Chemical pipeline leakage detection method and system based on optical fiber sensing

Abstract

The invention discloses a chemical pipeline leakage detection method and system based on optical fiber sensing in order to solve the problems of incomplete detection coverage, response lag, easiness in environmental interference, inaccurate positioning and the like in the prior art, and discloses a chemical pipeline leakage detection method and system based on optical fiber sensing. Obtaining an enhanced voiceprint signal; the method comprises the following steps: firstly, carrying out feature extraction to obtain leakage voiceprint features, and finally, carrying out leakage identification based on the leakage voiceprint features and by virtue of an artificial intelligence algorithm, and determining a leakage position by utilizing an optical signal, so that rapid detection and accurate positioning of chemical pipeline leakage can be realized. Therefore, compared with the traditional technology, the chemical pipeline leakage detection system does not need intensive manual inspection tour, not only is high in response speed and labor cost reduced, but also is high in environmental interference resistance, full in detection coverage and free of blind areas, and can accurately lock the leakage position, so that the efficiency and the accuracy of chemical pipeline leakage detection are improved.

Description

Technical Field

[0001] This invention belongs to the field of chemical pipeline leakage detection technology based on artificial intelligence, specifically relating to a chemical pipeline leakage detection method and system based on fiber optic sensing. Background Technology

[0002] In recent years, the rapid development of the chemical industry has led to a continuous expansion of the scale of chemical pipeline construction. The pipeline media are mostly flammable, explosive, toxic, and hazardous substances. Leaks can easily cause safety accidents, resulting in serious casualties and environmental damage. Therefore, rapid detection and accurate location of chemical pipeline leaks are crucial. Currently, existing solutions for detecting chemical pipeline leaks mainly include two types: one is the manual inspection mode, where staff periodically inspect the pipeline route, using visual observation and handheld detectors to determine if leaks are present; the other is the point-sensor monitoring mode, which involves installing pressure sensors, gas sensors, and other equipment at key pipeline nodes. When the sensors detect abnormal parameters, they issue alarm signals.

[0003] The aforementioned existing technologies have significant drawbacks. Manual inspection has limited coverage and slow response, making 24-hour real-time monitoring impossible. Furthermore, it is difficult and inefficient to operate in complex terrain or harsh environments. Point-based sensor monitoring, on the other hand, suffers from blind spots, high sensor deployment costs, and is susceptible to interference from environmental factors such as noise and temperature / humidity changes in chemical industrial parks, leading to a high false alarm rate. Additionally, point-based sensors struggle to accurately pinpoint leaks, failing to meet the core requirements of "real-time, comprehensive, accurate, and rapid" leak detection in chemical pipelines. Therefore, given these shortcomings, providing a fiber optic sensing-based leak detection method for chemical pipelines that offers full coverage without blind spots, high real-time performance, and accurate leak location has become an urgent problem to solve. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for detecting leaks in chemical pipelines based on fiber optic sensing, in order to solve the problems of incomplete detection coverage, delayed response, susceptibility to environmental interference, and inaccurate positioning in existing technologies.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, a method for detecting leaks in chemical pipelines based on fiber optic sensing is provided, including: The optical signal transmitted by the distributed optical fiber sensing system installed along the chemical pipeline is acquired, and the optical signal is processed by signal conversion to obtain the acoustic signal of the chemical pipeline. The signal decomposition method is used to perform iterative decomposition and denoising on the voiceprint signal. After each decomposition and denoising process, the pure component signal is obtained. The denoised voiceprint signal is reconstructed using each pure component signal. The input signal of the previous decomposition is subtracted from the pure component signal obtained in the previous decomposition and used as the input signal for the next decomposition. The denoised audio signal is subjected to global enhancement processing to obtain an enhanced audio signal; The enhanced voiceprint signal is subjected to feature extraction processing to obtain leakage voiceprint features; The leakage acoustic signature is input into the pipeline leakage identification model to obtain the leakage identification result of the chemical pipeline; The location of the leak in the chemical pipeline was determined based on the optical signal. Based on the leak identification results and the leak location, a pipeline leak detection result is generated.

[0006] Based on the above disclosure, this invention deploys a distributed optical fiber sensing system along the chemical pipeline. This system utilizes optical fibers as both the sensing medium and transmission channel to continuously sense acoustic signature signals around the pipeline. When a leak occurs in the chemical pipeline, the interaction between the leaking medium and the pipeline wall and surrounding environment generates characteristic acoustic signatures at specific frequencies. These acoustic signature signals are transmitted to the distributed optical fiber through the pipeline structure or surrounding environment. The optical fiber converts the acoustic signature signal into an optical signal and transmits it to the downstream end. Thus, by laying distributed optical fibers along the entire pipeline, the blind spot problem inherent in point-based monitoring can be avoided. After obtaining the optical signal and converting it into an acoustic signature signal, iterative decomposition and denoising processing can be performed, which involves subtracting the purified signal obtained from the previous decomposition from the input signal of the previous decomposition. The component signals are used as input signals for the next decomposition. Based on this, noise can be removed during the layer-by-layer decomposition process to continuously extract pure component signals. Then, the pure component signals obtained from the decomposition can be used to reconstruct the denoised waveform signal. Next, the denoised waveform signal is globally enhanced to obtain the enhanced waveform signal, and its features are extracted to obtain the leakage waveform features. Then, the leakage waveform features are input into the pipeline leakage identification model to obtain the leakage identification result of the chemical pipeline. After obtaining the leakage identification result, the optical time-domain reflectometry characteristics of distributed optical fiber can be used in combination with the received optical signal to calculate the leakage location. Finally, based on the leakage identification result and the leakage location, the pipeline leakage detection result can be generated, thereby realizing the leakage detection of chemical pipelines.

[0007] Through the above design, this invention acquires pipeline acoustic signals using distributed optical fiber eavesdropping technology. After denoising and enhancement processing, an enhanced acoustic signal is obtained. Next, feature extraction is performed to obtain leakage acoustic features. Finally, based on these features and aided by artificial intelligence algorithms, leak identification is performed, and the leak location is determined using optical signals. This enables rapid detection and precise location of leaks in chemical pipelines. Compared to traditional technologies, this invention eliminates the need for intensive manual inspections, offering faster response times and reduced labor costs. Furthermore, it exhibits strong resistance to environmental interference, provides full detection coverage without blind spots, and accurately pinpoints the leak location. Therefore, it improves the efficiency and accuracy of chemical pipeline leak detection, meeting the core requirements of "real-time, comprehensive, accurate, and rapid" leak detection. Consequently, this invention is highly suitable for large-scale application and promotion.

[0008] In one possible design, a signal decomposition method is used to iteratively decompose and denoise the voiceprint signal, including: Obtain the input signal at the k-th decomposition, where the initial value of k is 1, and when k is 1, the input signal at the k-th decomposition is the voiceprint signal; Variational mode decomposition is performed on the input signal at the k-th decomposition to obtain several mode components; Calculate the signal distribution distance between each modal component and the input signal at the k-th decomposition, and select the modal component with the smallest signal distribution distance as the pure modal component, and treat the remaining modal components as noise components. The noise component is locally denoised to obtain the denoised component. The pure modal component and the denoised component are then used to form the pure component signal after the kth decomposition. Determine whether the iterative decomposition stopping condition is met. The iterative decomposition stopping condition is that the first distance is greater than the second distance, and the first distance is the smallest signal distribution distance among the signal distribution distances of each modal component at the k-th decomposition, and the second distance is the largest signal distribution distance among the signal distribution distances of each modal component at the (k-1)-th decomposition. If not, then subtract the pure component signal after the kth decomposition from the input signal at the kth decomposition to obtain the input signal at the (k+1)th decomposition. Increment k by 1 and reacquire the input signal from the kth decomposition until the iterative decomposition stopping condition is met, thus obtaining several pure component signals. These pure component signals can then be used to reconstruct the denoised signal.

[0009] In a possible design, the signal distribution distance between each modal component and the input signal at the k-th decomposition is calculated, including: For any modal component, wavelet decomposition is performed on the any modal component and the input signal to obtain the first wavelet signal of the any modal component at different scales and the second wavelet signal of the input signal at different scales. Calculate the probability distribution distance between the first wavelet signal and the second wavelet signal at each scale; The distances of each probability distribution are weighted and summed to obtain the multi-scale weighted probability distribution distances. Bandpass filters of different frequency bands are constructed, and bandpass filters are used to perform bandpass filtering on any mode component and the input signal to obtain mode filter components and input filter signals of different frequency bands; The Barton coefficients between the modal filtering components and the input filtering signal for each frequency band are calculated, and the weighted sums of the Barton coefficients are obtained. Based on the weighted Bartholomew's coefficient, the energy distribution distance between any modal component and the input signal is determined; Calculate the structural distribution distance between any modal component and the input signal; Based on the multi-scale weighted probability distribution distance, the energy distribution distance, and the structural distribution distance, the signal distribution distance between any modal component and the input signal at the k-th decomposition is calculated.

[0010] In one possible design, local noise denoising is performed on the noise components, including: The noise components are subjected to singular value decomposition to obtain several SVD components; Calculate the first correlation coefficient between each SVD component and the noise component; Acquire calibration signal components, wherein the calibration signal components are obtained by iterative decomposition and denoising of the acoustic signature signal of a leak-free chemical pipeline; Calculate the second correlation coefficient between each SVD component and the calibration signal component; The difference between the second correlation coefficient and the first correlation coefficient of each SVD component is calculated and used as the initial singular value decomposition factor for each SVD component. From a number of initial singular value decomposition factors, the largest and smallest initial singular value decomposition factors are selected, and the singular value decomposition factor of each SVD component is calculated using the largest and smallest initial singular value decomposition factors. The SVD components are sorted in descending order of singular value decomposition factors to obtain a sorted sequence. Calculate the difference between the singular value decomposition factors of each pair of adjacent SVD components in the sorted sequence, and determine the sorting number of the SVD component corresponding to the largest difference. The signal is reconstructed using the first L SVD components in the sorted sequence to obtain the denoised component, where L is the sorting sequence number.

[0011] In one possible design, the denoised voiceprint signal is subjected to global enhancement processing to obtain an enhanced voiceprint signal, including: Perform a short-time Fourier transform on the denoised ripple signal to obtain the time-frequency matrix of the denoised ripple signal; Based on the time-frequency matrix, the first instantaneous energy of the denoised ripple signal after short-time Fourier transform in each STFT time frame is calculated. Obtain the pipeline leakage frequency band range, and based on the time-frequency matrix, calculate the second instantaneous energy of the denoised ripple signal within the pipeline leakage frequency band range at each STFT time frame after short-time Fourier transform. Using the first instantaneous energy and the second instantaneous energy, the initial signal weight of the denoised ripple signal at each STFT time frame is calculated; The initial signal weights of each STFT time frame are upsampled to obtain the signal weights of the denoised signal at each sampling time. The denoised audio signal is weighted using the weights of each signal to obtain the enhanced audio signal.

[0012] In one possible design, feature extraction processing is performed on the enhanced voiceprint signal to obtain leakage voiceprint features, including: The enhanced voiceprint signal is subjected to frequency domain feature extraction processing to obtain frequency domain features, wherein the frequency domain features include MFCC features; Multi-scale temporal feature extraction processing is performed on the enhanced voiceprint signal to obtain the temporal features of the enhanced voiceprint signal; The leakage acoustic signature is formed by using the frequency domain features and the time domain features.

[0013] In one possible design, the enhanced voiceprint signal undergoes multi-scale temporal feature extraction processing, including: The time-domain scaling factor was determined, and based on the time-domain scaling factor and the enhanced voiceprint signal, the time-domain voiceprint signal corresponding to different time-domain scaling factors was constructed. For any time-domain acoustic signature signal, obtain the first time-domain feature dimension and the second time-domain feature dimension; Based on the first time-domain feature dimension, a first time-domain feature vector and a second time-domain feature vector corresponding to any time-domain voiceprint signal are constructed, and based on the second time-domain feature dimension, a third time-domain feature vector and a fourth time-domain feature vector corresponding to any time-domain voiceprint signal are constructed. Calculate the first maximum distance between the first time-domain feature vector and the second time-domain feature vector, and the second maximum distance between the third time-domain feature vector and the fourth time-domain feature vector; Based on the first maximum distance, the similarity between the first time-domain feature vector and the second time-domain feature vector is calculated, and based on the second maximum distance, the similarity between the third time-domain feature vector and the fourth time-domain feature vector is calculated. Using the similarity between the first time-domain feature vector and the second time-domain feature vector, and the similarity between the third time-domain feature vector and the fourth time-domain feature vector, the fuzzy measure entropy of any time-domain voiceprint signal is calculated, and after all time-domain voiceprint signals have been polled, the multi-scale fuzzy measure entropy is obtained. The multi-scale fuzzy metric entropy is used as the temporal feature of the enhanced voiceprint signal.

[0014] Secondly, a chemical pipeline leakage detection system based on fiber optic sensing is provided, including: an AI processing unit and a distributed fiber optic sensing system set along the chemical pipeline, wherein the AI ​​processing unit includes a signal conversion unit, a signal denoising unit, a signal enhancement unit, a feature extraction unit, a leakage detection unit and a positioning unit. The signal conversion unit is used to acquire the optical signal transmitted by the distributed optical fiber sensing system installed along the chemical pipeline, and to perform signal conversion processing on the optical signal to obtain the acoustic signal of the chemical pipeline. The signal denoising unit is used to perform iterative decomposition and denoising processing on the voiceprint signal using a signal decomposition method, so as to obtain the pure component signal after each decomposition after the iterative decomposition and denoising processing, and to reconstruct the denoised voiceprint signal using each pure component signal. The input signal of the previous decomposition is subtracted from the pure component signal obtained in the previous decomposition and used as the input signal for the next decomposition. A signal enhancement unit is used to perform global enhancement processing on the denoised audio signal to obtain an enhanced audio signal. The feature extraction unit is used to perform feature extraction processing on the enhanced voiceprint signal to obtain leakage voiceprint features; The leakage detection unit is used to input the leakage acoustic features into the pipeline leakage identification model to obtain the leakage identification result of the chemical pipeline; The positioning unit is used to determine the location of the leak in the chemical pipeline based on the optical signal. The leak detection unit is also used to generate pipeline leak detection results based on the leak identification results and the leak location.

[0015] Thirdly, another fiber optic sensing-based chemical pipeline leak detection system is provided. Taking the device as an electronic device as an example, it includes a memory, a processor, and a transceiver that are connected in sequence. The memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the fiber optic sensing-based chemical pipeline leak detection method as described in the first aspect or any possible design of the first aspect.

[0016] Fourthly, a storage medium is provided, on which instructions are stored, which, when executed on a computer, perform the fiber optic sensing-based chemical pipeline leak detection method as described in the first aspect or any possible design of the first aspect.

[0017] Fifthly, a computer program product containing instructions is provided, which, when executed on a computer, causes the computer to perform the fiber optic sensing-based chemical pipeline leak detection method as described in the first aspect or any possible design of the first aspect.

[0018] Beneficial effects: (1) This invention collects pipeline acoustic signals through distributed optical fiber listening technology, and then obtains enhanced acoustic signals after denoising and enhancement processing; then, feature extraction is performed to obtain leakage acoustic features; finally, based on the leakage acoustic features and with the help of artificial intelligence algorithms, leakage is identified, and the leakage location is determined by using optical signals. In this way, rapid detection and accurate location of chemical pipeline leaks can be achieved. Therefore, compared with traditional technologies, this invention does not require intensive manual inspections, has a fast response speed, and reduces labor costs. At the same time, it has strong resistance to environmental interference, full detection coverage without blind spots, and can accurately locate the leakage location. Based on this, the efficiency and accuracy of chemical pipeline leak detection are improved, thereby meeting the core requirements of "real-time, comprehensive, accurate and fast" chemical pipeline leak detection. Therefore, this invention is very suitable for large-scale application and promotion. Attached Figure Description

[0019] Figure 1 A schematic flowchart illustrating the steps of a chemical pipeline leakage detection method based on fiber optic sensing provided in an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a chemical pipeline leakage detection system based on fiber optic sensing provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0021] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0022] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0023] Example: Before disclosing the leakage detection method provided in this embodiment, a chemical pipeline leakage detection system based on fiber optic sensing is provided. This system includes an AI processing unit, a smart inspection platform, and a distributed fiber optic sensing system installed along the chemical pipeline. Specifically, the distributed fiber optic sensing system includes distributed sensing fibers, which are arranged along the chemical pipeline, such as by attaching to the pipeline, wrapping around it, burying it, or suspending it. In this way, the distributed sensing fibers can be used as a sensing medium and transmission channel to sense acoustic signals around the pipeline throughout the entire process. That is, when a chemical pipeline leaks, the leaking medium interacts with the pipeline wall and the surrounding medium to generate characteristic acoustic signals of a specific frequency. These acoustic signals are transmitted to the distributed fibers through the pipeline structure or the surrounding environment. The fibers convert the acoustic signals into optical signals and transmit them to the back-end processing unit (AI processing unit).

[0024] Then, the AI ​​processing unit can use the leak detection method provided in this embodiment to perform signal denoising and enhancement, and extract the leak acoustic signature features. Then, it uses the trained artificial intelligence recognition model (i.e., the pipeline leak recognition model) to match and identify the features to obtain the leak identification result. Next, the AI ​​processing unit combines the optical time-domain reflectometry characteristics of distributed optical fiber and calculates the distance and location of the leak point based on the optical signal. Finally, it can use the aforementioned leak identification result and the distance and location of the leak point to generate the pipeline leak detection result and send it to the smart inspection platform. Based on this, staff can view the detection result in real time through the smart inspection platform. At the same time, the smart inspection platform will display the precise location of the leak point, the acoustic signature spectrum, and issue alarm prompts to notify staff to handle the situation in a timely manner.

[0025] Therefore, using this system for pipeline leak detection eliminates the need for intensive manual inspections compared to traditional technologies. It not only offers a faster response time and reduces labor costs, but also boasts strong resistance to environmental interference, provides full detection coverage without blind spots, and can accurately pinpoint the leak location. Based on these advantages, the efficiency and accuracy of chemical pipeline leak detection are improved, thus meeting the core requirements of "real-time, comprehensive, accurate, and rapid" chemical pipeline leak detection.

[0026] See Figure 1 As shown, the chemical pipeline leakage detection method based on fiber optic sensing provided in this embodiment can be operated, but is not limited to, on the distributed fiber optic sensing system, AI processing unit, and intelligent inspection platform. It is understood that the aforementioned execution entities do not constitute a limitation on the embodiments of this application. Accordingly, the operation steps of this method can be, but are not limited to, the steps S1 to S7 below.

[0027] S1. Acquire the optical signal transmitted by the distributed optical fiber sensing system installed along the chemical pipeline, and perform signal conversion processing on the optical signal to obtain the acoustic signature signal of the chemical pipeline; In specific implementation, as mentioned above, the distributed sensing optical fiber is laid along the entire pipeline. Therefore, when a series of ultra-short (approximately 10 nanoseconds) laser pulses are continuously emitted into the sensing optical fiber, the laser pulses will propagate in the optical fiber. Based on this, there is a data point every 1 meter (determined by the pulse width) on the optical fiber, and this data point contains the acoustic signature signal; Specifically, when a leak occurs at a certain point in the chemical pipeline, the characteristic acoustic wave generated by the leaking medium will act on the optical fiber near that point, causing micro-strain in the optical fiber. This micro-strain will modulate the phase or intensity of the "backscattered Rayleigh light" of the laser pulse that subsequently passes through that point. In this way, the optical fiber will transmit this data optical signal to the AI ​​processing unit, and the AI ​​processing unit will demodulate these modulated backscattered lights and convert them into electrical signals, thereby obtaining the acoustic signature signals corresponding to different data points.

[0028] After obtaining the acoustic fingerprint signal of the chemical pipeline, signal denoising can be performed, as shown in steps S2 and S3 below.

[0029] S2. A signal decomposition method is adopted to iteratively decompose and denoise the voiceprint signal, so that after each decomposition, a clean component signal is obtained. The denoised voiceprint signal is reconstructed using each clean component signal. The input signal of the previous decomposition is subtracted from the clean component signal obtained in the previous decomposition, and the result is used as the input signal for the next decomposition. In practical applications, due to the presence of various environmental noises, denoising is required before leakage identification to reduce the interference of noise on subsequent leakage identification. This embodiment provides an iterative decomposition and denoising method that first performs fine decomposition and then performs local denoising. The process can be, but is not limited to, the steps S21 to S27 below.

[0030] S21. Obtain the input signal at the kth decomposition, where the initial value of k is 1, and when k is 1, the input signal at the kth decomposition is the voiceprint signal; in this embodiment, the voiceprint signal in step S1 is used as the input signal at the first decomposition; then, variational mode decomposition is performed on the input signal to obtain several mode components, the process of which is shown in step S22 below.

[0031] S22. Perform variational mode decomposition on the input signal at the k-th decomposition to obtain several modal components. In specific applications, for example, but not limited to, performing variational mode decomposition on the input signal at the k-th decomposition according to a preset number of decomposition modes to obtain several modal components; at the same time, for example, the preset number of decomposition modes is 2, therefore, 2 modal components can be obtained.

[0032] After obtaining two modal components, the pure modal components can be screened, as shown in step S23 below.

[0033] S23. Calculate the signal distribution distance between each modal component and the input signal at the k-th decomposition, and select the modal component with the smallest signal distribution distance as the pure modal component, and treat the remaining modal components as noise components; in specific applications, this embodiment provides a method for calculating the signal distribution distance of multiple indices; specifically, taking any modal component as an example, the calculation process of its signal distribution distance with the input signal can be described, but is not limited to, the steps S23a to S23h below.

[0034] S23a. For any modal component, wavelet decomposition is performed on the any modal component and the input signal to obtain the first wavelet signal of the any modal component at different scales and the second wavelet signal of the input signal at different scales. In specific implementation, since traditional distance calculation only considers the global distribution and ignores the local features of the signal at different scales, this embodiment performs wavelet decomposition on the any modal component and the input signal to calculate the probability distribution distance between the signals at each different scale (for example, the scale can be set to 3, and wavelet decomposition is a common technique in signal processing, and its decomposition principle will not be elaborated here). The calculation process of the probability distribution distance is shown in step S23b below.

[0035] S23b. Calculate the probability distribution distance between the first wavelet signal and the second wavelet signal at each scale; in this embodiment, for the first wavelet signal and the second wavelet signal at any scale, the first wavelet signal and the second wavelet signal are first discretized and normalized to obtain the first normalized signal and the second normalized signal; then, the probability distribution distance between the first normalized signal and the second normalized signal is used as the probability distribution distance between the wavelet signal and the second wavelet signal at that scale.

[0036] Specifically, the process of discretizing and normalizing the first wavelet signal is as follows: sum the signal points in the first wavelet signal to obtain the signal sum; then, divide the signal points in the first wavelet signal by the signal sum to obtain the first normalized signal; simultaneously, the formula for calculating the probability distribution distance between the first normalized signal and the second normalized signal is: ; In the formula, This represents the distance between the probability distributions of the first and second normalized signals. Let q be the q-th signal point in the first normalized signal. This represents the q-th signal point in the second normalized signal. The length of the first normalized signal.

[0037] Thus, based on the aforementioned formula, after calculating the probability distribution distance between the first wavelet signal and the second wavelet signal at each scale, a weighted summation can be performed to obtain the multi-scale weighted probability distribution distance, as shown in step S23c below.

[0038] S23c. Weighted summation of the distances of each probability distribution is performed to obtain the multi-scale weighted probability distribution distance. In this embodiment, the weight of each scale can be calculated based on the energy proportion. For example, taking the first wavelet signal at any scale as an example, the sum of squares of all signal points of the first wavelet signal at any scale is calculated as the energy at that scale. Similarly, the energy of the remaining scales is calculated and summed to obtain the total energy. Finally, the energy at any scale is divided by the total energy to obtain the weight of that scale.

[0039] Thus, based on the weights calculated above, the weighted summation of the distances of each probability distribution can be completed; then, considering the differences in importance of different frequency bands, this embodiment introduces frequency band energy weighting to calculate the energy distribution distance, the process of which is shown in steps S23d to S23f below.

[0040] S23d. Construct bandpass filters for different frequency bands, and use the bandpass filters to perform bandpass filtering on any modal component and the input signal to obtain modal filter components and input filter signals of different frequency bands. In specific implementation, for example, but not limited to, using a Butterworth bandpass filter to perform bandpass filtering on any modal component and the input signal, that is, by setting the low cutoff frequency, high cutoff frequency, sampling frequency and filter order of the Butterworth bandpass filter, bandpass filtering of different frequency bands can be performed (e.g., setting the low cutoff frequency and high cutoff frequency to 0Hz and 50Hz, thereby separating the signal in the 0-50Hz frequency band), thereby dividing any modal component and the input signal into modal filter components and input filter signals of different frequency bands; then, the Bartholomew's coefficients of the modal filter components and input filter signals of different frequency bands can be calculated, thereby weighted summing of the Bartholomew's distances of different frequency bands to obtain the energy distribution distance between any modal component and the input signal.

[0041] The calculation process for the Bach coefficients of different frequency bands is shown in step S23e below.

[0042] S23e. Calculate the Barton coefficient between the modal filter component and the input filter signal for each frequency band, and sum the weighted Barton coefficients to obtain the weighted Barton coefficient. In this embodiment, for the modal filter component and the input filter signal in the same frequency band, the probability density of the modal filter component and the input filter signal can be calculated. Then, the Barton coefficient of the modal filter component and the input filter signal in the same frequency band can be calculated using the calculated probability density. Of course, the Barton coefficient is a commonly used intermediate parameter for signal similarity measurement, and its calculation principle will not be elaborated here.

[0043] Once the Barton coefficients between the modal filter components and the input filter signal in each frequency band are obtained, the Barton coefficients can be weighted and summed according to the pre-set frequency band weights. For example, but not limited to, the frequency bands can be divided into three bands: low frequency (0-50Hz), mid frequency (50-200Hz, the main leakage characteristic band), and high frequency (200-500Hz, mainly noise). In this way, a high weight (e.g., 0.5) can be set for the mid frequency band, while a low weight (0.3 and 0.2) can be set for the low frequency and high frequency bands.

[0044] Thus, after obtaining the weighted Bartholomew's coefficients, the energy distribution distance between any modal component and the input signal can be calculated based on these coefficients, as shown in step S23f below.

[0045] S23f. Based on the weighted Bartholomew's coefficient, determine the energy distribution distance between any modal component and the input signal; in a specific implementation, the energy distribution distance is: -ln(WBC), where WBC is the weighted Bartholomew's coefficient.

[0046] Based on this, after calculating the energy distribution distance between any modal component and the input signal, the structural distribution distance between them can be calculated, as shown in step S23g below.

[0047] S23g. Calculate the structural distribution distance between any modal component and the input signal. In specific applications, for example, but not limited to, first calculate the first signal mean and first signal standard deviation of any modal component; then, calculate the second signal mean and second signal standard deviation of the input signal; then, calculate the first difference between each signal point in any modal component and the first signal mean, and divide each first difference by the first signal standard deviation to obtain several first intermediate parameters; next, calculate the second difference between each signal point in the input signal and the second signal mean, and divide each second difference by the second signal standard deviation to obtain several second intermediate parameters; then, subtract the corresponding second difference from each first difference to obtain several parameter differences; finally, take the absolute value of several parameter differences and sum them to obtain the structural distribution distance; thus, the structural distribution distance can quantify the waveform structural differences between two signals at the same time point.

[0048] Thus, after calculating the structural distribution distance, the signal distribution distance between any modal component and the input signal can be calculated by combining the aforementioned energy distribution distance and the multi-scale weighted probability distribution distance, as shown in step S23h below.

[0049] S23h. Based on the multi-scale weighted probability distribution distance, the energy distribution distance, and the structural distribution distance, the signal distribution distance between any modal component and the input signal at the k-th decomposition is calculated; in this embodiment, the signal distribution distance between the two can be obtained by weighted summation of the aforementioned three distances.

[0050] Therefore, after calculating the signal distribution distance between the two modal components and the input signal at the kth decomposition through the aforementioned steps S23a to S23h, the modal component with the smallest signal distribution distance can be selected as the pure modal component, while the remaining modal component is used as the noise component. Then, the noise component can be denoised so that the denoised component and the pure modal component can be used to form the pure component signal after the kth decomposition. The process is shown in step S24 below.

[0051] S24. Perform local denoising on the noise components to obtain denoised components, and use the pure mode components and denoised components to form the pure component signal after the kth decomposition; In this embodiment, this embodiment provides an improved SVD denoising method, which can adaptively select SVD components to reconstruct the signal according to the actual signal, thereby achieving adaptive denoising; wherein, the improved SVD denoising process can be, but is not limited to, as shown in the following steps S24a to S24i.

[0052] S24a. Perform singular value decomposition on the noise components to obtain several SVD components; after completing the singular value decomposition of the noise components, the correlation coefficient can be calculated, as shown in step S24b below.

[0053] S24b. Calculate the first correlation coefficient between each SVD component and the noise component; in this embodiment, for example, but not limited to, the covariance and standard deviation method can be used to calculate the first correlation coefficient between each SVD component and the noise component; of course, the correlation coefficient is a commonly used tool for signal similarity, and its calculation process will not be described in detail.

[0054] After obtaining several first correlation coefficients, the calibration signal components can be obtained so that the correlation coefficients between each SVD component and the calibration signal components can be calculated. The process is shown in steps S24c and S24d below.

[0055] S24c. Obtain the calibration signal component, wherein the calibration signal component is obtained by iteratively decomposing and denoising the acoustic signal of the leak-free chemical pipeline; in specific implementation, iteratively decomposing and denoising the acoustic signal of the leak-free chemical pipeline is performed to obtain the modal component at each decomposition; then, based on the method of the aforementioned step S23, the calibration pure component and calibration noise component are screened out; finally, the calibration noise component at the kth decomposition is used as the aforementioned calibration signal component.

[0056] After obtaining the calibration signal components, the second correlation coefficient can be calculated, as shown in step S24d below.

[0057] S24d. Calculate the second correlation coefficient between each SVD component and the calibration signal component.

[0058] After calculating each of the second correlation coefficients, the initial singular value decomposition factor of each SVD component can be calculated by combining each of the first correlation coefficients, as shown in step S24e below.

[0059] S24e. Calculate the difference between the second correlation coefficient and the first correlation coefficient of each SVD component, and use it as the initial singular value decomposition factor for each SVD component.

[0060] After calculating the initial singular value decomposition factor of each SVD component based on the aforementioned step S24e, the final singular value decomposition factor of each SVD component can be calculated based on this, as shown in step S24f below.

[0061] S24f. From a number of initial singular value decomposition factors, select the largest and smallest initial singular value decomposition factors, and use the largest and smallest initial singular value decomposition factors to calculate the singular value decomposition factor of each SVD component. In this embodiment, for any SVD component, first subtract the smallest singular value decomposition factor from the initial singular value decomposition factor of that SVD component to obtain a first decomposition factor difference; then, subtract the smallest singular value decomposition factor from the largest singular value decomposition factor to obtain a second decomposition factor difference; finally, use the ratio between the first decomposition factor difference and the second decomposition factor difference as the final singular value decomposition factor of that SVD component.

[0062] Thus, after calculating the singular value decomposition factor of each SVD component based on the aforementioned method, the SVD components can be screened based on this, as shown in steps S24g and S24h below.

[0063] S24g. Sort each SVD component in descending order of singular value decomposition factor to obtain a sorted sequence. After sorting each SVD component, the difference between the singular value decomposition factors of two adjacent SVD components in the sorted sequence can be used to filter the SVD components. The process is shown in step S24h below.

[0064] S24h. The difference between the singular value decomposition factors of each pair of adjacent SVD components in the sorted sequence is calculated, and the sorting number of the SVD component corresponding to the largest difference is determined. In specific implementation, the difference between the singular value decomposition factor of the first SVD component and the singular value decomposition factor of the second SVD component in the sorted sequence is calculated first; then, the difference between the singular value decomposition factor of the second SVD component and the singular value decomposition factor of the third SVD component is calculated, and so on, until the difference between the singular value decomposition factor of the penultimate SVD component and the singular value decomposition factor of the last SVD component is calculated. Based on this, the sorting number of the SVD component corresponding to the largest difference can be determined. For example, the difference between the singular value decomposition factor of the fifth SVD component and the singular value decomposition factor of the sixth SVD component is the largest. At this time, the sorting number of the SVD component corresponding to the largest difference is 6, that is, L is 6. Therefore, the first 6 SVD components in the sorted sequence can be selected for signal reconstruction, and the process is shown in step S24i below.

[0065] S24i. Using the first L SVD components in the sorted sequence, the signal is reconstructed to obtain the denoised component, where L is the sorting number.

[0066] Based on the aforementioned steps S24a to S24i, the noise component is denoised. After obtaining the denoised component, it can be combined with the aforementioned pure modal component (the two are added together) to form the pure component signal at the kth decomposition. Then, it can be determined whether the iterative decomposition stopping condition is met, as shown in step S25 below.

[0067] S25. Determine whether the iterative decomposition stopping condition is met. The iterative decomposition stopping condition is that the first distance is greater than the second distance, and the first distance is the smallest signal distribution distance among the signal distribution distances of each modal component at the k-th decomposition, and the second distance is the largest signal distribution distance among the signal distribution distances of each modal component at the (k-1)-th decomposition. The aforementioned iterative stopping condition can be understood as follows: if the similarity between the two modal components obtained by the current decomposition and the input signal is very low (i.e., the Bach distance is very large, that is, larger than the maximum value of the Bach distance between the two components obtained by the previous decomposition), it indicates that the remaining signal is mainly noise and there are no components similar to the input signal. At this time, the iteration stops.

[0068] If the aforementioned iterative decomposition stopping condition is not met, the decomposition needs to continue, as shown in steps S26 and S27 below.

[0069] S26. If not, then subtract the pure component signal after the kth decomposition from the input signal at the kth decomposition to obtain the input signal at the (k+1)th decomposition. In this embodiment, subtracting the pure component signal after the kth decomposition from the original signal is equivalent to completing one iteration of decomposition denoising and extracting a pure component signal. At this time, the remaining signal is used as the input signal to continue iterative decomposition denoising until the aforementioned iterative decomposition stopping condition is met, and then the denoising of the entire voiceprint signal can be completed. The process is shown in step S27 below.

[0070] S27. Increment k by 1 and reacquire the input signal at the kth decomposition until the iterative decomposition stopping condition is met, thus obtaining several pure component signals, which can be used to reconstruct the denoised signal.

[0071] Therefore, through the aforementioned steps S21 to S27, this embodiment provides an iterative decomposition and denoising method based on first performing fine decomposition based on variational mode decomposition, and then using the modified SVD method for local denoising. Compared with traditional variational mode decomposition, this embodiment does not require pre-setting the number of decomposition modes to complete the screening of effective components. Therefore, it avoids the problem that manually setting the number of decomposition layers has a significant impact on the decomposition effect. At the same time, performing local denoising on the noise components after each decomposition can preserve the original effective signal to the greatest extent, thus further improving the effectiveness of denoising.

[0072] After completing the iterative decomposition and denoising, global enhancement processing can be performed, as shown in step S3 below.

[0073] S3. Perform global enhancement processing on the denoised ripple signal to obtain an enhanced acoustic ripple signal; in specific implementation, in order to further suppress the broadband noise or residual interference that was not removed in the aforementioned denoising, and to perform targeted enhancement according to the target acoustic ripple frequency band (i.e., leakage frequency band), this embodiment performs energy weight-based spectrum shaping on the denoised ripple signal at the full frequency band level to achieve signal enhancement. The specific process can be, but is not limited to, the steps S31 to S36 below.

[0074] S31. Perform a short-time Fourier transform on the denoised ripple signal to obtain the time-frequency matrix of the denoised ripple signal; in this embodiment, each row of the time-frequency matrix represents the value (named time-frequency value) of the denoised ripple signal at the same frequency in different STFT time frames after the denoised ripple signal has undergone a short-time Fourier transform and the modulus of its real and imaginary parts has been taken; for example, the first row is represented as: X(T1,fN), X(T2,fN), ..., X(TZ,fN), where X(TZ,fN) represents the denoised ripple signal after a short-time Fourier transform and the modulus of its real and imaginary parts has been taken; The time-frequency matrix is ​​the value of the denoised signal after undergoing a short-time Fourier transform and taking the modulus of its real and imaginary parts, at a time frame TZ and a frequency of fN. Similarly, each column of the time-frequency matrix represents the value of the denoised signal after undergoing a short-time Fourier transform and taking the modulus of its real and imaginary parts at different frequencies within the same STFT time frame. That is, the first column can be represented as: X(T1,f1), X(T1,f2), ..., X(T1,fN). Therefore, X(T1,fN) represents the value at a time frame T1 and a frequency of fN.

[0075] Thus, after obtaining the time-frequency matrix, the instantaneous energy can be calculated, as shown in step S32 below.

[0076] S32. Based on the time-frequency matrix, calculate the first instantaneous energy of the denoised ripple signal after short-time Fourier transform in each STFT time frame; in this embodiment, summation is performed on each column of the time-frequency matrix to obtain the first instantaneous energy corresponding to each STFT time frame; then, the pipeline leakage frequency band range is obtained, and based on this, the second instantaneous energy within the pipeline leakage frequency band leakage range is determined, as shown in step S33 below.

[0077] S33. Obtain the pipeline leakage frequency band range, and based on the time-frequency matrix, calculate the second instantaneous energy of the denoised ripple signal within the pipeline leakage frequency band range at each STFT time frame after short-time Fourier transform. In this embodiment, it is assumed that the pipeline leakage frequency band range is n1 to n2 (this frequency band range can be preset according to historical leakage acoustic signals). For time frame T1, the time-frequency values ​​corresponding to frequencies n1 to n2 are selected from the first column of the time-frequency matrix, that is, the time-frequency values ​​between X(T1,fn1) and X(T1,fn2) are obtained and summed to obtain the second instantaneous energy within the pipeline leakage frequency band range at time frame T1. Thus, after calculating the second instantaneous energy within the pipeline leakage frequency band range at each STFT time frame based on the aforementioned method, the initial signal weight of the denoised ripple signal at each STFT time frame can be calculated, as shown in step S34 below.

[0078] S34. Using each first instantaneous energy and each second instantaneous energy, calculate the initial signal weight of the denoised ripple signal in each STFT time frame. In this embodiment, for any STFT time frame, divide the second instantaneous energy of that STFT time frame by its corresponding first instantaneous energy to obtain the weight factor of that STFT time frame. After polling all STFT time frames, the weight factor of the denoised ripple signal in each STFT time frame can be obtained. Then, normalize each weight factor to obtain the initial signal weight of the denoised ripple signal in each STFT time frame. Finally, upsampling can be performed to restore the length of the weight to the signal length corresponding to the denoised ripple signal.

[0079] The upsampling process is shown in step S35 below.

[0080] S35. Upsample the initial signal weights of each STFT time frame to obtain the signal weights of the denoised signal at each sampling time; in specific implementation, for sampling times at the Tz-th and Tz+1-th time frames... The signal weights can be obtained by upsampling using, but are not limited to, the following formula.

[0081] ; In the formula, Indicates the sampling time The corresponding signal weights, This represents the signal weights corresponding to the (Tz+1)th time frame and the (Tz)th time frame.

[0082] Thus, the signal weight of the noise-removed ripple signal at each sampling time can be calculated using the aforementioned formula; then, the signal can be weighted based on the signal weight at each sampling time, as shown in step S36 below.

[0083] S36. Using the weights of each signal, the denoised signal is weighted to obtain the enhanced voiceprint signal; in specific implementation, for example, but not limited to, the denoised signal can be transposed first, and then the transposed denoised signal can be multiplied by the signal weight at the corresponding sampling time to obtain the enhanced voiceprint signal.

[0084] After the global enhancement of the noise-reduced signal is completed through the aforementioned steps S31 to S36, feature extraction processing can be performed, as shown in step S4 below.

[0085] S4. Perform feature extraction processing on the enhanced voiceprint signal to obtain leakage voiceprint features; in specific implementation, for example, but not limited to, first perform frequency domain feature extraction processing on the enhanced voiceprint signal to obtain frequency domain features; then, perform multi-scale time domain feature extraction processing on the enhanced voiceprint signal to obtain time domain features of the enhanced voiceprint signal; finally, the leakage voiceprint features can be formed by using the frequency domain features and the time domain features.

[0086] In this embodiment, the frequency domain features described may include, but are not limited to, MFCC features.

[0087] Meanwhile, this embodiment uses the entropy of the multi-scale fuzzy measure of the enhanced voiceprint signal as a temporal feature to describe the complexity and irregularity of the signal. That is, it quantifies the complexity and irregularity of the enhanced voiceprint signal at different time resolutions. In this way, by comparing the significant difference in this feature between the normal state and the leakage state (especially small leaks that are difficult to detect), the leakage can be accurately identified.

[0088] Optionally, the extraction process of the aforementioned time-domain features can be, but is not limited to, the steps S41 to S47 below.

[0089] S41. Determine the time-domain scaling factor, and construct the time-domain voiceprint signal corresponding to different time-domain scaling factors based on the time-domain scaling factor and the enhanced voiceprint signal.

[0090] In this embodiment, the temporal voiceprint signal corresponding to any temporal scale factor can be represented as: ; In the formula, Represents any of the time-domain scale factors, Let represent the time-domain voiceprint signal corresponding to any time-domain scale factor, and b represent the b-th data point in the time-domain voiceprint signal. V represents the d-th signal point in the enhanced voiceprint signal, and V represents the length of the enhanced voiceprint signal. In this embodiment, the time-domain scaling factor can be set to multiple consecutive values, such as 1 to 3. Of course, its value can be set according to actual use, which will not be elaborated here.

[0091] Thus, based on the aforementioned formula, after constructing the temporal voiceprint signals corresponding to different temporal scale factors, the temporal feature dimension can be obtained, as shown in step S42 below.

[0092] S42. For any time-domain voiceprint signal, obtain the first time-domain feature dimension and the second time-domain feature dimension; in specific implementation, for example, the first time-domain feature dimension is 2 and the second time-domain feature dimension is 3. Based on this, after obtaining the two time-domain feature dimensions, the time-domain feature vector can be constructed, and the process is shown in step S43 below.

[0093] S43. Based on the first time-domain feature dimension, construct the first time-domain feature vector and the second time-domain feature vector corresponding to any time-domain voiceprint signal, and based on the second time-domain feature dimension, construct the third time-domain feature vector and the fourth time-domain feature vector corresponding to any time-domain voiceprint signal.

[0094] For example, the expression for the first time-domain eigenvector is: ; In the formula, Represents the first time-domain eigenvector. Represents the first time-domain feature dimension. This represents the b-th data point in any of the time-domain acoustic signature signals. This represents the (b+m-1)th data point in any given time-domain acoustic signature signal. For m consecutive The mean of b, where b = 1, 2, 3, ..., B-m+1, where B is the length of any time-domain acoustic signature signal.

[0095] in, In the formula, Let b be the (b+r)th data point in any given time-domain acoustic signature signal.

[0096] Thus, it can be seen from the above formula that if m is 2 and b is 1, Then it contains Two data points; similarly, when b is any other value, The same applies to the corresponding data points, which will not be elaborated upon here.

[0097] Meanwhile, the construction process of the second time-domain feature vector is the same as that of the first time-domain feature vector, except that the value of b is different. That is, the value is taken from different data points in any time-domain voiceprint signal until the value has the same length as the first time-domain feature vector (such as taking the value from 2 to B-m+2, that is, taking the value from the second data point of any time-domain voiceprint signal as the starting point).

[0098] Of course, the construction process of the third and fourth time-domain feature vectors under the second time-domain feature dimension is the same, and will not be repeated here.

[0099] After obtaining the aforementioned four temporal feature vectors, the maximum distance can be calculated, as shown in step S44 below.

[0100] S44. Calculate the first maximum distance between the first time-domain feature vector and the second time-domain feature vector, and the second maximum distance between the third time-domain feature vector and the fourth time-domain feature vector.

[0101] In this embodiment, the first maximum distance can be calculated using, for example, but not limited to, the following formula; ; In the formula, This represents the relationship between the data point with index b in the first time-domain feature vector and the second time-domain feature vector. The index is the first maximum distance between c data points, where b,c = 1, 2, 3, ..., U, and U is the length of the first temporal feature vector. Based on this, it has been explained earlier that when the index is b, it contains multiple data points. Therefore, for each data point with index b, the distance between the index and the data points is... For each data point with index c, calculate the difference, take the absolute value, and then take the largest one as the [value]. .

[0102] Thus, the aforementioned Then it exists Two data points, then There are also two data points (e.g., for example, for...) Therefore, it is Then take the one with the largest absolute value as the [value]. .

[0103] Therefore, based on the aforementioned formula, the first maximum distance between the first time-domain feature vector and the second time-domain feature vector, and the second maximum distance between the third time-domain feature vector and the fourth time-domain feature vector can be calculated; then, similarity can be calculated based on this, as shown in step S45 below.

[0104] S45. Based on the first maximum distance, calculate the similarity between the first time-domain feature vector and the second time-domain feature vector, and based on the second maximum distance, calculate the similarity between the third time-domain feature vector and the fourth time-domain feature vector.

[0105] In this embodiment, the similarity between the first time-domain feature vector and the second time-domain feature vector can be calculated using, for example, but not limited to, the following formula.

[0106] ; In the formula, This represents the relationship between the data point with index b in the first time-domain feature vector and the second time-domain feature vector. The index represents the similarity between c data points. It is a constant (with a value of 2). The similarity limit is represented by a value equal to 0.15 times the standard deviation of any time-domain voiceprint signal.

[0107] Thus, based on the aforementioned formula, the similarity between the data points corresponding to different indices in the first time-domain feature vector and the second time-domain feature vector can be calculated; of course, the calculation of the similarity between the third time-domain feature vector and the fourth time-domain feature vector is also the same, and will not be elaborated here.

[0108] After obtaining the similarity between feature vectors in different time domains, the entropy of the fuzzy measure is calculated, as shown in step S46 below.

[0109] S46. Calculate the fuzzy measure entropy of any time-domain voiceprint signal by using the similarity between the first time-domain feature vector and the second time-domain feature vector, and the similarity between the third time-domain feature vector and the fourth time-domain feature vector. After polling all time-domain voiceprint signals, obtain the multi-scale fuzzy measure entropy.

[0110] In this embodiment, the first comprehensive average similarity between the first time-domain feature vector and the second time-domain feature vector, and the second comprehensive average similarity between the third time-domain feature vector and the fourth time-domain feature vector are calculated first.

[0111] As previously explained, there is a similarity between the data points corresponding to different indices in the first time-domain feature vector and the second time-domain feature vector. Therefore, we initialize b to 1 and calculate the sum of the similarities between the data point with index b in the first time-domain feature vector and the data points with each index in the second time-domain feature vector (i.e., index c takes values ​​from 1 to U). Then, we calculate the average to obtain the first average similarity. Next, we increment b by 1 until b equals U, obtaining several first average similarities. We sum these several first average similarities to obtain the total similarity. Finally, we divide the total similarity by U to obtain the first comprehensive average similarity between the first time-domain feature vector and the second time-domain feature vector. Of course, the calculation process for the second comprehensive average similarity is the same, and will not be repeated here.

[0112] After obtaining the two comprehensive average similarities, the fuzzy measure entropy of any time-domain voiceprint signal can be calculated based on this. The calculation formula is: ln(ψ1)-ln(ψ2), where ψ1 and ψ2 represent the first comprehensive average similarity and the second comprehensive average similarity, respectively.

[0113] Thus, in the aforementioned manner, the fuzzy measure entropy of the remaining time-domain voiceprint signals can be calculated. Finally, based on this, the time-domain features of the enhanced voiceprint signal can be constructed, as shown in step S47 below.

[0114] S47. The multi-scale fuzzy measure entropy is used as the temporal feature of the enhanced voiceprint signal; in this embodiment, the temporal feature can be obtained by using the fuzzy measure entropy to form a vector.

[0115] After obtaining the time-domain features through the aforementioned steps S41 to S47, they can be combined with the frequency-domain features to form the leakage acoustic signature features, that is, the two are spliced ​​together to obtain the leakage acoustic signature features; then, the trained neural network model can be used to detect pipeline leaks, as shown in step S5 below.

[0116] S5. Input the leakage acoustic features into the pipeline leakage identification model to obtain the leakage identification result of the chemical pipeline; in this embodiment, the pipeline leakage identification model may be, but is not limited to, a trained CNN-LSTM hybrid model, that is, it is trained by taking the sample leakage features of several sample chemical pipelines as input and the leakage identification result of each sample chemical pipeline as input, and the leakage identification result is either pipeline leakage or no leakage.

[0117] Thus, after obtaining the leak identification results, the leak can be located, as shown in step S6 below.

[0118] S6. Determine the leak location of the chemical pipeline based on the optical signal. In this embodiment, as previously explained, there is a data point every 1 meter (determined by the pulse width) on the optical fiber, and each data point contains an acoustic signature signal. Therefore, when the leak identification result is a pipeline leak based on an acoustic signature signal, it indicates that there is a leak at the location corresponding to the data point. Based on this, the optical time-domain reflectometry characteristics of the distributed optical fiber can be used to locate the leak, i.e., the transmission time of the optical signal corresponding to the data point (i.e., the interval between the emission of the laser pulse and the return time of the optical signal corresponding to the data point). Then, based on the transmission time, the distance between the leak point and the laser pulse emission point is calculated, thereby obtaining the leak location.

[0119] Specifically, the formula for calculating the interval distance is: distance L = (speed of light × t) / (2 × fiber refractive index), where t represents the transmission time.

[0120] Thus, after determining the location of the leak, the leak identification results can be combined to generate pipeline leak detection results, as shown in step S7 below.

[0121] S7. Based on the leak identification result and the leak location, a pipeline leak detection result is generated. In this embodiment, the AI ​​processing unit sends the pipeline leak detection result to the intelligent inspection platform. The intelligent inspection platform will issue an alarm based on the pipeline leak detection result and display the precise location of the leak point and the acoustic spectrum map so that staff can perform pipeline maintenance in a timely manner.

[0122] Therefore, through the fiber optic sensing-based chemical pipeline leak detection method described in detail in steps S1 to S7 above, this invention collects pipeline acoustic signals using distributed fiber optic listening technology. Then, after denoising and enhancement processing, an enhanced acoustic signal is obtained. Next, feature extraction is performed to obtain leak acoustic features. Finally, based on the leak acoustic features and with the aid of artificial intelligence algorithms, leak identification is performed, and the leak location is determined using optical signals. Thus, rapid detection and precise location of chemical pipeline leaks can be achieved. Compared to traditional technologies, this invention eliminates the need for intensive manual inspections, resulting in faster response times and reduced labor costs. Furthermore, it exhibits strong resistance to environmental interference, provides full detection coverage without blind spots, and accurately pinpoints the leak location. Based on these advantages, the efficiency and accuracy of chemical pipeline leak detection are improved, thus meeting the core requirements of "real-time, comprehensive, accurate, and rapid" chemical pipeline leak detection. Therefore, this invention is highly suitable for large-scale application and promotion.

[0123] like Figure 2 As shown, the second aspect of this embodiment provides a hardware system for implementing the chemical pipeline leakage detection method based on optical fiber sensing described in the first aspect of the embodiment. The system includes: an AI processing unit and a distributed optical fiber sensing system arranged along the chemical pipeline. The AI ​​processing unit includes a signal conversion unit, a signal denoising unit, a signal enhancement unit, a feature extraction unit, a leakage detection unit, and a positioning unit.

[0124] The signal conversion unit is used to acquire the optical signal transmitted by the distributed optical fiber sensing system installed along the chemical pipeline, and to perform signal conversion processing on the optical signal to obtain the acoustic signal of the chemical pipeline.

[0125] The signal denoising unit is used to perform iterative decomposition and denoising processing on the voiceprint signal using a signal decomposition method. After each iteration of decomposition and denoising processing, the pure component signal after each decomposition is obtained, and the denoised voiceprint signal is reconstructed using each pure component signal. The input signal of the previous decomposition is subtracted from the pure component signal obtained in the previous decomposition, and the result is used as the input signal for the next decomposition.

[0126] The signal enhancement unit is used to perform global enhancement processing on the denoised audio signal to obtain an enhanced audio signal.

[0127] The feature extraction unit is used to perform feature extraction processing on the enhanced voiceprint signal to obtain leakage voiceprint features.

[0128] The leakage detection unit is used to input the leakage acoustic features into the pipeline leakage identification model to obtain the leakage identification result of the chemical pipeline.

[0129] The positioning unit is used to determine the location of the leak in the chemical pipeline based on the optical signal.

[0130] The leak detection unit is also used to generate pipeline leak detection results based on the leak identification results and the leak location.

[0131] The working process, working details and technical effects of the system provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0132] like Figure 3 As shown, the third aspect of this embodiment provides another chemical pipeline leak detection system based on fiber optic sensing. Taking the device as an electronic device as an example, it includes: a memory, a processor, and a transceiver connected in sequence. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the chemical pipeline leak detection method based on fiber optic sensing as described in the first aspect of the embodiment.

[0133] For specific examples, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0134] In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. For example, the processor may not be limited to microprocessors of the STM32F105 series, reduced instruction set computer (RISC) microprocessors, x86 architecture processors, or processors with integrated neural network processing units (NPUs). The transceiver may be, but is not limited to, a Wi-Fi transceiver, a Bluetooth transceiver, a General Packet Radio Service (GPRS) transceiver, a ZigBee (a low-power LAN protocol based on the IEEE 802.15.4 standard) transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver. Furthermore, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0135] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0136] The fourth aspect of this embodiment provides a storage medium that stores instructions containing the fiber optic sensing-based chemical pipeline leak detection method described in the first aspect of the embodiment. That is, the storage medium stores instructions that, when executed on a computer, perform the fiber optic sensing-based chemical pipeline leak detection method as described in the first aspect of the embodiment.

[0137] The storage medium refers to a carrier for storing data, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or memory sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0138] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0139] The fifth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the chemical pipeline leakage detection method based on fiber optic sensing as described in the first aspect of this embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0140] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting leaks in chemical pipelines based on fiber optic sensing, characterized in that, include: The optical signal transmitted by the distributed optical fiber sensing system installed along the chemical pipeline is acquired, and the optical signal is processed by signal conversion to obtain the acoustic signal of the chemical pipeline. The signal decomposition method is used to perform iterative decomposition and denoising on the voiceprint signal. After each decomposition and denoising process, the pure component signal is obtained. The denoised voiceprint signal is reconstructed using each pure component signal. The input signal of the previous decomposition is subtracted from the pure component signal obtained in the previous decomposition and used as the input signal for the next decomposition. The denoised audio signal is subjected to global enhancement processing to obtain an enhanced audio signal; The enhanced voiceprint signal is subjected to feature extraction processing to obtain leakage voiceprint features; The leakage acoustic signature is input into the pipeline leakage identification model to obtain the leakage identification result of the chemical pipeline; The location of the leak in the chemical pipeline was determined based on the optical signal. Based on the leak identification results and the leak location, a pipeline leak detection result is generated.

2. The method according to claim 1, characterized in that, The speaker signal is iteratively decomposed and denoised using a signal decomposition method, including: Obtain the input signal at the k-th decomposition, where the initial value of k is 1, and when k is 1, the input signal at the k-th decomposition is the voiceprint signal; Variational mode decomposition is performed on the input signal at the k-th decomposition to obtain several mode components; Calculate the signal distribution distance between each modal component and the input signal at the k-th decomposition, and select the modal component with the smallest signal distribution distance as the pure modal component, and treat the remaining modal components as noise components. The noise component is locally denoised to obtain the denoised component. The pure modal component and the denoised component are then used to form the pure component signal after the kth decomposition. Determine whether the iterative decomposition stopping condition is met. The iterative decomposition stopping condition is that the first distance is greater than the second distance, and the first distance is the smallest signal distribution distance among the signal distribution distances of each modal component at the k-th decomposition, and the second distance is the largest signal distribution distance among the signal distribution distances of each modal component at the (k-1)-th decomposition. If not, then subtract the pure component signal after the kth decomposition from the input signal at the kth decomposition to obtain the input signal at the (k+1)th decomposition. Increment k by 1 and reacquire the input signal from the kth decomposition until the iterative decomposition stopping condition is met, thus obtaining several pure component signals. These pure component signals can then be used to reconstruct the denoised signal.

3. The method according to claim 2, characterized in that, Calculate the signal distribution distance between each modal component and the input signal at the k-th decomposition, including: For any modal component, wavelet decomposition is performed on the any modal component and the input signal to obtain the first wavelet signal of the any modal component at different scales and the second wavelet signal of the input signal at different scales. Calculate the probability distribution distance between the first wavelet signal and the second wavelet signal at each scale; The distances of each probability distribution are weighted and summed to obtain the multi-scale weighted probability distribution distances. Bandpass filters of different frequency bands are constructed, and bandpass filters are used to perform bandpass filtering on any mode component and the input signal to obtain mode filter components and input filter signals of different frequency bands; The Barton coefficients between the modal filtering components and the input filtering signal for each frequency band are calculated, and the weighted sums of the Barton coefficients are obtained. Based on the weighted Bartholomew's coefficient, the energy distribution distance between any modal component and the input signal is determined; Calculate the structural distribution distance between any modal component and the input signal; Based on the multi-scale weighted probability distribution distance, the energy distribution distance, and the structural distribution distance, the signal distribution distance between any modal component and the input signal at the k-th decomposition is calculated.

4. The method according to claim 2, characterized in that, Local denoising of noise components includes: The noise components are subjected to singular value decomposition to obtain several SVD components; Calculate the first correlation coefficient between each SVD component and the noise component; Acquire calibration signal components, wherein the calibration signal components are obtained by iterative decomposition and denoising of the acoustic signature signal of a leak-free chemical pipeline; Calculate the second correlation coefficient between each SVD component and the calibration signal component; The difference between the second correlation coefficient and the first correlation coefficient of each SVD component is calculated and used as the initial singular value decomposition factor for each SVD component. From a number of initial singular value decomposition factors, the largest and smallest initial singular value decomposition factors are selected, and the singular value decomposition factor of each SVD component is calculated using the largest and smallest initial singular value decomposition factors. The SVD components are sorted in descending order of singular value decomposition factors to obtain a sorted sequence. Calculate the difference between the singular value decomposition factors of each pair of adjacent SVD components in the sorted sequence, and determine the sorting number of the SVD component corresponding to the largest difference. The signal is reconstructed using the first L SVD components in the sorted sequence to obtain the denoised component, where L is the sorting sequence number.

5. The method according to claim 1, characterized in that, The denoised audio signal is subjected to global enhancement processing to obtain an enhanced audio signal, including: Perform a short-time Fourier transform on the denoised ripple signal to obtain the time-frequency matrix of the denoised ripple signal; Based on the time-frequency matrix, the first instantaneous energy of the denoised ripple signal after short-time Fourier transform in each STFT time frame is calculated. Obtain the pipeline leakage frequency band range, and based on the time-frequency matrix, calculate the second instantaneous energy of the denoised ripple signal within the pipeline leakage frequency band range at each STFT time frame after short-time Fourier transform. Using the first instantaneous energy and the second instantaneous energy, the initial signal weight of the denoised ripple signal at each STFT time frame is calculated; The initial signal weights of each STFT time frame are upsampled to obtain the signal weights of the denoised signal at each sampling time. The denoised audio signal is weighted using the weights of each signal to obtain the enhanced audio signal.

6. The method according to claim 1, characterized in that, The enhanced voiceprint signal is subjected to feature extraction processing to obtain leakage voiceprint features, including: The enhanced voiceprint signal is subjected to frequency domain feature extraction processing to obtain frequency domain features, wherein the frequency domain features include MFCC features; Multi-scale temporal feature extraction processing is performed on the enhanced voiceprint signal to obtain the temporal features of the enhanced voiceprint signal; The leakage acoustic signature is formed by using the frequency domain features and the time domain features.

7. The method according to claim 6, characterized in that, Multi-scale temporal feature extraction processing is performed on the enhanced voiceprint signal, including: The time-domain scaling factor was determined, and based on the time-domain scaling factor and the enhanced voiceprint signal, the time-domain voiceprint signal corresponding to different time-domain scaling factors was constructed. For any time-domain acoustic signature signal, obtain the first time-domain feature dimension and the second time-domain feature dimension; Based on the first time-domain feature dimension, a first time-domain feature vector and a second time-domain feature vector corresponding to any time-domain voiceprint signal are constructed, and based on the second time-domain feature dimension, a third time-domain feature vector and a fourth time-domain feature vector corresponding to any time-domain voiceprint signal are constructed. Calculate the first maximum distance between the first time-domain feature vector and the second time-domain feature vector, and the second maximum distance between the third time-domain feature vector and the fourth time-domain feature vector; Based on the first maximum distance, the similarity between the first time-domain feature vector and the second time-domain feature vector is calculated, and based on the second maximum distance, the similarity between the third time-domain feature vector and the fourth time-domain feature vector is calculated. Using the similarity between the first time-domain feature vector and the second time-domain feature vector, and the similarity between the third time-domain feature vector and the fourth time-domain feature vector, the fuzzy measure entropy of any time-domain voiceprint signal is calculated, and after all time-domain voiceprint signals have been polled, the multi-scale fuzzy measure entropy is obtained. The multi-scale fuzzy metric entropy is used as the temporal feature of the enhanced voiceprint signal.

8. A chemical pipeline leak detection system based on fiber optic sensing, characterized in that, include: The AI ​​processing unit and the distributed optical fiber sensing system installed along the chemical pipeline, wherein the AI ​​processing unit includes a signal conversion unit, a signal denoising unit, a signal enhancement unit, a feature extraction unit, a leak detection unit and a location unit; The signal conversion unit is used to acquire the optical signal transmitted by the distributed optical fiber sensing system installed along the chemical pipeline, and to perform signal conversion processing on the optical signal to obtain the acoustic signal of the chemical pipeline. The signal denoising unit is used to perform iterative decomposition and denoising processing on the voiceprint signal using a signal decomposition method, so as to obtain the pure component signal after each decomposition after the iterative decomposition and denoising processing, and to reconstruct the denoised voiceprint signal using each pure component signal. The input signal of the previous decomposition is subtracted from the pure component signal obtained in the previous decomposition and used as the input signal for the next decomposition. A signal enhancement unit is used to perform global enhancement processing on the denoised audio signal to obtain an enhanced audio signal. The feature extraction unit is used to perform feature extraction processing on the enhanced voiceprint signal to obtain leakage voiceprint features; The leakage detection unit is used to input the leakage acoustic features into the pipeline leakage identification model to obtain the leakage identification result of the chemical pipeline; The positioning unit is used to determine the location of the leak in the chemical pipeline based on the optical signal. The leak detection unit is also used to generate pipeline leak detection results based on the leak identification results and the leak location.

9. A chemical pipeline leak detection system based on fiber optic sensing, characterized in that, include: The system comprises a memory, a processor, and a transceiver connected in sequence for communication. The memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the chemical pipeline leakage detection method based on fiber optic sensing as described in any one of claims 1 to 7.

10. A chemical pipeline leak detection system based on fiber optic sensing, characterized in that, include: A storage medium, wherein the storage medium stores instructions that, when executed on a computer, perform the chemical pipeline leakage detection method based on fiber optic sensing as described in any one of claims 1 to 7.

Images