A fluid pipeline leakage detection method based on improved fuzzy twin support vector machine

Abstract

The invention belongs to the field of fluid pipeline leakage detection and relates to a fluid pipeline leakage detection method based on an improved fuzzy twin support vector machine, comprising: obtaining vibration and sound signals of the fluid pipeline at different leakage degrees; extracting features of the vibration and sound signals at different leakage degrees, and normalizing the extracted feature data; inputting the normalized data into an improved fuzzy twin support vector machine model to obtain pipeline leakage detection results; the invention utilizes the spatial relative distance ratio of different samples to divide the sample space into regions, introduces the fuzzy data set theory in the noise and support vector aliasing regions to distinguish them, and re-assigns the membership of normal and noise data regions, thereby reducing the influence of noise data on the identification model and solving the problem that the FTSVM cannot distinguish between support vectors and noise data, resulting in poor model robustness.

Description

Technical Field

[0001] The invention belongs to the field of fluid pipeline leakage detection and relates to a fluid pipeline leakage detection method based on an improved fuzzy twin support vector machine. Background Art

[0002] Pipelines are important means of transportation for fluid media such as natural gas, oil and tap water, and are also important urban infrastructure. However, due to the corrosion, cracks, dents of pipeline materials, human damage and environmental factors, pipeline leakage accidents occur from time to time, which not only affects people's normal life, but also may cause huge economic losses and environmental disasters. Therefore, real-time monitoring of pipeline operation status is very important for maintaining safe operation of pipeline networks, avoiding waste of resources and achieving sustainable production. At present, there are two main problems in leakage detection of fluid pipelines. On the one hand, with the continuous expansion of pipeline scale, the amount of data in the fluid pipeline detection system has increased significantly, and the traditional identification model has low leakage identification efficiency when facing a large number of data sets, which affects the timeliness of leakage detection. On the other hand, when collecting pipeline leakage signals, due to the influence of environmental or human factors, the leakage sound signal always contains some noise data. However, due to its poor robustness, the general identification model cannot correctly distinguish between noise signals and normal leakage signals, resulting in a decrease in the accuracy of leakage identification.

[0003] At present, in the field of fluid pipeline leakage identification, researchers usually use data-driven intelligent diagnosis methods to classify and identify the fault features after leakage signal processing, and often use machine learning methods such as support vector machine (SVM), artificial neural network (ANN), k-nearest neighbor (KNN), etc. Among them, SVM and its improved version are an intelligent diagnosis method based on statistical theory, with a solid theoretical foundation and excellent performance, and have significant advantages in nonlinear and high-dimensional pattern recognition. Fuzzy twin support vector machine (FTSVM) is an effective machine learning model. By introducing fuzzy membership function, it can overcome the influence of abnormal data on identification results. In addition, because the model solves two smaller-scale quadratic programming problems, its computational efficiency is significantly improved compared with SVM, which is about 4 times faster than SVM.

[0004] However, FTSVM still has the following two problems when facing large and noisy data sets:

[0005] (1) Although the FTSVM recognition model improves computational efficiency by generating two classification hyperplanes, the model itself still needs to solve two quadratic programming problems, and its computational efficiency needs to be further improved.

[0006] (2) In FTSVM, the membership function in the sample space describes the space between the input data and the class center, ignoring the position of the sample data in the feature space and simply misclassifying the marginal support vectors as noise data, which limits the robustness of the model. Summary of the invention

[0007] In order to solve the problems existing in the above-mentioned prior art, the present invention proposes a fluid pipeline leakage detection method based on an improved fuzzy twin support vector machine, the method comprising: obtaining vibration and sound signals of the fluid pipeline at different leakage degrees; performing feature extraction on the vibration and sound signals at different leakage degrees, and normalizing the extracted feature data; aggregating the normalized feature data, and dividing the set into a training set and a test set; constructing an improved fuzzy twin support vector machine model, inputting the data in the training set into the improved fuzzy twin support vector machine model for training, and using the data in the test set to test the trained improved fuzzy twin support vector machine model to obtain the optimal improved fuzzy twin support vector machine model; obtaining the vibration and sound signal to be detected, and inputting the vibration and sound signal to be detected into the trained improved fuzzy twin support vector machine model to obtain the pipeline leakage detection result.

[0008] Preferably, the training process of the improved fuzzy twin support vector machine model includes:

[0009] Step 1: Obtain sample features of vibration-generated acoustic signals of different leakage degrees; calculate the relative distance ratio of two samples;

[0010] Step 2: Divide the samples into regions according to the relative distance ratio;

[0011] Step 3: According to the divided areas, the improved fuzzy membership function is used to weight the sample features to obtain a weight matrix with different leakage states;

[0012] Step 4: Use the least squares method and weight matrix to construct the objective function of the least squares FTSVM identification model;

[0013] Step 5: Solve the objective function to obtain the optimal hyperplane parameters corresponding to each type of sample;

[0014] Step 6: Construct a decision function based on the optimal hyperplane parameters;

[0015] Step 7: Input the test sample data into the decision function to obtain the pipeline leakage identification result and complete the model training.

[0016] Preferably, the formula for calculating the relative distance ratio of two samples is:

[0017]

[0018] Among them, x c+ represents the center of the positive sample class, l + represents the number of positive samples, x i represents the sample data set, x c- represents the center of the negative sample class, l - represents the number of negative samples, α + Represents the relative distance ratio of positive samples, δ represents a constant greater than 0 to prevent the denominator from being 0, α - Represents the relative distance ratio of negative samples.

[0019] Preferably, the improved fuzzy membership function expression is:

[0020] When the sample is in the normal data area, the fuzzy membership function is:

[0021] s i =(1-q)(1-||x i -x c+ || / r + +δ)

[0022] When the sample is a noise data area, the fuzzy membership function is:

[0023]

[0024] When the sample is a mixed area of ​​noise and support vector, the fuzzy membership function is:

[0025]

[0026] Among them, s i Represents the specific fuzzy membership value of a single sample; x i is the i-th sample data; q is a constant value, x c+ is the sample center of the positive sample; r + =max{x i -x c+ |||y i = +1} is the maximum radius of the positive sample; α i is the relative distance value between different classes constructed based on the centers of positive and negative samples; μ i is the membership calculation value; v i Calculates the non-membership value.

[0027] The expression of the objective function of the least squares FTSVM identification model is:

[0028]

[0029] st-(K(X2,A t )w1+e2b1)+ξ=e2

[0030]

[0031] st-(K(X1,A t )w2+e1b2)+η=e1

[0032] Among them, X1 and X2 correspond to positive and negative sample data respectively; w i and b i are weights and biases respectively; A t =[X1;X2]; ξ and η represent slack variables; C1, C2, C3, C4 are penalty coefficients; K is the RBF kernel function; e1 and e2 are unit vectors.

[0033] Preferably, solving the objective function includes: substituting the constraint condition into the objective function to obtain the objective function to be solved; and calculating w i and b i The parameters are differentiated to obtain the parameters of the hyperplane.

[0034] Furthermore, the hyperplane parameter expression is:

[0035]

[0036] Among them, C1, C2, C3, C4 are penalty coefficients, T = [s2K(X2, A t )s2e2] is the intermediate parameter matrix, T t Represents the transposed matrix of T, R = [K(X1,A t )e1] is the intermediate parameter matrix, R t represents the transposed matrix of R, s2 represents the specific fuzzy membership value of the first sample, I is the unit matrix, s1 represents the specific fuzzy membership value of the second sample, e1 and e2 are unit vectors, and K is the RBF kernel function.

[0037] Preferably, the expression of the decision function is:

[0038]

[0039] in, represents weight, K represents RBF kernel function, x represents the leakage sample data to be tested, A t =[X1;X2] t , b i Indicates deviation.

[0040] Beneficial effects of the present invention:

[0041] The present invention improves the fuzzy twin support vector machine in two aspects. On the one hand, the sample space is divided into regions by using the spatial relative distance ratio of different samples, and the fuzzy data set theory is introduced in the noise and support vector aliasing regions for distinction. The membership degree of the normal and noise data regions is also reassigned. This reduces the influence of noise data on the identification model and solves the shortcoming of FTSVM that it cannot distinguish between support vectors and noise data, resulting in poor model robustness. On the other hand, by expanding FTSVM to the least squares form, the two quadratic programming problems that the original model needs to solve are transformed into two linear equations problems, thereby improving the computational efficiency of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Attached Figure 1 It is an overall flow chart of the present invention in a specific implementation;

[0043] Attached Figure 2 This is the flow chart of the ImF-LSTSVM algorithm in the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] A specific implementation method of a fluid pipeline leakage detection method based on an improved fuzzy twin support vector machine, such as Figure 1 As shown, the method includes: obtaining vibration and sound signals of fluid pipelines at different leakage degrees; extracting features of vibration and sound signals at different leakage degrees, and normalizing the extracted feature data; collecting the normalized feature data, and dividing the collection into a training set and a test set; constructing an improved fuzzy twin support vector machine model, inputting the data in the training set into the improved fuzzy twin support vector machine model for training, and using the data in the test set to test the trained improved fuzzy twin support vector machine model to obtain the optimal improved fuzzy twin support vector machine model; obtaining the vibration and sound signal to be detected, and inputting the vibration and sound signal to be detected into the trained improved fuzzy twin support vector machine model to obtain the pipeline leakage detection result.

[0046] like Figure 2 As shown in FIG. 1 , the training process of the improved fuzzy twin support vector machine model includes:

[0047] Step 1: Obtain sample features of vibration-generated acoustic signals of different leakage levels; calculate the relative distance ratio of two samples; that is, input the training sample data set X+ ,X - , if it is a multi-classification task, two sample data sets of different classes are used as X + ,X - , to construct k(k-1) / 2 sub-classifiers.

[0048] The formula for calculating the sample center of a two-sample data set is:

[0049]

[0050] Among them, x c+ represents the center of the positive sample class, l + represents the number of positive samples, x i represents the sample data set, x c- represents the center of the negative sample class, l - represents the number of negative samples, α + Represents the relative distance ratio of positive samples, δ represents a constant greater than 0 to prevent the denominator from being 0, α - Represents the relative distance ratio of negative samples.

[0051] Step 2: Divide the samples into regions according to the relative distance ratio; the results of the region division include: normal data, noise data, and noise and support vector aliasing data division, the expression is:

[0052]

[0053] Step 3: According to the divided areas, the improved fuzzy membership function is used to weight the sample features to obtain a weight matrix with different leakage states;

[0054] According to the division of the sample space, the fuzzy membership of the sample is reconstructed, and the improved fuzzy membership function is used to perform weighted processing on the sample features under different leakage states. The specific formula is expressed as follows:

[0055] When the sample is in the normal data region, s i =(1-q)(1-||x i -x c+ || / r + +δ)

[0056] When the sample is in the noisy data region,

[0057] When the sample is located in the aliasing region of noise and support vector, the fuzzy data set theory is introduced to distinguish noise and support vector, and the s i The value is reassigned and the expression is:

[0058]

[0059] where s i Represents the specific fuzzy membership value of a single sample; x i is the i-th sample data; q is a constant value, the specific value is 0.1; x c+ is the sample center of the positive sample; r + =max{||x i -x c+ |||y i = +1}, is the maximum radius of the positive sample; α i is the relative distance value between different classes constructed based on the centers of positive and negative samples; μ i is the membership calculation value; v i The new membership value makes the normal data, noise data and support vector data samples have different weight values, thereby reducing the impact of noise data on model performance.

[0060] Step 4: Use the least squares method and weight matrix to construct the objective function of the least squares FTSVM identification model; its expression is:

[0061]

[0062] st-(K(X2,A t )w1+e2b1)+ξ=e2

[0063]

[0064] st-(K(X1,A t )w2+e1b2)+η=e1

[0065] Where X1 and X2 correspond to positive and negative sample data respectively; w i and b i are weights and biases respectively; A t =[X1;X2]; ξ and η represent slack variables; C1, C2, C3, C4 are penalty coefficients; K is the RBF kernel function, K(X,A t )=exp(-||XC t || 2 / 2γ 2 ), where γ is the kernel parameter.

[0066] Step 5: Solve the objective function to obtain the optimal hyperplane parameters corresponding to each type of sample; specifically, solve the optimal hyperplane of ImF-LSTSVM by bringing the equality constraint into the objective function:

[0067]

[0068] Solving the above equations, we can get the parameter expressions of the two hyperplanes:

[0069]

[0070] Where R = [K(X1,A t )e1],T=[s2K(X2,A t )s2e2]

[0071]

[0072] Among them, C1, C2, C3, C4 are penalty coefficients, T = [s2K(X2, A t )s2e2],T t Represents the transposed matrix of T, R = [K(X1,A t )e1],R t represents the transposed matrix of R, s2 represents the specific fuzzy membership value of the first sample, I is the unit matrix, s1 represents the specific fuzzy membership value of the second sample, e1 and e2 are unit vectors, and K is the RBF kernel function.

[0073] Step 6: Construct a decision function based on the optimal hyperplane parameters; the expression of the decision function is:

[0074]

[0075] in, represents weight, K represents RBF kernel function, x represents test sample, A t =[X1;X2] t , b i Indicates deviation.

[0076] Step 7: Input the test sample data into the decision function to obtain the pipeline leakage identification result and complete the model training.

[0077] In practical applications, it is necessary to collect various leakage conditions of fluid pipelines (such as no leakage, small leakage, medium leakage and severe leakage), and extract corresponding features from the leakage sound signal to form a feature data set. After the feature extraction is completed, the improved membership assignment method can be used to calculate the fuzzy membership of each data sample point and form a weight matrix, and then the feature data is input into the ImF-LSTSVM model, and the training optimization is performed through the "one-to-one" multi-classification strategy. After the training and optimization, the ImF-LSTSVM model has the ability to identify the leakage of fluid pipelines. When it is necessary to detect and analyze the leakage of the fluid pipeline, the vibration sound signal of the current fluid pipeline is collected, and then the features of the signal are extracted and the data is sent to the ImF-LSTSVM to obtain the leakage of the current fluid pipeline. Since the model uses the least squares method and the improved fuzzy membership function to improve FTSVM, the ImF-LSTSVM identification model has higher computational efficiency and better robust performance, and can still achieve high efficiency and accuracy in leakage identification when the fluid pipeline leakage is in a situation of severe noise.

[0078] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A fluid pipeline leakage detection method based on an improved fuzzy twin support vector machine, characterized in that: include: Obtain vibration and acoustic signals of fluid pipelines at different leakage levels; Extract features from vibration and sound signals with different leakage degrees, and normalize the extracted feature data; aggregate the normalized feature data and divide the aggregate into a training set and a test set; construct an improved fuzzy twin support vector machine model, input the data in the training set into the improved fuzzy twin support vector machine model for training, and use the data in the test set to test the trained improved fuzzy twin support vector machine model to obtain the optimal improved fuzzy twin support vector machine model; obtain the vibration and sound signal to be detected, input the vibration and sound signal to be detected into the trained improved fuzzy twin support vector machine model to obtain the pipeline leakage detection result; The training process of the improved fuzzy twin support vector machine model includes: Step 1: Obtain sample characteristics of vibration and acoustic signals of different leakage degrees; calculate the relative distance ratio of two samples; Step 2: Divide the samples into regions according to the relative distance ratio; Step 3: According to the divided areas, the improved fuzzy membership function is used to weight the sample features to obtain a weight matrix with different leakage states; Step 4: Use the least squares method and weight matrix to construct the objective function of the least squares FTSVM identification model; Step 5: Solve the objective function to obtain the optimal hyperplane parameters corresponding to each type of sample; Step 6: Construct a decision function based on the optimal hyperplane parameters; Step 7: Input the test sample data into the decision function to obtain the pipeline leakage identification result and complete the model training.

2. The fluid pipeline leakage detection method based on improved fuzzy twin support vector machine according to claim 1 is characterized in that: The formula for calculating the relative distance ratio of two samples is: Among them, x c+ represents the center of the positive sample class, l + represents the number of positive samples, x i represents the sample data set, x c- represents the center of the negative sample class, l - represents the number of negative samples, α + Represents the relative distance ratio of positive samples, δ represents a constant greater than 0 to prevent the denominator from being 0, α - Represents the relative distance ratio of negative samples.

3. The fluid pipeline leakage detection method based on improved fuzzy twin support vector machine according to claim 1 is characterized in that: The improved fuzzy membership function expression is: When the sample is in the normal data area, the fuzzy membership function is: s i =(1-q)(1-||x i -x c+ || / r + +δ) When the sample is a noise data area, the fuzzy membership function is: When the sample is a mixed area of ​​noise and support vector, the fuzzy membership function is: Among them, s i Represents the specific fuzzy membership value of a single sample; x i is the i-th sample data; q is a constant value, x c+ is the sample center of the positive sample; r + =max{||x i -x c+ |||y i = +1} is the maximum radius of the positive sample; α i is the relative distance value between different classes constructed based on the centers of positive and negative samples; μ i is the membership calculation value; v i Calculates the non-membership value.

4. The fluid pipeline leakage detection method based on improved fuzzy twin support vector machine according to claim 1 is characterized in that: The expression of the objective function of the least squares FTSVM identification model is: st-(K(X2,A t )w1+e2b1)+ξ=e2 st-(K(X1,A t )w2+e1b2)+η=e1 Among them, X1 and X2 correspond to positive and negative sample data respectively; w i and b i are weights and biases respectively; A t =[X1;X2]; ξ and η represent slack variables; C1, C2, C3, C4 are penalty coefficients; K is the RBF kernel function; e1 and e2 are unit vectors.

5. The fluid pipeline leakage detection method based on improved fuzzy twin support vector machine according to claim 1, characterized in that: Solving the objective function includes: substituting the constraint conditions into the objective function to obtain the objective function to be solved; solving w i and b i The parameters are differentiated to obtain the parameters of the hyperplane.

6. The fluid pipeline leakage detection method based on improved fuzzy twin support vector machine according to claim 5 is characterized in that: The hyperplane parameter expression is: Among them, C1, C2, C3, C4 are penalty coefficients, T = [s2K(X2, A t )s2e2],T t Represents the transposed matrix of T, R = [K(X1,A t )e1],R t represents the transposed matrix of R, s2 represents the specific fuzzy membership value of the first sample, I is the unit matrix, s1 represents the specific fuzzy membership value of the second sample, e1 and e2 are unit vectors, and K is the RBF kernel function.

7. The fluid pipeline leakage detection method based on improved fuzzy twin support vector machine according to claim 1, characterized in that: The expression of the decision function is: in, represents weight, K represents RBF kernel function, x represents the leakage sample data to be tested, A t =[X1;X2] t , b i Indicates deviation.

Images