Pipeline leakage denoising method based on improved four-vector optimization algorithm parameter optimization

Abstract

The invention relates to the field of pipeline leakage detection, in particular to a pipeline leakage denoising method based on parameter optimization of an improved four-vector optimization algorithm. The method comprises the following steps: collecting a leakage infrasonic wave original signal X (t), performing low-pass filtering to obtain a signal use good point set and a random difference variation strategy optimization FVIM algorithm, obtaining a GDFVIM algorithm, and adaptively selecting an optimal decomposition layer number K and a penalty factor alpha of VMD decomposition by using the GDFVIM algorithm; based on the selected optimal decomposition layer number K and the penalty factor alpha, performing VMD decomposition on the signal to obtain K IMF components; and calculating the mutual information amount between each IMF component and the leakage infrasonic wave original signal X (t), selecting an effective IMF component by using the mutual information amount, and reconstructing to obtain a denoised leakage infrasonic wave signal. The accuracy and efficiency of parameter selection can be improved, and the performance of pipeline leakage detection is enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of pipeline leakage detection, in particular to a pipeline leakage denoising method based on improved four-vector optimization algorithm parameter optimization. BACKGROUND

[0002] Pipeline systems are inevitably threatened by various factors due to long-term complex natural environment and working conditions, resulting in leakage accidents. However, due to the interference of on-site operating equipment and environment, noise is mixed in the pipeline leakage signal, resulting in errors in the detection results. The VMD method has been widely used in pipeline leakage detection, and the method has good noise robustness.

[0003] However, when using VMD for signal processing, the values of the decomposition layer (K) and the penalty factor (a) need to be set in advance, and the values of K and a have a great influence on the decomposition effect. There are currently three methods to determine K and a: one is based on empirical knowledge or center frequency observation method, which requires multiple attempts, low efficiency and poor adaptability; the second is the evaluation index selection method, which establishes a suitable evaluation index according to the signal characteristics to select parameters (such as the patent with the application number CN202411300098.2 and the patent name of Acquisition method of pipeline leakage judgment standard based on multivariate variational mode decomposition), but the K value selected by this method is not universal.

[0004] Therefore, it is urgent to design a parameter optimization method that can improve the accuracy and efficiency of parameters, so as to better denoise the pipeline leakage signal. SUMMARY

[0005] The technical problem to be solved by the present application is to overcome the defects of the prior art and provide a pipeline leakage denoising method based on improved four-vector optimization algorithm parameter optimization, which can improve the accuracy and efficiency of parameter selection and enhance the performance of pipeline leakage detection.

[0006] In order to solve the above technical problems, the technical scheme of the present application is as follows: a pipeline leakage denoising method based on improved four-vector optimization algorithm parameter optimization, comprising:

[0007] Collecting the leakage infrasound wave original signal X(t) and performing low-pass filtering to obtain the signal

[0008] Using the GDFVIM algorithm to adaptively select the optimal decomposition layer number K and the penalty factor a of VMD decomposition, and using the GDFVIM algorithm to adaptively select the optimal decomposition layer number K and the penalty factor a of VMD decomposition;

[0009] Based on the selected optimal decomposition layer number K and the penalty factor a, the signal VMD decomposition is performed to obtain K IMF components;

[0010] The mutual information between each IMF component and the original leak infrasound signal X(t) is calculated, and the effective IMF component is selected using the mutual information, and the denoised leak infrasound signal is obtained after reconstruction.

[0011] Further, the position equation of the GDFVIM algorithm is:

[0012]

[0013] In the formula, X n,i new represents the updated position of the nth best agent in the ith dimension; P n,i represents the current position of the nth best agent in the ith dimension; represents the current average position of all agents in the ith dimension; θ represents an adaptive coefficient; u1, u2, u3 represent random numbers uniformly distributed in [0, 1] generated by the optimal point set; F represents a difference scaling factor, X r1,i , X r2,i represents different individuals randomly selected in the population.

[0014] Further, in the process of adaptively selecting the optimal decomposition level K and the penalty factor α of VMD decomposition using the GDFVIM algorithm, the envelope entropy is used as the fitness function.

[0015] Further, the calculation formula of the mutual information is:

[0016]

[0017] In the formula, MI(X, Y) represents the mutual information of X and Y, H is the number of samples, and f(x) is the overall multivariate kernel density function.

[0018] Further, the effective IMF component is selected using the mutual information, specifically:

[0019] The mean of the mutual information between each IMF component and the original leak infrasound signal X(t) is taken as the threshold;

[0020] The IMF component with mutual information greater than the threshold with the original leak infrasound signal X(t) is taken as the effective IMF component.

[0021] By adopting the technical scheme, the application can increase the search diversity of the algorithm, help the algorithm to explore the solution space more effectively, help to avoid the algorithm from falling into local optimum, and improve the probability of finding the global optimal solution by using the point set and the random difference mutation strategy to optimize the four-vector optimization (FVIM) algorithm; the accuracy and efficiency of parameter selection can be improved by using the optimal parameters K and a of the VMD (variational mode decomposition) adaptively selected by the GDFVIM; the IMFs (intrinsic mode functions) containing the leakage signal are screened out by using the MI (mutual information) method based on the optimal parameters K and a, and the denoised signal is reconstructed, so that the denoising effect is good and the performance of the pipeline leakage detection is enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is a flowchart of a pipeline leakage denoising method based on improved four-vector optimization algorithm parameter optimization of the application;

[0023] Figure 2 is a pipeline leakage infrasound wave original signal X(t) acquisition experimental device diagram;

[0024] Figure 3 is a GDFVIM-VMD decomposition result diagram;

[0025] Figure 4 is a GDFVIM-VMD decomposition component spectrum diagram;

[0026] Figure 5 is an IMF component and original signal mutual information diagram;

[0027] Figure 6 is an original signal time domain diagram;

[0028] Figure 7 is a FVIM-VMD decomposed reconstructed signal time domain diagram;

[0029] Figure 8 is a GDFVIM-VMD decomposed reconstructed signal time domain diagram. DETAILED DESCRIPTION

[0030] In order to make the content of the application more easily understood, the application will be further described in detail below according to specific embodiments and in combination with the drawings.

[0031] As shown in Figure 1 , a pipeline leakage denoising method based on improved four-vector optimization algorithm parameter optimization, comprising:

[0032] Step S1, collect the leakage infrasound wave original signal X(t), use the FIR low-pass filter to perform low-pass filtering on the leakage infrasound wave original signal X(t), and obtain the signal

[0033] Specifically, the acquisition process of the signal is often accompanied by various interferences, which can cause the signal to leak, greatly affecting the accuracy and reliability of signal analysis. Therefore, a finite impulse response (FIR) low-pass filter is used to process the original leakage signal. The FIR filter is designed based on the window function method to filter the original leakage signal, and the order of the FIR low-pass filter can be N = 50, and the cutoff frequency can be f c = 80.

[0034] Step S2, using the set of good points and the random differential mutation strategy to optimize the four-vector optimization (FVIM) algorithm, and obtaining an improved four-vector optimization (GDFVIM) algorithm;

[0035] Specifically, the introduction of the set of good points and the random differential mutation strategy in the four-vector optimization algorithm can increase the search diversity of the algorithm, help the algorithm explore the solution space more effectively, and help avoid the algorithm from falling into local optimum, thereby improving the probability of finding the global optimal solution.

[0036] Step S3, using the GDFVIM algorithm to adaptively select the optimal decomposition layer number K and the penalty factor a of the variational mode decomposition (VMD) decomposition;

[0037] Specifically, the FVIM optimization process includes three main stages, which dominate the operation of FVIM, namely, the initialization stage, the iteration stage and the optimal solution stage. FVIM determines the optimal solution of the problem by systematically comparing the obtained solution with the value specified in the objective function.

[0038] FVIM proposes and utilizes different mathematical models to control the behavior of the search agent. Four different agent movement and positioning models are shown in equations (1)-(4). The center position is found by averaging the results obtained from the previous models through equation (5).

[0039]

[0040] where X n,i represents the updated position of the nth best agent in the ith dimension, n = 1, 2, 3, 4. P n,i is the current position of the nth best agent in the ith dimension. represents the current average position of all agents in the ith dimension. θ is an adaptive coefficient. ξ1, ξ2, ξ3 represent random numbers uniformly distributed in the range [0, 1], which are used to introduce randomness into the algorithm to ensure that individuals explore different regions of the search space.

[0041] In the FVIM algorithm, θ is used to bridge the gap between exploration and exploitation for the same purpose as the inertia weight (W) of PSO, as shown in equation (6).

[0042]

[0043] where Max_iter is the maximum number of iterations, and θ plays a balancing role between exploration and exploitation. Too large θ will increase the step size in the iteration process, causing the swarm to move away from the optimal region. Conversely, too small θ prioritizes local fine-tuning at the expense of exploration. However, determining the ideal step size depends on the nature of the problem.

[0044] In the original FVIM iteration process, the optimal individual falls into a local optimum, while the GDFVIM combines the good point set and the random differential mutation strategy in the differential evolution algorithm to ensure the diversity of the initial population and enable the algorithm to jump out of the local optimum, balancing global and local search capabilities. The improved position equation is:

[0045]

[0046] where X n,i new represents the updated position of the nth best individual in the ith dimension; u1, u2, and u3 are generated from the good point set, ensuring a more uniform distribution in the [0, 1] interval, making the individual position update more reasonable and balancing exploration and exploitation. represents the current average position of all individuals in the ith dimension; F represents the differential scaling factor, X r1,i , X r2,i represents randomly selected different individuals in the population. Therefore, GDFVIM exhibits superior exploration of the solution space, helping to avoid the algorithm falling into a local optimum and improving the probability of finding a global optimal solution.

[0047] When using GDFVIM to optimize VMD parameters, the envelope entropy is selected as the fitness function, and the envelope entropy calculation formula is as follows:

[0048]

[0049] where: E z is the envelope entropy of the zth IMF component; j is the discrete sequence index, i.e., the jth sampling point of the envelope signal; J is the total number of envelope signal sampling points; a z (j) is the envelope signal obtained by Hilbert transform of the zth IMF component;

[0050] Step S4, based on the selected optimal decomposition layer number K and the penalty factor α, VMD is performed on the signal to obtain K IMF components;

[0051] Step S5, calculate the mutual information between each IMF component and the original signal of the leaked infrasound wave X(t);

[0052] Wherein, the mutual information (MI) is an index to measure the degree of interdependence between two random variables, and can measure the information shared by two variables. Therefore, MI is used to select the IMF component related to the leakage signal for reconstruction. The smaller the mutual information between each IMF component and the original signal, the less effective signal it contains, i.e. the higher the noise content. The mutual information MI calculation expression is as follows:

[0053] Since the joint distribution P(x, y) is difficult to obtain and is not universally applicable, kernel density estimation is used instead. In this embodiment, normal multivariate kernel density estimation is used instead, which is defined as follows:

[0054]

[0055] In the formula, MI(X, Y) represents the mutual information of X and Y, H is the number of samples, and f(x) is the overall multivariate kernel density function.

[0056] Step S6, use the mutual information to select the effective IMF component; specifically:

[0057] Take the average of the mutual information between each IMF component and the original signal of the leaked infrasound wave X(t) as the threshold;

[0058] Take the IMF component with mutual information greater than the threshold with the original signal of the leaked infrasound wave X(t) as the effective IMF component.

[0059] Step S7, reconstruct to obtain the denoised leaked infrasound wave signal.

[0060] The above-mentioned scheme related to the embodiments will be described in detail below in conjunction with specific embodiments.

[0061] Example 1

[0062] 1. The experimental parameters are as follows: the pipeline system in this test is a U-shaped PE pipe, the pipeline Φ110x10m, the total length is 3599cm, the initial pressure of the pipeline is 0.3MPa, and the flow rate is 16m / s. Two CASI-NGP-A infrasound sensors are placed upstream and downstream of the pipeline, and the leakage point is located between the two sensors, with a leakage aperture of 3mm. The infrasound sensors are connected to a digital network transmission instrument through a cable, and then transmitted to the analysis software on the PC end through a network cable. The experimental pipeline layout is shown in Figure 2 The infrasound sensor is used to collect the original signal of the pipeline leakage, and the original signal of the pipeline leakage X(t) is obtained.

[0063] 2、Firstly, the original signal collected is preprocessed. The signal collection process is often accompanied by various interferences, which may cause signal leakage phenomenon, greatly affecting the accuracy and reliability of signal analysis. Therefore, the original leakage signal is processed using a finite impulse response (FIR) low-pass filter. The FIR filter based on window function method is used to filter the original leakage signal. The order N of the FIR low-pass filter is 50, the cutoff frequency f c = 80, and the filtered original signal is obtained

[0064] 3、Based on GDFVIM, the optimal parameters of VMD are adaptively selected, and the iteration number is 50 times, and the envelope entropy is used as the fitness function. The optimization result shows that when K = 7 and a = 861, the system obtains the optimal fitness value 7.4012. The VMD decomposition result is shown in The decomposition result is shown in Figure 3 The IMF component spectrum is shown in Figure 4

[0065] 4、The mutual information (MI) of the 7 IMF components generated by VMD decomposition and the original signal is quantitatively analyzed, and a component screening mechanism is established. As shown in Table 1, the mutual information MI i and the average mutual information MI m of each IMF and the original signal are calculated, and the threshold of MI i > MI m is set as the effective component determination threshold. Through the calculation of MI m = 1.1828, IMF4-IMF6 are selected for signal reconstruction to obtain According to Figure 5 , it can be directly observed that IMF4-IMF6 exhibit significant correlation characteristics. Figure 6 is the time domain graph of the original signal, Figure 8 is the time domain graph of the reconstructed signal, and the time domain comparison curve of Figure 6 and Figure 8 clearly shows that the reconstructed signal retains the leakage characteristics while effectively suppressing high-frequency noise.

[0066] Table 1 MI of IMF component and original signal

[0067]

[0068] Comparative Example 1

[0069] ​The difference between Comparative Example 1 and Example 1 is that the VMD optimal parameters are adaptively selected based on FVIM. The modal number K is set to be in the range of [1, 10], the penalty factor a is set to be in the range of [10, 2500], the iteration number is 50 times, and the envelope entropy is used as the fitness function. The optimization results show that when K = 8 and a = 643, the system obtains the optimal fitness value of 7.4129.

[0070] The difference between Comparative Example 1 and Example 1 is that the VMD optimal parameters are adaptively selected based on FVIM. The modal number K is set to be in the range of [1, 10], the penalty factor a is set to be in the range of [10, 2500], the iteration number is 50 times, and the envelope entropy is used as the fitness function. The optimization results show that when K = 8 and a = 643, the system obtains the optimal fitness value of 7.4129. After implementing FVIM-VMD decomposition, mutual information (MI) quantitative analysis is performed on the 8 IMF components generated after decomposition and the original signal, and the mutual information MI i and the average mutual information MI m of each IMF and the original leakage signal are calculated. MI i >MI m is set as the effective component judgment threshold. By calculating MI m = 1.05216, IMF3 (MI3 = 1.326) and IMF6 (MI6 = 1.112) are selected for signal reconstruction to obtain Figure 7 is the time threshold graph of the reconstructed signal.

[0071] A multi-dimensional evaluation system is used to verify the effectiveness of the method, and the root mean square error (RMSE), signal-to-noise ratio (SNR), and mean absolute error (MAE) of the reconstructed signals in Example 1 and Comparative Example 1 are calculated as standard evaluation of denoising effect, and the results are shown in Table 2.

[0072] Table 2 Denoising Evaluation Table

[0073]

[0074] As shown in the experimental data in Table 2, after processing by the method, the RMSE is reduced from 0.00404 to 0.00390, with a decrease of 3.5%; the SNR is improved from 8.43 dB to 9.51 dB, with a relative improvement of 12.8%; the MAE is optimized from 0.00320 to 0.00306, with a decrease of 4.4%; the optimization effect is relatively significant.

[0075] Among them, the denoising effect of Comparative Example 1 is not ideal, the main reason is that the four-vector optimization algorithm (FVIM) has the defects of slow convergence speed and low convergence accuracy, which leads to inaccurate parameter selection, affects the accuracy and efficiency of leakage detection, and reduces the feature extraction ability and noise separation effect of the leakage signal. In Comparative Example 1, the FVIM is improved by using the good point set (GPS) and the random differential mutation strategy (DE), which solves this problem.

[0076] ​With the above ideal embodiments according to the present application as the inspiration, through the above description, relevant staff can make various changes and modifications without deviating from the technical idea of the present application. The technical scope of the present application is not limited to the content of the specification, and must be determined according to the scope of the claims.

Claims

1. A pipeline leakage denoising method based on parameter optimization of an improved four-vector optimization algorithm, characterized in that: include: Collect the original signal X(t) of the leakage infrasound wave and perform low-pass filtering to obtain the signal The FVIM algorithm is optimized using the good point set and random difference mutation strategy to obtain the GDFVIM algorithm, which is then used to adaptively select the optimal decomposition level K and penalty factor α of VMD decomposition. Based on the selected optimal decomposition layer K and penalty factor α, the signal Perform VMD decomposition to obtain K IMF components; The mutual information between each IMF component and the original leakage infrasound signal X(t) is calculated, and the effective IMF components are selected using the mutual information. After reconstruction, the denoised leakage infrasound signal is obtained.

2. The pipeline leakage denoising method based on parameter optimization of the improved four-vector optimization algorithm according to claim 1 is characterized in that: The position equation of the GDFVIM algorithm is: Where, X n,i new represents the updated position of the nth best agent in the i-th dimension; P n,i represents the current position of the nth best agent in the i-th dimension; represents the current average position of all agents in the i-th dimension; θ represents an adaptive coefficient; u1, u2, u3 represent random numbers uniformly distributed in the range [0, 1] generated by the good point set; F represents the differential scaling factor, X r1,i 、X r2,i represents different individuals randomly selected from the population.

3. The pipeline leakage denoising method based on parameter optimization of the improved four-vector optimization algorithm according to claim 1 is characterized in that: In the process of adaptively selecting the optimal decomposition level K and penalty factor α of VMD decomposition using the GDFVIM algorithm, the envelope entropy is used as the fitness function.

4. The pipeline leakage denoising method based on parameter optimization of the improved four-vector optimization algorithm according to claim 1 is characterized in that: The calculation formula of mutual information is: Where MI(X,Y) represents the mutual information between X and Y, H is the number of samples, and f(x) is the overall multivariate kernel density function.

5. The pipeline leakage denoising method based on parameter optimization of the improved four-vector optimization algorithm according to claim 1 is characterized in that: Use mutual information to select effective IMF components, specifically: The mean of the mutual information between each IMF component and the original signal X(t) of the leakage infrasound wave is used as the threshold; The IMF component whose mutual information with the original signal X(t) of the leakage infrasound wave is greater than the threshold is taken as the effective IMF component.

Images