A pipeline leakage identification method combining OVMD and GBDT algorithm

Abstract

The present application relates to the field of offshore platform pipeline leakage detection, and proposes a pipeline leakage identification method combining OVMD and GBDT algorithm. First, a pipeline leakage model is established, leakage signals under different working conditions are collected, and real background noise is collected on the offshore platform, which is injected into the laboratory leakage signal to simulate the actual working environment. The original sample data set constructed is decomposed using the OVMD method to obtain multiple signal component functions, feature indicators are extracted, the GBDT algorithm is used to classify and identify the extracted feature indicators, and a confusion matrix and precision-recall curve are formed. The present application effectively improves the identification accuracy of the leakage signal, solves the problems of difficulty in acoustic emission signal collection, low identification accuracy of existing methods, poor robustness and complex implementation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of offshore platform pipeline leakage detection, in particular to a pipeline leakage identification method combining OVMD and GBDT algorithm. BACKGROUND

[0002] With the increase of the service life of the marine oil and gas platform pipeline system, its safety problem is increasingly prominent. In the complex marine environment, both natural gas pipelines and crude oil processing facilities will vibrate due to various vibration sources. This long-term accumulated vibration effect may cause pipeline structure fatigue and even damage. If it is not found and prevented in time, once a leakage accident occurs, it will pose a major safety threat to the marine oil and gas platform, and even may cause disastrous consequences such as explosion, causing incalculable economic losses and social impact. Therefore, the leakage protection of offshore oil and gas pipelines has become one of the focus problems in the industry. However, the traditional theoretical analysis method is difficult to implement in many actual situations, and it is difficult to fully ensure the safety and reliability of the equipment. Therefore, the industry gradually introduces and widely applies acoustic emission technology for real-time monitoring and diagnosis. Acoustic emission detection technology is a dynamic, real-time and non-destructive evaluation method, which captures the acoustic signals spontaneously generated by the internal structure of the material under stress or environmental changes, accurately detects the health status of the overall structure, and can realize early warning of potential damage. This technology does not need to rely on external excitation source, has high defect recognition ability, comprehensive evaluation performance and wide application adaptability, etc.

[0003] In order to use acoustic emission technology to identify pipeline leakage, the first thing is to collect the pipeline leakage signal. However, due to actual reasons, it is difficult to collect high-quality pipeline leakage signals. In addition, due to the extremely complex working environment of the offshore platform pipeline, and the strong noise interference, such as mechanical vibration noise generated by strong pipeline vibration, sea wave noise and wind noise existing in the marine environment, etc. These interferences further increase the difficulty of collecting high-quality acoustic signals.

[0004] For the collected acoustic emission signals, machine learning and deep learning are one of the methods for signal recognition. However, the traditional method has the problem of low recognition accuracy and low robustness under strong noise interference. In addition, the traditional machine learning method needs to expand the information dimension of the signal to get a relatively high accuracy, which increases the complexity of the processing process.

[0005] Therefore, it is necessary to provide a new technical solution to solve the above problems, and the present application improves the traditional variational mode decomposition (VMD) and proposes an optimal variational mode decomposition (OVMD) combined with a pipeline leakage identification method of gradient boosting decision tree (GBDT) algorithm. SUMMARY

[0006] The technical problem to be solved by the present application is to provide a way to inject actual background noise into laboratory leakage signals to solve the problem of difficult acquisition of high-quality acoustic emission signals, and to provide a new method to solve the problems of insufficient recognition accuracy, low robustness and complex implementation process of existing methods.

[0007] To achieve the above purpose, the present application is realized by the following technical scheme: a pipeline leakage identification method (OVMD-GBDT) of optimal variational mode decomposition (OVMD) combined with gradient boosting decision tree (GBDT) algorithm, comprising the following steps:

[0008] S1, building a pipeline leakage model, setting acoustic emission sensors on the measured pipeline, and collecting multiple groups of pipeline leakage signals;

[0009] S2, selecting at least two places on the offshore platform to set the acoustic emission sensors, and collecting actual background noise signals;

[0010] S3, injecting the actual background noise signal into the pipeline leakage signal to construct an original sample data set;

[0011] S4, decomposing the original sample data set constructed in step S3 using the OVMD method to obtain multiple signal component functions, the OVMD method comprising: introducing the Pearson correlation coefficient to select the optimal mode number , using peacock optimization algorithm to optimize the penalty factor and the fidelity coefficient ;

[0012] S5, extracting feature indicators from the signal component functions obtained in step S4 to form a feature vector set;

[0013] S6, using the GBDT algorithm to classify and identify the extracted feature vector set to form a confusion matrix and a precision-recall curve.

[0014] Preferably, the pipeline leakage model in step S1 includes: a pipeline, a pressure device and the acoustic emission sensor, the acoustic emission sensor is connected to a preamplifier, the preamplifier is connected to an acoustic emission detection device, and the acoustic emission detection device is connected to an industrial computer.

[0015] Preferably, the pipeline leakage model in step S1 includes two areas: one is an experimental section, in which multiple control valves are evenly distributed for simulating and controlling leakage conditions; the other is a buffer section, which is connected to a pressure device through a flexible rubber pipe, and the pressure device adopts an electric pressure self-controlled pressure test pump.

[0016] Preferably, the acoustic emission sensor installed for collecting the actual background noise signal in step S2 is the same as that in step S1, and the sampling mode is a long-term timing sampling mode.

[0017] Preferably, in step S3, the pipeline leakage signal in step S1 is first evenly divided according to the length of the sampling point to obtain multiple leakage signal samples, and then the actual background noise signal collected in S2 is also divided into sampling point segments with the same number as the pipeline leakage signal, and noise signal samples with the same number as the leakage signal samples are constructed. Finally, by injecting the noise signal samples into the leakage signal samples, the original sample data set containing the leakage signal samples and the noise signal samples is formed.

[0018] Preferably, the Pearson correlation coefficient is introduced in step S4 to select the optimal modal number. , mode selection is performed according to the following model:

[0019] ;

[0020] in, is the Pearson correlation coefficient function, which is calculated as follows:

[0021] ; is the original signal; is the first one obtained by VMD decomposition modal components; is the residual signal after decomposition; It is The correlation coefficient between the modal components and the original signal; is the correlation coefficient between the residual signal and the original signal; is the residual correlation coefficient attenuation factor. According to the model setting, the optimal modal number can be automatically selected. , the specific process is as follows:

[0022] (1) Set the initial mode number , a preliminary VMD decomposition is performed on the signal to be analyzed;

[0023] (2) For the current modal number, the Pearson correlation coefficient between each modal component and the original signal is calculated , and the correlation coefficient between the residual signal and the original signal is further calculated .

[0024] (3) If the correlation of the residual signal is greater than the smallest correlation coefficient among all modal components, i.e. , it indicates that there is still under-decomposition, so K is increased by 1 and the second step is returned to continue iteration.

[0025] (4) If is satisfied, and , the current modal number is the reasonable upper limit and can be saved as the optimal value.

[0026] (5) After the stop condition is satisfied, the current is output as the final estimated optimal modal number, and the program is terminated.

[0027] Preferably, in the S4 step, the peacock optimization algorithm is introduced to optimize the penalty factor and the fidelity coefficient , including: generating multiple candidate parameter combinations according to the optimal modal number obtained in the S4 step , and performing VMD decomposition on them as individuals, calculating the corresponding residual signals and obtaining evaluation indicators, selecting the position with the optimal fitness, and then generating a new position based on the peacock optimization strategy and performing residual calculation and indicator evaluation again.

[0028] Preferably, in the S5 step, the feature indicators of the signal component function are extracted as the time domain features and frequency domain features of the signal component function obtained in the S4 step, and the feature vector set of the original sample data set is constructed; wherein the time domain features include: mean, standard deviation, minimum value, maximum value, root mean square value, skewness, kurtosis, peak value, margin, pulse index; the frequency domain features include spectral mean, spectral standard deviation, spectral energy, spectral center, spectral flatness, spectral roll-off point, spectral skewness, frequency kurtosis, spectral drop, spectral coefficient of variation.

[0029] Preferably, in the S6 step, the GBDT method is used for classification, 70% of the feature vector set is used as the training set and 30% is used as the test set, and the confusion matrix and precision-recall curve are formed as the identification result.

[0030] The application provides a pipeline leakage identification method (OVMD-GBDT) combining optimal variational modal decomposition (OVMD) and gradient boosting decision tree (GBDT) algorithm.

[0031] 1. The application solves the problem of difficult high-quality pipeline leakage signal acquisition to a certain extent.

[0032] 2. The application provides an improved VMD method, i.e., an OVMD method, which solves the problem that the traditional VMD method is very sensitive to modal number , penalty factor and fidelity coefficient . The optimal modal number can be conveniently selected as the final estimated modal number by introducing the Pearson correlation coefficient to estimate the modal number . The penalty factor and fidelity coefficient are optimized by introducing the peacock optimization algorithm to improve the accuracy of signal decomposition and the effect of feature extraction, and the residual signal is minimized.

[0033] 3. The application solves the problem that the traditional VMD method reduces the identification efficiency due to too much information sequence after obtaining multiple signal component functions.

[0034] 4. The application solves the problem that the traditional machine learning method has low identification accuracy in the case of low signal-to-noise ratio. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is a schematic diagram of the overall process of the present application;

[0036] Figure 2 is a schematic diagram of the pipeline leakage model of the present application;

[0037] Figure 3 is a frequency spectrum diagram after noise signal injection of the present application;

[0038] Figure 4 is a time-frequency diagram after noise signal injection of the present application;

[0039] Figure 5 is a flowchart for calibrating the number of modes K based on correlation coefficients of the present application;

[0040] Figure 6 is a flowchart for optimizing the penalty factor a and the fidelity coefficient t based on the peacock optimization algorithm of the present application;

[0041] Figure 7 is a confusion matrix diagram of the present application;

[0042] Figure 8 is a precision-recall curve diagram of the present application.

[0043] The reference signs are: 1, pipeline; 2, control valve; 3, acoustic emission sensor; 4, preamplifier; 5, acoustic emission detection device; 6, industrial computer. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the specification of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0045] Please refer to the attached Figure 1 , Figure 1 is a schematic diagram of the overall process of the pipeline leakage identification method of the present application according to the OVMD combined with the GBDT algorithm. The method provided by the present application can include the following steps:

[0046] S1, build a pipeline leakage model, set acoustic emission sensors on the measured pipeline, and collect multiple groups of pipeline leakage signals. The built pipeline leakage model mainly includes a pipeline 1, a pressure device, a pressure gauge and a series of control valves 2, wherein the pipeline 1 is used to simulate the working pipeline of the offshore platform, the pressure device is used to provide pressure to make the fluid flow in the pipeline 1, the pressure gauge is used to reflect the current pressure in the pipeline, and the control valve 2 is used to control the size of the leakage to realize the simulation of different leakage conditions.

[0047] S2, select at least two pipeline settings on the offshore platform to set the acoustic emission sensor, collect the actual background noise signal. Select the actual investigation of "Lianwan 3-1" offshore platform, collect the actual background noise signal at 19 meters platform and 29 meters platform.

[0048] S3, inject the actual background noise signal into the pipeline leakage signal to construct the original sample data set. Because the actual background noise signal and the pipeline leakage signal are both continuous signals, the signal power is used to determine the noise percentage, and then the appropriate sample number is selected for injection to obtain the original sample data set.

[0049] S4, decompose the original sample data set constructed in step S3 using the OVMD method to obtain a plurality of signal component functions. The OVMD method is improved based on the traditional variational mode decomposition (VMD) method, and the core is the modal number The rate is calibrated based on the Pearson correlation coefficient, and the peacock optimization algorithm (POA) is introduced to optimize the penalty factor And the fidelity coefficient A modal number estimation model based on the Pearson correlation coefficient is proposed, which aims to minimize the effective modal number To prevent the phenomenon of false modes caused by too large Value. The penalty factor And the fidelity coefficient Optimization to minimize the residual signal as much as possible, thereby improving the accuracy of subsequent analysis or prediction tasks.

[0050] S5, extract the feature index of the signal component function from the plurality of signal component functions obtained in step S4 to form a feature vector set to improve the recognition efficiency.

[0051] S6, classify and identify the feature vector set extracted above using the GBDT algorithm to form a confusion matrix and a precision-recall curve for reflecting the recognition accuracy and performance of the method. All data are input into the GBDT algorithm for classification and identification, and the confusion matrix is used to reflect the data recognition of the model, and the precision-recall curve is used to reflect the recognition accuracy of the model.

[0052] The overall idea of the method is to obtain a variety of leakage signals under various leakage conditions in a laboratory controlled environment first, then collect the actual background noise signal of the pipeline, inject the actual background noise signal into the leakage signal, and construct a dataset simulating the working conditions in a real environment. The traditional variational mode decomposition (VMD) method is improved to obtain the OVMD method, and then the OVMD method is used to decompose the constructed dataset to obtain the signal component function. Select the characteristic index to extract the characteristics in the signal component function, and then input it into the gradient boosting tree algorithm (GBDT) for classification and recognition to obtain the confusion matrix and precision-recall curve for reflecting the recognition accuracy and method performance.

[0053] Please refer to the accompanying drawings Figure 2 , Figure 2 is a schematic diagram of the pipeline leakage model in the present application, which shows the overall structure of the pipeline leakage model.

[0054] The pipeline leakage model mainly consists of a pipeline 1, a pressure device, a pressure gauge and a series of control valves 2. The selected pipeline model strictly follows the GB / T 8163-2018 standard, uses seamless steel pipe material, and is suitable for high-pressure fluid transportation. The pipeline 1 has a diameter of 105 mm, a wall thickness of 5 mm and a total length of 6.2 m. The system is divided into two key areas. First, the experimental section: 5 meters long, with multiple high-pressure welded needle-shaped control valves 2 evenly distributed inside, used to simulate and control the leakage condition; then the buffer section: 1.2 meters long, connected to the pressure device through a flexible rubber pipeline. The pressure device uses a Z4DSY type electric pressure self-control test pressure pump with excellent performance. The acoustic emission sensor 3 includes two R15 sensors, which are installed at both ends of the pipeline 1 10 cm away from the boundary of the experimental section, and industrial-grade white vaseline is used as the coupling medium between the pipeline surface and the sensor, and then fixed with adhesive tape.

[0055] The data acquisition device uses a type of acoustic emission detection device 5 produced by the United States Physical Acoustics Company (PAC) with the model number PCI-2. The device has a wide and precise sampling frequency range (1 kHz to 3 MHz) and is built-in with a high-efficiency data storage function, which can store acoustic emission waveform information on the hard disk at a speed of 10 million sampling points per second. The data acquisition stage uses the AEwin software that comes with the PCI-2 device for operation and control. The acoustic emission sensor 3 used in the experiment is of the R15 type, with a center frequency set to 150 kHz. In order to further optimize the signal quality and enhance the contrast between the signal and the noise, the experiment configures a 2 / 4 / 6 channel preamplifier 4, which can be adjusted to 20 / 40 / 60 dB in gain. The acquisition parameters are set and optimized in advance on the AEwin software platform of the industrial computer 6. Then, a lead fracture experiment is conducted to verify the coupling state of the sensor and the pipeline and the stability of signal transmission, and to confirm whether the data acquisition device is functioning properly. The final gain is set to 40db, and the sampling frequency is set to 1MHz. After debugging, the pipeline leakage acoustic emission experiment is carried out according to the method in Figure 1 , and multiple sets of pipeline leakage signals are collected.

[0056] In order to collect the actual background noise signals of the pipeline, a detailed on-site survey is conducted at the "Lianwan 3-1" offshore platform, and the pipelines of the 19-meter platform and the 29-meter platform are selected as the collection objects. The 19-meter platform has a typical mechanical vibration noise characteristic due to its high pipeline vibration intensity, and is prone to fatigue damage. The 29-meter platform is in the cooling process after the compressor is pressurized, and the pipeline at this location bears a large pressure, especially at the elbow part, which is prone to damage due to stress concentration. Therefore, these two locations with typical noise characteristics and potential risks are selected as the collection points of the on-site noise signals. The type and parameter settings of the installed acoustic emission sensors are consistent with those in the laboratory, and industrial white vaseline is used as the coupling agent, and the sensors are fixed with adhesive tape; the sampling mode is long-term timed sampling mode.

[0057] Please refer to Figure 3 and Figure 4 , which are the frequency spectrum and time-frequency diagram respectively after the actual background noise signal is injected into the pipeline leakage signal in the present invention.

[0058] Since the noise level in the laboratory is relatively low and cannot reflect the actual noise situation on the offshore platform, the measured on-site noise is injected into the leakage signal collected in the laboratory to verify the acoustic emission pipeline leakage detection method. Since both are continuous signals, the noise percentage is determined by the signal power, i.e.

[0059] ;

[0060] where is the noise power, is the signal power, , are the noise and leakage signal root mean square values respectively, is the sample number. Therefore, the leakage signal fused with noise can be obtained by the following formula:

[0061] ;

[0062] wherein is the collected leakage signal; is the measured noise signal.

[0063] In order to make the sample of the noise-containing leakage signal, the laboratory-collected pipeline leakage signal is first uniformly segmented according to the length of 1024 sampling points, so as to obtain 900 leakage signal samples. The actual noise signal collected on the offshore platform is also divided into paragraphs with a length of 1024 sampling points, and then 900 noise signal samples are constructed. Finally, by injecting the noise signal sample into the leakage signal sample, an original sample data set containing 1800 groups of samples is formed.

[0064] Please refer to Figure 5 , Figure 5 is the flow chart of the mode number calibration of the traditional VMD method in the application.

[0065] The mode number calibration based on the correlation coefficient is carried out in the following way: in the VMD decomposition process, the selection of the mode number has a significant influence on the final decomposition effect. If the set value is too large, the signal is easy to be over-decomposed, and false modes are generated; and when the setting is too small, the phenomenon of insufficient decomposition may occur. Therefore, aiming at this problem, an adaptive mode number determination method based on the correlation coefficient is proposed. The method uses the Pearson correlation coefficient to measure the linear correlation between two signals and , and the calculation formula is as follows:

[0066] ;

[0067] The application proposes a mode number estimation model based on the Pearson correlation coefficient, the target of which is to minimize the effective mode number , and the mode selection is carried out according to the following model:

[0068] ;

[0069] wherein, is the Pearson correlation coefficient function; is the original signal; is the first modal component obtained by VMD decomposition; is the residual signal after decomposition; is the first modal component; is the correlation coefficient between the first modal component and the original signal; is the correlation coefficient between the residual signal and the original signal; is the residual correlation coefficient decay factor. According to the model setting, the optimal modal number can be automatically selected, and the specific process is as follows:

[0070] (1) Set the initial modal number , and perform preliminary VMD decomposition on the signal to be analyzed;

[0071] (2) For the current modal number, calculate the Pearson correlation coefficient between each modal component and the original signal , and further calculate the correlation coefficient between the residual signal and the original signal ;

[0072] (3) If the correlation of the residual signal is greater than the smallest correlation coefficient among all modal components, i.e. , it indicates that there is still under-decomposition phenomenon, so K is increased by 1, and the second step is returned to continue iteration;

[0073] (4) If is satisfied, and , the current modal number is a reasonable upper limit, which can be saved as the optimal value.

[0074] (5) After the stop condition is satisfied, the current is output as the final estimated optimal modal number, and the program is terminated.

[0075] Please refer to Figure 6 , Figure 6 is the flowchart of the peacock optimization algorithm (POA) used in the present application to optimize the penalty factor and the fidelity coefficient .

[0076] The peacock optimization algorithm is introduced to optimize the fidelity coefficient τ and the penalty factor α in the following way: on the basis of the determined modal number , two key parameters in VMD decomposition still need to be further determined: the penalty factor and the fidelity coefficient . These two parameters have a great influence on the decomposition quality and are directly related to the final obtained modal components and the residual signal ​accuracy. To improve the precision of signal decomposition and feature extraction effect, it is necessary to optimize and , minimize the residual signal as much as possible, so as to improve the accuracy of subsequent analysis or prediction task. For this purpose, a parameter estimation method based on peacock optimization algorithm (POA) is introduced. The method takes minimizing the energy of residual signal as the goal, and the optimization index is VMD residual evaluation function :

[0077] ;

[0078] wherein, is the length of the original signal; is the original input signal; is the parameter decomposition obtained by the first modal function. The peacock optimization algorithm (POA) realizes the optimization search by simulating the process of male peacock showing tail feathers to attract female peacock. POA algorithm contains two key links of display behavior and updating mechanism, and has the advantages of fast convergence speed and strong global search ability. Assuming that in D-dimensional space, the population is composed of peacock individuals, the position of each peacock represents a set of parameter combinations to be optimized (penalty factor and fidelity coefficient), and each position corresponds to a fitness value (i.e. ). The peacock individuals attract the positions of superior individuals through display behavior, and gradually update the positions and optimize the parameters. The specific updating formula is as follows:

[0079] ;

[0080] In the formula: is the position of the th peacock individual in the th iteration, is the position of the optimal individual in the current population, is a random number in [0,1], is a standard normal random number, is a decay factor. In the VMD algorithm, the parameter combination will significantly affect the decomposition result. In order to realize efficient parameter optimization, each group is regarded as a "individual" in the POA algorithm, and the residual error index generated thereby is taken as the fitness function for evaluation. With the help of POA algorithm, global search and update of parameter space can be realized, and better solution can be obtained. In this embodiment, in order to realize efficient parameter optimization, each group is regarded as a "individual" in the POA algorithm, and the residual error index The fitness function is evaluated. The POA algorithm is used to search and update the parameter space globally, and a better solution can be obtained. The basic steps are:

[0081] (1) The optimal modal number of the current signal is determined by using the modal number rate limiting method based on the correlation coefficient in the S4 step ;

[0082] (2) A plurality of candidate parameter combinations are generated according to the value , and are used as individuals for VMD decomposition, calculation of the corresponding residual signal and obtaining of the evaluation index ; ;

[0083] (3) The position with the optimal fitness is set as ;

[0084] (4) A new position is generated according to the peacock optimization strategy , and residual calculation and index evaluation are performed again;

[0085] (5) If a better individual appears, update ;

[0086] (6) If the stopping criterion is not met, return to step (4) and continue; if the termination condition is met, take the current parameter as the optimal parameter of OVMD.

[0087] Please refer to Figure 7 and Figure 8 , Figure 7 and Figure 8 to show the confusion matrix and precision-recall curve after classification and recognition by the gradient boosting tree algorithm (GBDT).

[0088] Using the improved VMD method to decompose the signal can obtain a plurality of signal component functions. However, if the signal component functions are directly recognized, the recognition efficiency will be reduced due to too many information sequences. In order to improve the recognition efficiency, the embodiment proposes a method for extracting feature indexes from the signal component functions for recognition. The extracted features include 10 time domain features and 10 frequency domain features. The 10 time domain features are: mean, standard deviation, minimum value, maximum value, root mean square value, skewness, kurtosis, peak value, margin, and pulse index. The 10 frequency domain features are: spectral mean, spectral standard deviation, spectral energy, spectral center, spectral flatness, spectral roll-off point, spectral skewness, frequency kurtosis, spectral drop, and spectral coefficient of variation. From the signal component functions of the improved VMD, the above 20 indexes are selected for feature extraction, and 80 feature vectors are extracted for each sample to form a feature vector set of the original sample data set. Then, the feature vector set is input into the GBDT algorithm for classification and recognition:

[0089] The core idea of ​​GBDT is to optimize the model by gradually building decision trees, so that each new tree added at a time is based on reducing the error of the previous model. The step model is as follows:

[0090] ;

[0091] in, It is the previous round model, Indicates the decision trees; the final GBDT model can be expressed as a CART additive model:

[0092] ;

[0093] in, Indicates the A decision tree, Representative The parameters of the decision tree, Represents the number of decision trees. In order to optimize the general loss function, GBDT uses the gradient value of the loss function to build a model. superior, is the eigenvector, is the corresponding category. The algorithm flow of GBDT classification problem is as follows:

[0094] (1) Initialize model parameters: ;

[0095] (2) For , calculate the probability that each sample belongs to each category:

[0096] ;

[0097] (3) For , calculate the probability residual:

[0098] ;

[0099] And Fit a regression tree and get the mth round The leaf node area of ​​the regression tree of the class ,in .

[0100] (4) For calculate The value on:

[0101] ;

[0102] Finally, the calculated Models of each type:

[0103] ;

[0104] The feature vectors extracted from the signal component function are trained and classified by using the GBDT algorithm. When the 1800 sets of feature vectors after feature extraction are classified by using the GBDT algorithm, 70% of each type is used as a training set, and 30% is used as a test set, that is, 1260 sets of training set samples and 540 sets of test set samples. According to the attached table, it can be seen that the training set and the test set are well separated. Figure 7 It can be concluded that the method described in this embodiment correctly identifies 268 noisy leakage signals and 269 measured noise signals, identifies 2 measured noise signals as noisy leakage signals, and identifies 1 noisy leakage signal as a measured noise signal, with an accuracy of 99.44%. According to the attached table, it can be seen that the training set and the test set are well separated. Figure 8 It can be concluded that the precision and recall of the prediction results are very close to 1. The above proves that the method described in this embodiment has a high recognition accuracy.

[0105] Through the method described in this embodiment, a data set simulating the real offshore platform pipeline leakage signal is successfully constructed, and the problems of low recognition accuracy and insufficient robustness of traditional machine learning methods are solved, providing an important basis for the structural analysis and optimization of offshore platforms.

[0106] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, substitutions and alterations can be made hereto without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A pipeline leakage identification method combining OVMD and GBDT algorithm, characterized in that, The method comprises the following steps: S1, a pipeline leakage model is built, an acoustic emission sensor is arranged on the measured pipeline, and multiple groups of pipeline leakage signals are collected; S2, at least two acoustic emission sensors are arranged on the pipeline on the offshore platform, and actual background noise signals are collected; S3, the actual background noise signals are injected into the pipeline leakage signals to construct an original sample data set; S4, decomposing the original sample data set constructed in step S3 using an OVMD method to obtain a plurality of signal component functions, the OVMD method comprising: introducing a Pearson correlation coefficient to select an optimal modal number , performing modal selection according to the following model: ； wherein is the Pearson correlation coefficient function, which is calculated as follows: ; is the original signal; is the first modal component obtained by VMD decomposition; is the residual signal after decomposition; is the correlation coefficient of the first modal component and the original signal; is the correlation coefficient of the residual signal and the original signal; is the residual correlation coefficient decay factor; according to the model setting, the optimal modal number can be automatically selected, and the specific process is as follows: (1) Set initial mode number performing a preliminary VMD decomposition on the signal to be analyzed; (2) For the current modal number, calculate the Pearson correlation coefficient between each modal component and the original signal , and further calculate the correlation coefficient between the residual signal and the original signal ; (3) If the correlation of the residual signal Greater than the smallest correlation coefficient among all modal components, that is, , indicating that there is still under-decomposition, then K needs to be increased by 1 and return to the second step to continue iteration; (4) If , and , the current modal number is a reasonable upper limit and can be saved as the optimal value; (5) After the stop condition is met, output the current as the final estimated optimal modal number, and terminate the program; Optimizing algorithm for peacock to use penalty factor With fidelity coefficient Optimization is performed, including: according to the obtained optimal modal number Generate multiple candidate parameter combinations , and carry out VMD decomposition as an individual, calculate the corresponding residual signal and obtain the evaluation index, select the position with the optimal fitness, and then generate a new position based on the peacock optimization strategy and perform residual calculation and index evaluation again; S5, feature indexes of the signal component function obtained in step S4 are extracted to form a feature vector set; S6, the GBDT algorithm is used for classification and identification of the extracted feature vector set to form a confusion matrix and a precision-recall rate curve. 2.The pipeline leakage identification method of claim 1, wherein, The pipeline leakage model in the S1 step comprises a pipeline, a pressure device and the acoustic emission sensor, the acoustic emission sensor is connected to a preamplifier, the preamplifier is connected to an acoustic emission detection device, and the acoustic emission detection device is connected to an industrial computer.

3. The pipeline leakage identification method of claim 2, wherein, The pipeline leakage model in the S1 step comprises two regions, one is an experimental section, a plurality of control valves are uniformly distributed in the experimental section for simulating and controlling the leakage condition, and the other is a buffer section connected to the pressure device through a flexible rubber pipeline, and the pressure device adopts an electric pressure self-control pressure test pump.

4. The pipeline leak identification method of claim 1, wherein, The acoustic emission sensors for collecting the actual background noise signals in the S2 step are consistent with those in the step S1, and the sampling mode is a long-term timing sampling mode.

5. The method according to claim 1, wherein, In the S3 step, the pipeline leakage signals in the step S1 are uniformly divided according to the length of the sampling points to obtain a plurality of leakage signal samples, the actual background noise signals collected in the S2 step are also divided into sampling point paragraphs with the same number as the pipeline leakage signals, noise signal samples with the same number as the leakage signal samples are constructed, finally, the noise signal samples are injected into the leakage signal samples to form the original sample data set containing the leakage signal samples and the noise signal samples.

6. The method according to claim 1, wherein, In the S5 step, the feature indexes of the signal component function are extracted, the time domain features and frequency domain features of the signal component function obtained in the S4 step are extracted, and the feature vector set of the original sample data set is constructed; wherein the time domain features include mean, standard deviation, minimum value, maximum value, root mean square value, skewness, kurtosis, peak value, margin, pulse index; the frequency domain features include spectral mean, spectral standard deviation, spectral energy, spectral center, spectral flatness, spectral roll-off point, spectral skewness, frequency kurtosis, spectral drop, spectral coefficient of variation.

7. The method according to claim 1, wherein, In the S6 step, the GBDT method is used for classification, 70% of the feature vector set is used as a training set, 30% is used as a test set, a confusion matrix and a precision-recall rate curve are formed as an identification result.

Images