there are outliers, and if so, handle them;

[0053] Performing standardization processing on the numerical data;

[0054] The preprocessing module performs preprocessing operations on the image data in the following manner:

[0055] Check the image data for damage, duplication, or incompleteness and delete low-quality or unqualified images;

[0056] Crop the image;

[0057] Data normalization converts pixel values ​​to the range [0,1].

[0058] Optionally, the training module constructs and trains a support vector machine model in the following manner:

[0059] Using known gas network leakage and normal numerical data as training sets and test sets, constructing and using the training sets to train the support vector machine model, using the test sets to evaluate the trained support vector machine model, and tuning the support vector machine model parameters;

[0060] The training module constructs and trains a deep convolutional neural network model in the following manner:

[0061] The hyperparameters of the deep convolutional neural network model are set, the deep convolutional neural network model is trained using known gas pipeline leakage and image data under normal conditions as a training set, and the weights and biases of the deep convolutional neural network model are optimized using an optimizer.

[0062] Optionally, the fusion module fuses the numerical data classification and discrimination result and the image data classification and discrimination result according to DS evidence theory in the following manner to output a comprehensive decision result:

[0063] Converting the probability output by the support vector machine model into first evidence;

[0064] Converting the probability output by the deep convolutional neural network model into a second piece of evidence;

[0065] Use Dempster combination rule to fuse classification results:

[0066] Bel(A∩B)=∑Bel(A i )*Bel(B j );.

[0067] Among them: Bel(A∩B) is the evidence of the merger, Bel(A i ) is the first evidence from the output of the support vector machine model, Bel(B j ) is the second evidence output from the deep convolutional neural network model;

[0068] Bel′(leakage) and Bel′(no leakage) are calculated by the above formula, and the result is the final probability of leakage and no leakage;

[0069] Compare Bel′(leakage) and Bel′(no leakage). If Bel′(leakage)-Bel′(no leakage)>0, make a comprehensive decision that the pipeline is leaking; if Bel′(leakage)-Bel′(no leakage)<0, make a comprehensive decision that the pipeline is not leaking; if Bel′(leakage)-Bel′(no leakage)=0, make a comprehensive decision that “it is impossible to determine whether there is a leak”.

[0070] It can be seen that, through the pipeline leakage warning method and device for multi-source heterogeneous gas pipeline network data fusion provided by the present invention, the data layer pre-processes the multi-source heterogeneous data, and adopts different pre-processing methods for numerical data and image data to achieve pre-processing of the physical attribute data and image data of the gas pipeline network; the feature layer constructs a support vector machine model and a deep convolutional neural network model to extract the features of numerical data and image data respectively, and outputs the classification probability to provide an evidence basis for the decision layer; the decision layer is based on the DS evidence theory, and the judgment results of the model are fused and decided to obtain the final decision judgment on whether the gas pipeline is leaking. The present invention adopts different feature extraction schemes for heterogeneous data respectively, realizes the accurate feature extraction of multi-source heterogeneous data, fully integrates the features of heterogeneous data through evidence theory, eliminates the uncertainty of data sources, and finally achieves effective warning of potential risks of gas pipeline leakage. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0072] Figure 1 A flow chart of a pipeline leakage early warning method for multi-source heterogeneous gas pipeline network data fusion provided by an embodiment of the present invention;

[0073] Figure 2 A schematic diagram of the logical architecture of a pipeline leakage early warning method for multi-source heterogeneous gas pipeline network data fusion provided in an embodiment of the present invention;

[0074] Figure 3 A schematic diagram of the structure of a pipeline leakage warning device for multi-source heterogeneous gas pipeline network data fusion provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0075] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0076] Figure 1 A flow chart of a pipeline leakage early warning method for multi-source heterogeneous gas pipeline network data fusion provided by an embodiment of the present invention is shown. Figure 2 The logical architecture diagram of the pipeline leakage warning method for multi-source heterogeneous gas pipeline network data fusion provided by the embodiment of the present invention is shown in FIG. Figure 1 and Figure 2 The pipeline leakage early warning method for multi-source heterogeneous gas pipeline network data fusion provided by the embodiment of the present invention includes:

[0077] S1, obtaining multi-source heterogeneous data from a gas pipeline network data center, and dividing the multi-source heterogeneous data into numerical data and image data according to the structural type;

[0078] S2, preprocessing operations are performed on numerical data and image data respectively;

[0079] S3, builds and trains support vector machine models and deep convolutional neural network models;

[0080] S4, inputting the preprocessed numerical data into a support vector machine model to obtain a numerical data classification and discrimination result, and inputting the preprocessed image data into a deep convolutional neural network model to obtain an image data classification and discrimination result;

[0081] S5, according to the DS evidence theory, the classification and discrimination results of numerical data and image data are integrated to output a comprehensive decision result.

[0082] Specifically, the present invention proposes a pipeline leakage warning solution that integrates multi-source heterogeneous gas pipeline network data, so as to extract data with low redundancy, high fault tolerance and rich effective information from massive data, achieve consistent interpretation and description of the gas pipeline network environment, improve the data throughput efficiency and real-time performance of the gas pipeline network system, and effectively warn of potential gas pipeline leakage risks.

[0083] The present invention is mainly divided into a three-layer structure, including: a data layer extracts data and preprocesses, a feature layer extracts features and outputs judgment results, and a decision layer outputs the final decision.

[0084] As an optional implementation manner of an embodiment of the present invention, obtaining multi-source heterogeneous data from a gas pipeline network data center includes: obtaining multi-source heterogeneous data from the gas pipeline network data center, the multi-source heterogeneous data including: gas pipeline network pressure, flow, temperature, humidity, sound, and image data; and performing format conversion and standardization processing on the obtained multi-source heterogeneous data.

[0085] As an optional implementation manner of an embodiment of the present invention, dividing multi-source heterogeneous data into numerical data and image data according to structural type includes: performing feature extraction on the multi-source heterogeneous data to extract numerical features and image features; combining the numerical features into numerical data, and combining the image features into image data.

[0086] Specifically, at the data level, data such as pipeline pressure, flow, temperature, humidity, geographic location, and monitoring images are obtained from the gas pipeline company's data center, and divided into numerical data and image data according to the structural type, and different preprocessing operations are performed for the two types of data.

[0087] As an optional implementation of an embodiment of the present invention, preprocessing operations on numerical data include: checking whether the numerical data has missing values, and if so, filling the missing values ​​with the median; using a data standardization method to detect whether there are outliers, and if so, processing the outliers; standardizing the numerical data; preprocessing operations on image data include: checking whether the image data is damaged, repeated, or incomplete and deleting low-quality or unqualified images; cropping the image; converting pixel values ​​to the [0,1] interval through data standardization.

[0088] Specifically, for numerical data, the present invention first checks whether the data has missing values ​​and fills the missing values ​​with the median. Then, a data standardization method is used to detect whether there are outliers. For each eigenvalue X, calculate Where X is the original value of the feature, μ is the feature mean, and σ is the data standard deviation. According to data distribution and expert experience, the present invention determines the threshold α=0.55. When Z>α, the feature is judged as an outlier and the average value is used to replace the outlier. Finally, the Min-Max scaling method is used: Calculation Where X scaled is the scaled value, X min is the minimum value of the data, X max Is the maximum value of the data. scaled Instead of the original feature value X, the data is scaled to the interval [0,1] to facilitate model training.

[0089] The present invention first checks the image data for damage, duplication, or incompleteness, and deletes low-quality or unqualified images. The image is then cropped. Generally, the information-rich area is close to the center of the image, so the center of the image is selected as the center and the image is cropped to a size of 128*128. Finally, refer to the Min-Max scaling in the above Min-Max scaling method to convert the image pixel value to the [0,1] interval.

[0090] As an optional implementation manner of an embodiment of the present invention, constructing and training a support vector machine model includes: using numerical data of known gas pipeline leakage and normal conditions as training sets and test sets, constructing and training a support vector machine model using the training set, using the test set to evaluate the trained support vector machine model, and tuning the support vector machine model parameters; constructing and training a deep convolutional neural network model includes: setting hyperparameters of the deep convolutional neural network model, using image data of known gas pipeline leakage and normal conditions as training sets to train the deep convolutional neural network model, and using an optimizer to optimize the weights and biases of the deep convolutional neural network model.

[0091] Specifically, at the feature level, a support vector machine (SVM) model and a deep convolutional neural network (DCNN) model are constructed and trained respectively. The preprocessed numerical data are input into the SVM, and the preprocessed image data are input into the DCNN to obtain the classification judgment results of the model.

[0092] The processed image data and numerical data are divided into training set and test set in the ratio of 7:3. During the division process, it is necessary to ensure that the distribution ratio of leakage samples and non-leakage samples in the training set and the test set is the same.

[0093] The support vector machine (SVM) model and the deep convolutional neural network (DCNN) model were constructed and trained. The preprocessed numerical data were input into the SVM, and the preprocessed image data were input into the DCNN to obtain the classification judgment results of the model.

[0094] In specific implementation, the present invention constructs and trains an SVM classification model and a CNN classification model, with classification labels of "leakage" and "no leakage", and outputs corresponding pipeline leakage judgment results.

[0095] The present invention uses Python's Scikit-Learn library to establish an SVM model, and the initial parameters of the model are selected as follows: regularization parameter C = 0.1, the kernel function selects the radial basis function kernel, the kernel function coefficient γ is set to 0.1, and probability = True is set to obtain the predicted probability value. Each sample in each numerical data training set is represented as a feature vector, which is input into the model for training, and the performance of the trained model is evaluated using the test set data, and the model parameters are tuned. After optimizing the model, the classification results and accuracy of the model on the test set are obtained.

[0096] Design the network structure to establish a DCNN model. The specific structure of the network is: input layer (128, 128, 3); convolution layer 32 3*3 convolution kernels, step size 1, using ReLU activation function, output (128, 128, 32); pooling layer uses 2x2 size maximum pooling operation, step size 2, output (64, 64, 32); convolution layer 64 3*3 size convolution kernels, step size 1, using ReLU activation function, output (64, 64, 64); pooling layer uses 2x2 size maximum pooling operation, step size 2, output (32, 32, 64); convolution layer 128 3x3 size convolution kernels, step size 1 , using ReLU activation function, output (32,32,128); the pooling layer uses a 2x2 maximum pooling operation with a step size of 2 and an output of (16,16,128); the flattening layer flattens the output of the last pooling layer into a one-dimensional vector and outputs (16*16*128,), that is, a one-dimensional vector of length 32768; the fully connected layer contains 64 neurons, uses the ReLU activation function, and outputs (2,); the Dropout layer randomly drops some neurons with a probability of 0.5 to reduce overfitting; the output layer: 2 neurons, using the Softmax activation function, representing the probability distribution of two categories (leakage and non-leakage). Set the model learning rate to 0.01, the batch size of each training round to 64, input the training set data in the processed image data into the model for training, and select the Adam optimizer to optimize the model weights and biases. The optimized training model is tested on the test set to obtain the classification results and accuracy of the model.

[0097] At the decision-making level, considering the interdependence and conflict of data, as well as the uncertainty and incompleteness of data, the Dempster combination rule in evidence theory is used to fuse the recognition results of the two classifiers to obtain a comprehensive decision result as the final recognition type, specifically:

[0098]

[0099] in,

[0100] As an optional implementation of the embodiment of the present invention, according to the DS evidence theory, the classification and discrimination results of numerical data and the classification and discrimination results of image data are fused to output the comprehensive decision result, including: converting the probability output by the support vector machine model into the first evidence; converting the probability output by the deep convolutional neural network model into the second evidence; using the Dempster combination rule to fuse the classification results: Bel(A∩B)=∑Bel(A i )*Bel(B j ); where: Bel(A∩B) is the evidence for the merger, Bel(A i ) is the first evidence from the output of the support vector machine model, Bel(B j ) is the second evidence output from the deep convolutional neural network model; Bel′(leakage) and Bel′(no leakage) are calculated by the above formula, and the calculated result is the final probability of leakage and no leakage; Bel′(leakage) and Bel′(no leakage) are compared. If Bel′(leakage)-Bel′(no leakage)>0, a comprehensive decision result of pipeline leakage is made; if Bel′(leakage)-Bel′(no leakage)<0, a comprehensive decision result of no pipeline leakage is made; if Bel′(leakage)-Bel′(no leakage)=0, a comprehensive decision result of “cannot determine whether there is leakage” is made.

[0101] Specifically, the present invention combines the classification results of the classifier based on evidence theory and makes the final decision steps as follows:

[0102] 1. Convert the probabilities output by the SVM classifier and the DCNN classifier into evidence: Where E is the calculated evidence and P is the predicted probability output by each classifier.

[0103] 2. Use Dempster combination rule to fuse the classification results. The combination rule formula is: Bel(A∩B)=∑Bel(A i )*Bel(B j ). Among them, Bel(A∩B) is the evidence of the merger, Bel(A i ) and Bel(N j ) represent the evidence from the SVM classifier and the DCNN classifier respectively. Bel (leakage) and Bel (no leakage) are calculated according to the formula.

[0104] 3. To obtain the final classification decision result, normalize the evidence in step 2. The normalization formula is: Bel′ (leakage) and Bel′ (no leakage) are calculated according to the formula, and the calculation result is the final probability of leakage and no leakage.

[0105] 4. Compare the magnitudes of Bel′(leak) and Bel′(no leak). If Bel′(leak) - Bel′(no leak) > 0, then make a judgment of pipeline leakage; if Bel′(leak) - Bel′(no leak) < 0, then make a judgment that the pipeline has no leakage. Specifically, in extremely special cases, theoretically, there may be a situation where Bel′(leak) - Bel′(no leak) = 0. If such a situation occurs, then make a judgment of "unable to determine whether there is leakage".

[0106] In specific implementation, the evidence theory is used to fuse the recognition results of SVM and DCNN as the final decision. Suppose in a set of data of a gas pipeline in a certain place, the leakage probability and non - leakage probability given by the SVM classifier are P S泄漏 = 0.7, P S未泄漏 = 0.3; the leakage probability and non - leakage probability given by the DCNN classifier are P DC泄漏 = 0.6, P DC未泄漏 = 0.4.

[0107] Convert the leakage probability and non - leakage probability into evidence. Among them, the evidence of the SVM classifier is:

[0108] The evidence of the DCNN classifier is:

[0109] After calculating the evidence, use the Dempster combination rule to fuse the classification results:

[0110] Bel(leak) = E S (leak) * E DC (leak) = 3.495, Bel(no leak) = E S (no leak) * E DC (no leak) = 0.2881.

[0111] Perform a normalization operation on the fusion result:

[0112]

[0113] Since Bel′(leak) > Bel′(no leak), then issue a pipeline leakage warning, and the probability of pipeline leakage is 0.9248.

[0114] If Bel′(leak) < Bel′(no leak), then it indicates that the pipeline is normal and there is no leakage.

[0115] Theoretically, in extremely rare cases, it may occur that Bel′(leak) = Bel′(no leak). At this time, make a judgment of "unable to determine whether there is leakage".

[0116] It can be seen that the pipeline leakage warning method of multi-source heterogeneous gas pipeline network data fusion of the present invention includes preprocessing of multi-source heterogeneous data at the data level, using different preprocessing methods for numerical data and image data, and realizing preprocessing of physical attribute data and image data of the gas pipeline network; constructing a support vector machine model and a deep convolutional neural network model at the feature level to extract the features of numerical data and image data respectively, and output classification probabilities to provide an evidence basis for the decision-making layer; based on the DS evidence theory, the decision-making layer makes a fusion decision on the judgment results of the model to obtain the final decision judgment on whether the gas pipeline is leaking. The present invention adopts different feature extraction schemes for heterogeneous data respectively, realizes accurate feature extraction of multi-source heterogeneous data, fully integrates the features of heterogeneous data through evidence theory, eliminates the uncertainty of data sources, and finally realizes effective warning of potential risks of gas pipeline leakage.

[0117] Figure 3 The structure diagram of the pipeline leakage warning device for multi-source heterogeneous gas pipeline network data fusion provided by an embodiment of the present invention is shown. The pipeline leakage warning device for multi-source heterogeneous gas pipeline network data fusion applies the above method. The following is only a brief description of the structure of the pipeline leakage warning device for multi-source heterogeneous gas pipeline network data fusion. For other matters not covered, please refer to the relevant description in the pipeline leakage warning method for multi-source heterogeneous gas pipeline network data fusion, see Figure 3 The pipeline leakage early warning device for multi-source heterogeneous gas pipeline network data fusion provided by the embodiment of the present invention includes:

[0118] An acquisition module is used to acquire multi-source heterogeneous data from a gas network data center and divide the multi-source heterogeneous data into numerical data and image data according to the structural type;

[0119] A preprocessing module, used for performing preprocessing operations on numerical data and image data respectively;

[0120] Training module, used to build and train support vector machine models and deep convolutional neural network models;

[0121] The discrimination module is used to input the preprocessed numerical data into the support vector machine model to obtain the numerical data classification discrimination result, and input the preprocessed image data into the deep convolutional neural network model to obtain the image data classification discrimination result;

[0122] The fusion module is used to fuse the classification and discrimination results of numerical data and the classification and discrimination results of image data according to the DS evidence theory, and output a comprehensive decision result.

[0123] As an optional implementation of the embodiment of the present invention, the acquisition module acquires multi-source heterogeneous data of the gas pipeline network data center in the following manner:

[0124] Obtain multi-source heterogeneous data from the gas pipeline network data center, including: gas pipeline network pressure, flow, temperature, humidity, sound, and image data;

[0125] Perform format conversion and standardization on the acquired multi-source heterogeneous data.

[0126] As an optional implementation of the embodiment of the present invention, the acquisition module divides the multi-source heterogeneous data into numerical data and image data according to the structural type in the following manner:

[0127] Perform feature extraction on multi-source heterogeneous data to extract numerical features and image features;

[0128] Combine numerical features into numerical data, and combine image features into image data.

[0129] As an optional implementation of the embodiment of the present invention, the preprocessing module performs preprocessing operations on the numerical data in the following manner:

[0130] Check if there are missing values ​​in the numerical data, and fill the missing values ​​with the median if there are;

[0131] Use data normalization methods to detect whether there are outliers, and if so, handle them;

[0132] Standardize numerical data;

[0133] The preprocessing module performs preprocessing operations on image data in the following ways:

[0134] Check image data for corruption, duplication, incompleteness and remove low-quality or substandard images;

[0135] Crop the image;

[0136] Data normalization converts pixel values ​​to the range [0,1].

[0137] As an optional implementation of the embodiment of the present invention, the training module constructs and trains the support vector machine model in the following manner:

[0138] Using known gas pipeline leakage and normal numerical data as training and test sets, construct and train a support vector machine model using the training set, use the test set to evaluate the trained support vector machine model, and tune the support vector machine model parameters;

[0139] The training module builds and trains a deep convolutional neural network model in the following way:

[0140] The hyperparameters of the deep convolutional neural network model are set, the image data of known gas pipeline leakage and normal conditions are used as training sets to train the deep convolutional neural network model, and the optimizer is used to optimize the weights and biases of the deep convolutional neural network model.

[0141] As an optional implementation of the embodiment of the present invention, the fusion module fuses the numerical data classification and discrimination results and the image data classification and discrimination results according to the DS evidence theory in the following manner to output a comprehensive decision result:

[0142] Convert the probability output by the support vector machine model into the first evidence;

[0143] Convert the probability output by the deep convolutional neural network model into the second evidence;

[0144] Use Dempster combination rule to fuse classification results:

[0145] Bel(A∩B)=∑Bel(A i )*Bel(B j );.

[0146] Among them: Bel(A∩B) is the evidence of the merger, Bel(A i ) is the first evidence from the output of the support vector machine model, Bel(B j ) is the second evidence from the output of the deep convolutional neural network model;

[0147] Bel′(leakage) and Bel′(no leakage) are calculated by the above formula, and the result is the final probability of leakage and no leakage;

[0148] Compare Bel′(leakage) and Bel′(no leakage). If Bel′(leakage)-Bel′(no leakage)>0, make a comprehensive decision that the pipeline is leaking; if Bel′(leakage)-Bel′(no leakage)<0, make a comprehensive decision that the pipeline is not leaking; if Bel′(leakage)-Bel′(no leakage)=0, make a comprehensive decision that “it is impossible to determine whether there is a leak”.

[0149] It can be seen that the pipeline leakage warning device for multi-source heterogeneous gas pipeline network data fusion of the present invention includes preprocessing of multi-source heterogeneous data at the data level, using different preprocessing methods for numerical data and image data, and realizing preprocessing of physical attribute data and image data of the gas pipeline network; constructing a support vector machine model and a deep convolutional neural network model at the feature level to extract the features of numerical data and image data respectively, and output classification probabilities to provide an evidence basis for the decision-making layer; based on the DS evidence theory, the decision-making layer makes a fusion decision on the judgment results of the model to obtain the final decision judgment on whether the gas pipeline is leaking. The present invention adopts different feature extraction schemes for heterogeneous data respectively, realizes accurate feature extraction of multi-source heterogeneous data, fully integrates the features of heterogeneous data through evidence theory, eliminates the uncertainty of data sources, and finally realizes effective warning of potential risks of gas pipeline network leakage.

[0150] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A pipeline leakage warning method based on multi-source heterogeneous gas pipeline network data fusion, It is characterized in that include: Acquire multi-source heterogeneous data from a gas network data center, and divide the multi-source heterogeneous data into numerical data and image data according to structural types; Performing preprocessing operations on the numerical data and the image data respectively; Build and train support vector machine models and deep convolutional neural network models; Input the preprocessed numerical data into the support vector machine model to obtain the numerical data classification and discrimination results, and input the preprocessed image data into the deep convolutional neural network model to obtain the image data classification and discrimination results; The numerical data classification and discrimination results and the image data classification and discrimination results are integrated according to the DS evidence theory to output a comprehensive decision result.

2. The method according to claim 1, It is characterized in that The method of obtaining multi-source heterogeneous data from the gas pipeline network data center includes: Acquire the multi-source heterogeneous data from the gas pipeline network data center, wherein the multi-source heterogeneous data includes: gas pipeline network pressure, flow, temperature, humidity, sound, and image data; The acquired multi-source heterogeneous data is format converted and standardized.

3. The method according to claim 2, It is characterized in that The dividing the multi-source heterogeneous data into numerical data and image data according to the structural type comprises: Performing feature extraction on the multi-source heterogeneous data to extract numerical features and image features; The numerical features are combined into the numerical data, and the image features are combined into the image data.

4. The method according to claim 1 or 3, It is characterized in that The preprocessing operation on the numerical data comprises: Check whether the numerical data has missing values, and if so, fill the missing values ​​with the median; Use data normalization methods to detect whether there are outliers, and if so, handle them; Performing standardization processing on the numerical data; The preprocessing operation on the image data comprises: Check the image data for damage, duplication, or incompleteness and delete low-quality or unqualified images; Crop the image; Data normalization converts pixel values ​​to the range [0,1].

5. The method according to claim 1, It is characterized in that The constructing and training of the support vector machine model includes: Using known gas network leakage and normal numerical data as training sets and test sets, constructing and using the training sets to train the support vector machine model, using the test sets to evaluate the trained support vector machine model, and tuning the support vector machine model parameters; The construction and training of the deep convolutional neural network model includes: The hyperparameters of the deep convolutional neural network model are set, the deep convolutional neural network model is trained using known gas pipeline leakage and image data under normal conditions as a training set, and the weights and biases of the deep convolutional neural network model are optimized using an optimizer.

6. The method according to claim 1 or 5, It is characterized in that The output of the comprehensive decision result by fusing the numerical data classification and discrimination result and the image data classification and discrimination result according to the DS evidence theory includes: Converting the probability output by the support vector machine model into first evidence; Converting the probability output by the deep convolutional neural network model into a second piece of evidence; Use Dempster combination rule to fuse classification results: Call(A∩B)=∑Call(A i )*Call(B j );。 Among them: Bel(A∩B) is the evidence of the merger, Bel(A i ) is the first evidence from the output of the support vector machine model, Bel(B j ) is the second evidence output from the deep convolutional neural network model; Bel′(leakage) and Bel′(no leakage) are calculated by the above formula, and the result is the final probability of leakage and no leakage; Compare Bel′(leakage) and Bel′(no leakage). If Bel′(leakage)-Bel′(no leakage)>0, make a comprehensive decision that the pipeline is leaking; if Bel′(leakage)-Bel′(no leakage)<0, make a comprehensive decision that the pipeline is not leaking; if Bel′(leakage)-Bel′(no leakage)=0, make a comprehensive decision that "it is impossible to determine whether there is a leak".

7. A pipeline leakage warning device based on multi-source heterogeneous gas pipeline network data fusion, It is characterized in that include: An acquisition module, used for acquiring multi-source heterogeneous data from a gas network data center, and dividing the multi-source heterogeneous data into numerical data and image data according to structural types; A preprocessing module, used for performing preprocessing operations on the numerical data and the image data respectively; Training module, used to build and train support vector machine models and deep convolutional neural network models; A discrimination module, used for inputting the preprocessed numerical data into the support vector machine model to obtain a classification discrimination result of the numerical data, and inputting the preprocessed image data into the deep convolutional neural network model to obtain a classification discrimination result of the image data; The fusion module is used to fuse the numerical data classification and discrimination results and the image data classification and discrimination results according to the DS evidence theory, and output a comprehensive decision result.

8. The device according to claim 7, It is characterized in that The acquisition module acquires multi-source heterogeneous data from the gas pipeline network data center in the following manner: Acquire the multi-source heterogeneous data from the gas pipeline network data center, wherein the multi-source heterogeneous data includes: gas pipeline network pressure, flow, temperature, humidity, sound, and image data; The acquired multi-source heterogeneous data is format converted and standardized.

9. The device according to claim 8, It is characterized in that The acquisition module divides the multi-source heterogeneous data into numerical data and image data according to the structural type in the following manner: Performing feature extraction on the multi-source heterogeneous data to extract numerical features and image features; The numerical features are combined into the numerical data, and the image features are combined into the image data.

10. The method according to claim 7 or 9, It is characterized in that The preprocessing module performs preprocessing operations on the numerical data in the following manner: Check whether the numerical data has missing values, and if so, fill the missing values ​​with the median; Use data normalization methods to detect whether there are outliers, and if so, handle them; Performing standardization processing on the numerical data; The preprocessing module performs preprocessing operations on the image data in the following manner: Check the image data for damage, duplication, or incompleteness and delete low-quality or unqualified images; Crop the image; Data normalization converts pixel values ​​to the range [0,1].

11. The device according to claim 7, It is characterized in that The training module constructs and trains the support vector machine model in the following manner: Using known gas network leakage and normal numerical data as training sets and test sets, constructing and using the training sets to train the support vector machine model, using the test sets to evaluate the trained support vector machine model, and tuning the support vector machine model parameters; The training module constructs and trains a deep convolutional neural network model in the following manner: The hyperparameters of the deep convolutional neural network model are set, the deep convolutional neural network model is trained using known gas pipeline leakage and image data under normal conditions as a training set, and the weights and biases of the deep convolutional neural network model are optimized using an optimizer.

12. The device according to claim 7 or 11, It is characterized in that The fusion module fuses the numerical data classification and discrimination results and the image data classification and discrimination results according to the DS evidence theory in the following manner to output a comprehensive decision result: Converting the probability output by the support vector machine model into first evidence; Converting the probability output by the deep convolutional neural network model into a second piece of evidence; Use Dempster combination rule to fuse classification results: Call(A∩B)=∑Call(A i )*Call(B j );。 Among them: Bel(A∩B) is the evidence of the merger, Bel(A i ) is the first evidence from the output of the support vector machine model, Bel(B j ) is the second evidence output from the deep convolutional neural network model; Bel′(leakage) and Bel′(no leakage) are calculated by the above formula, and the result is the final probability of leakage and no leakage; Compare Bel′(leakage) and Bel′(no leakage). If Bel′(leakage)-Bel′(no leakage)>0, make a comprehensive decision that the pipeline is leaking; if Bel′(leakage)-Bel′(no leakage)<0, make a comprehensive decision that the pipeline is not leaking; if Bel′(leakage)-Bel′(no leakage)=0, make a comprehensive decision that "it is impossible to determine whether there is a leak".

Citation Information

Patent Citations

  • Photovoltaic array fault diagnosis method based on composite information

    CN108647716A

  • Fused partial discharge type identification method based on DS evidence theory

    CN112686093A

  • Rainfall runoff prediction method based on multi-modal fusion

    CN117035148A

Cited By

  • Fuel gas monitoring fault analysis method and system and storage medium

    CN119961834A

  • Feature-level gas leakage prediction and early warning method based on multi-task neural network

    CN120873798A

  • A feature-level gas leakage prediction and early warning method based on a multi-task neural network

    CN120873798B