BOTDA temperature extraction method based on kernel extreme learning machine

Abstract

The present invention discloses a temperature extraction method for a Brillouin optical time-domain analysis system (BOTDA) based on a nuclear extreme learning machine, comprising: using the Brillouin optical time-domain analysis system to collect Brillouin gain spectrum parameters of a test optical fiber; using the nuclear extreme learning machine to analyze the parameters collected by the Brillouin optical time-domain analysis system; using the obtained real matrix as training data for the nuclear extreme learning machine; training the nuclear extreme learning machine using the training data to extract accurate temperature information; and utilizing a faster processing speed to improve system performance. The present invention introduces a nuclear extreme learning machine algorithm to improve the temperature extraction accuracy and efficiency of the Brillouin optical time-domain analysis system, which is beneficial to the application of the Brillouin optical time-domain analysis system in actual detection.

Description

Technical Field

[0001] The present invention belongs to the field of optical fiber distributed sensing and machine learning, relates to a distributed optical fiber sensing system, and in particular to a Brillouin optical time domain analysis technology based on stimulated Brillouin scattering of a nuclear extreme learning machine. Background Art

[0002] In the fields of civil engineering, construction, power distribution, etc., in order to measure the temperature, stress and other parameters of facilities (such as bridges, tunnels, oil and gas pipelines, transmission cables, etc.) over long distances (i.e., to conduct safety monitoring of major facilities related to national economy and people's livelihood), test optical fibers are often deployed. By testing the temperature or force parameters of the test optical fibers, the temperature or stress parameters of the facilities over long distances can be indirectly measured.

[0003] Currently, the length of test optical fibers typically spans several kilometers, tens of kilometers, or even hundreds of kilometers. In existing technologies, a Brillouin optical time-domain analyzer (BOTDA) is often used to collect the operating temperature of the deployed test optical fibers over long distances. BOTDA can collect information parameters of the test optical fibers, such as the Brillouin gain spectrum information of the test optical fibers at equal intervals over their length. However, the information parameters collected by BOTDA are all reflected in the form of electrical signals and cannot be directly read.

[0004] Therefore, existing technologies often use Lorentz curve fitting algorithms to analyze the information parameters collected by BOTDA. That is, using the Lorentz curve to analyze and convert the electrical signal parameters collected by BOTDA into readable digital information parameters. However, the Lorentz curve fitting algorithm has a slow analysis speed, low response speed, and low analysis efficiency.

[0005] Brillouin optical time-domain analysis (BOTDA), as an important DFS technology, has broad application prospects in the safety monitoring of major facilities related to national economy and people's livelihood, such as power transmission cables, oil and gas pipelines, and civil structures. Brillouin optical time-domain analysis technology is a mainstream technology in distributed fiber optic sensing based on Brillouin scattering, and has the characteristics of long sensing distance and relatively simple structure.

[0006] Distributed fiber optic sensing technology is developing rapidly and is now widely used in safety monitoring for large-scale facilities. It enables real-time health status monitoring and allows for the rapid and precise location of potential hazards to ensure safe operation. Distributed fiber optic sensing systems based on Brillouin scattering can measure temperature and strain. These systems, with their long sensing distances and high accuracy, are being applied in various fields and are a research hotspot in the field of fiber optic sensing. However, improving their performance in practical applications is crucial.

[0007] In the past, the main research was on extracting the temperature of the Brillouin optical time-domain analysis system based on traditional fitting methods, such as using the Lorentz curve fitting algorithm to extract the temperature of the Brillouin optical time-domain analysis system. However, this data processing method has a slow data processing speed.

[0008] Currently, machine learning methods have certain advantages in extracting temperature information from Brillouin optical time-domain analysis systems. Among them, methods based on kernel extreme learning machines significantly speed up data processing at a slight loss of accuracy.

[0009] Therefore, when extracting the temperature of the test fiber, it is necessary to try algorithms other than Lorentz curve fitting. In other words, studying the temperature extraction method of the Brillouin optical time-domain analysis system using a nuclear extreme learning machine is of great significance in practical applications. Summary of the Invention

[0010] Purpose of the Invention: In view of the above-mentioned deficiencies in existing data processing technologies, the present invention proposes a temperature extraction method for a Brillouin optical time-domain analysis system (BOTDA) based on a nuclear extreme learning machine. The Brillouin optical time-domain analysis system is used to collect temperature parameters of a test optical fiber, and a nuclear extreme learning machine is used to analyze the parameters collected by the Brillouin optical time-domain analysis system. More accurate temperature information is extracted from the collected Brillouin gain spectrum, and faster processing speed is utilized to improve system performance.

[0011] Technical solution: The purpose of the present invention is achieved through the following means.

[0012] A temperature extraction method based on nuclear extreme learning machine (BOTDA) is proposed. The Brillouin optical time domain analysis system is used to collect the temperature parameters of the test optical fiber, and the nuclear extreme learning machine is used to analyze the parameters collected by the Brillouin optical time domain analysis system. The Brillouin optical time domain analysis system includes a distributed feedback laser, a coupler, an acousto-optic modulator, an electro-optic modulator, a first erbium-doped fiber amplifier, an isolator, a polarization scrambler, a circulator, a photodetector, a piezoelectric oscillator and a mixer. The laser light emitted by the distributed feedback laser is divided into two paths by the coupler, namely a pump pulse light and a continuous probe light. For photometry, the pump pulse light is modulated by an acousto-optic modulator, amplified by the first erbium-doped fiber amplifier, and enters the circulator after the polarization is eliminated by the polarization scrambler; the continuous detection light is modulated by the microwave signal through the electro-optic modulator, and the output end of the electro-optic modulator is connected to the isolator. The pump pulse light and the continuous detection light undergo stimulated Brillouin scattering amplification effect in the test fiber. In order to achieve the purpose of frequency reduction, the local oscillator light generated by the piezoelectric oscillator and the signal detected by the photodetector are mixed in the mixer to generate an intermediate frequency signal. The intermediate frequency signal is collected and the temperature information is extracted using a nuclear extreme learning machine.

[0013] Furthermore, in the temperature extraction method based on the nuclear extreme learning machine (BOTDA), the Brillouin optical time domain analysis system further includes a polarization controller, a second erbium-doped fiber amplifier, a logarithmic detector, and a data acquisition card.

[0014] In the temperature extraction method based on the nuclear extreme learning machine (BOTDA), the output end of the distributed feedback laser is connected to the input end of the coupler, the output end of the coupler is respectively connected to the input ends of the electro-optic modulator and the acousto-optic modulator, a polarization controller is connected between the coupler and the electro-optic modulator, the output end of the acousto-optic modulator is connected to the input end of the first erbium-doped fiber amplifier, the output end of the first erbium-doped fiber amplifier is connected to the input end of the polarization scrambler, and the output end of the polarization scrambler is connected to one input end of the circulator; the output end of the electro-optic modulator is connected to the input end of the isolator, the output end of the isolator is connected to the other input end of the circulator through a test optical fiber, the output end of the circulator is connected to the input end of the photodetector through the second erbium-doped fiber amplifier, the output end of the photodetector is connected to the input end of the mixer, the output end of the mixer is connected to the input end of the logarithmic detector, the piezoelectric oscillator is also connected to the mixer, and the output end of the logarithmic detector is connected to a data acquisition card.

[0015] Furthermore, in the temperature extraction method of BOTDA based on nuclear extreme learning machine, the data acquisition card delivers the collected data to the computer, and the temperature parameters of the collected test optical fiber are extracted in the computer by the nuclear extreme learning machine algorithm. The training data is trained by the nuclear extreme learning machine. The training process is as follows: the training data is input into the nuclear extreme learning machine. The training data includes the Brillouin gain spectrum matrix [x i1 , x i2 ,...,x in ] T ∈R n And the label vector corresponding to the temperature information [c i1 , c i2 ,...,c im ] T ∈R m , where n represents the input feature dimension and m represents the output feature dimension; the gain spectrum data in the training data is mapped by the kernel function to generate a high-dimensional feature kernel matrix Ω ELM , the output weight matrix B is solved together with the one-hot label matrix C; the temperature extraction method prediction process of BOTDA based on the nuclear extreme learning machine is as follows: the Brillouin gain spectrum data to be predicted is modulo-normalized and pre-processed, and then input into the nuclear extreme learning machine; the high-dimensional features obtained by the kernel function operation are multiplied by the output weight matrix to obtain the classification evaluation value Y. For the nuclear extreme learning machine, the temperature classification information is mapped to a 1*K vector y t In the end, y tThe class with the highest score in the SoftMax distribution is taken as the predicted class.

[0016] After adopting the above technical solution, the beneficial effects of the present invention are as follows:

[0017] This paper proposes a temperature extraction method for a Brillouin optical time-domain analysis system based on a nuclear extreme learning machine. Compared to traditional Lorentz curve fitting methods for extracting temperature information, this method, based on a nuclear extreme learning machine, can extract more accurate temperature information from the collected Brillouin gain spectrum. Its processing speed is also significantly superior to Lorentz fitting, thereby improving system performance. Furthermore, this method eliminates the need for cumbersome preprocessing, significantly reducing computation time and hardware requirements, improving real-time performance and feasibility, and facilitating the widespread application of Brillouin optical time-domain analysis systems in practical monitoring applications.

[0018] The accompanying drawings are as follows:

[0019] Figure 1 This is the system block diagram of the Brillouin optical time-domain analysis system (BOTDA for short).

[0020] Figure 2 Extracting temperature diagram for kernel extreme learning machine

[0021] Figure 3 Schematic diagram of the network structure of the kernel extreme learning machine in the method of the present invention.

[0022] Figure 4 A comparison chart of BOTDA temperature extraction errors using kernel extreme learning machine at different signal-to-noise ratios.

[0023] Figure 5 Comparison of temperature extraction errors of BOTDA using kernel extreme learning machine at different sweep frequency step sizes

[0024] In the figure: 1-distributed feedback laser, 2-coupler, 3-polarization controller, 4-acousto-optic modulator, 5-electro-optic modulator, 6-first erbium-doped fiber amplifier, 7-second erbium-doped fiber amplifier, 8-isolator, 9-polarization scrambler, 10-circulator, 11-photodetector, 12-piezoelectric oscillator, 13-mixer, 14-logarithmic detector, 15-data acquisition card DETAILED DESCRIPTION

[0025] The implementation of the present invention will be further described below with reference to the accompanying drawings.

[0026] like Figure 1As shown, a Brillouin optical time domain analysis system (BOTDA for short) includes: a distributed feedback laser 1, a coupler 2, a polarization controller 3, an acousto-optic modulator 4, an electro-optic modulator 5, a first erbium-doped fiber amplifier 6, a second erbium-doped fiber amplifier 7, an isolator 8, a polarization scrambler 9, a circulator 10, a photodetector 11, a piezoelectric oscillator 12, a mixer 13, a logarithmic detector 14 and a data acquisition card 15.

[0027] The connection relationship of the various hardware components constituting the above-mentioned Brillouin optical time-domain analysis system is as follows: the output end of the distributed feedback laser 1 is connected to the input end of the coupler 2, the output end of the coupler 2 is respectively connected to the input ends of the electro-optic modulator 5 and the acousto-optic modulator 4. A polarization controller 3 is also connected between the coupler 2 and the electro-optic modulator 5. The output end of the acousto-optic modulator 4 is connected to the input end of the first erbium-doped fiber amplifier 6, the output end of the first erbium-doped fiber amplifier 6 is connected to the input end of the polarization scrambler 9, and the output end of the polarization scrambler 9 is connected to one input end of the circulator 10; the output end of the electro-optic modulator 5 is connected to the input end of the isolator 8. The output end of the isolator 8 is connected to the other input end of the circulator 10 through the test optical fiber. The output end of the circulator 10 is connected to the input end of the photodetector 11 through the second erbium-doped fiber amplifier 7. The output end of the photodetector 11 is connected to the input end of the mixer 13. The output end of the mixer 13 is connected to the input end of the logarithmic detector 14. The piezoelectric oscillator 12 is also connected to the mixer 13. The output end of the logarithmic detector 14 is connected to the data acquisition card 15. The data acquisition card 15 delivers the collected data to the computer, and the temperature parameters of the collected test optical fiber are extracted in the computer through the kernel extreme learning machine algorithm.

[0028] In the Brillouin optical time domain analysis system, the two ends of the test optical fiber are connected to the output end of the isolator and one input end of the circulator respectively.

[0029] exist Figure 1In the embodiment, the laser light emitted by the distributed feedback laser 1 is split into two paths by the coupler 2. In this embodiment, the coupler 2 is a 50:50 optical coupler. The laser light split into two paths by the coupler 2 is a pump pulse light and a continuous probe light. The pump pulse light is modulated by an acousto-optic modulator 4 with an extinction ratio of 50dB, amplified by a first erbium-doped fiber amplifier 6, and then enters a circulator 10 after passing through a polarization scrambler 9 to eliminate polarization. The continuous probe light is modulated by a microwave signal via an electro-optic modulator 5. The microwave signal frequency ranges from 10.560 GHz to 10.760 GHz, ensuring that the probe light scans within the Brillouin gain range. The output of the electro-optic modulator 5 is connected to an isolator 8 to process the reverse propagating light beam. Finally, the modulated signal undergoes sideband filtering before entering the test fiber. The pump pulse light and continuous probe light undergo stimulated Brillouin scattering amplification in the test fiber. The scattered light is converted into an electrical signal after passing through a photodetector. The local oscillator light generated by the piezoelectric oscillator 12 and the signal detected by the photodetector 11 are then mixed in a mixer 13 to generate an intermediate frequency signal, achieving frequency reduction and facilitating signal acquisition. The signal passes through a logarithmic detector 14, and data is collected using a data acquisition card 15. Finally, temperature information is extracted from the collected Brillouin gain spectrum.

[0030] A kernel extreme learning machine (KELM) method was used to extract temperature from a Brillouin optical time-domain analysis system. Compared to the Lorentz curve fitting method, the KELM method improved the accuracy of temperature extraction at different signal-to-noise ratios (at different locations) along the optical fiber. The KELM performance degraded more slowly when the frequency sweep step size was increased.

[0031] The principle of kernel extreme learning machine to extract temperature is as follows Figure 2 As shown, temperature information is extracted during the training and testing phases of the KELM algorithm. Based on theoretical analysis and properties of the actual Brillouin gain spectrum, which exhibits a linear shape between the Lorentzian and Gaussian curves, it can be fitted using a pseudo-Voigt curve. Therefore, the pseudo-Voigt curve serves as a training example. After processing the KELM algorithm, the input curve is converted to different temperature scales during the training phase.

[0032] In order to verify the robustness of the kernel extreme learning machine algorithm, the parameters of the fiber signal-to-noise ratio and the scanning frequency step size were changed in the experiment, such as Figure 2 In the experiment, the measured Brillouin gain spectrum along the tested fiber is used as input to the algorithm to extract temperature information.

[0033] Figure 4 and Figure 5 This is a comparison chart of the experimental results of the kernel extreme learning machine algorithm. Among them: Figure 4Figure 2 compares the BOTDA temperature extraction error using a kernel extreme learning machine at different signal-to-noise ratios. As distance increases, the signal-to-noise ratio decreases along with the fiber length. At various signal-to-noise ratios, the kernel extreme learning machine achieves a smaller temperature extraction error than the original extreme learning machine, significantly outperforming the Lorentz fitting method while maintaining comparable accuracy. Figure 5 Figure 1 compares the temperature extraction error of BOTDA using a kernel extreme learning machine at different frequency sweep step sizes. As the frequency sweep step size increases, the temperature extraction error gradually increases. Compared to the original extreme learning machine algorithm, the kernel extreme learning machine has a smaller increase in error and better robustness.

[0034] In summary, the temperature extraction method of the Brillouin optical time-domain analysis system based on the kernel extreme learning machine proposed in the present invention has the following characteristics:

[0035] 1) The kernel extreme learning machine has better accuracy in temperature extraction of the Brillouin optical time-domain analysis system at different distances along the light line (under different signal-to-noise ratios);

[0036] 2) The KELM model's performance in temperature extraction for Brillouin optical time-domain analysis systems degrades more slowly when the frequency sweep step size is increased. This improves the accuracy and efficiency of temperature extraction for Brillouin optical time-domain analysis systems in practical applications, providing a new solution for temperature extraction in Brillouin optical time-domain analysis systems.

[0037] What has been stated above is merely a preferred embodiment of the method of the present invention. It should be noted that, without departing from the essence of the scheme and apparatus of the present invention, certain changes may be made in actual implementation (such as the instrument parameters of the Brillouin optical time-domain analysis system) and should also be included in the scope of protection of the present invention.

Claims

1. A temperature extraction method based on a nuclear extreme learning machine (BOTDA), wherein a Brillouin optical time domain analysis system is used to collect Brillouin gain spectrum parameters of a test optical fiber, and a nuclear extreme learning machine is used to analyze the parameters collected by the Brillouin optical time domain analysis system. The Brillouin optical time domain analysis system comprises a distributed feedback laser (1), a coupler (2), an acousto-optic modulator (4), an electro-optic modulator (5), a first erbium-doped fiber amplifier (6), an isolator (8), a polarization scrambler (9), a circulator (10), a photodetector (11), a piezoelectric oscillator (12) and a mixer (13), and is characterized in that: The laser light emitted by the distributed feedback laser (1) is divided into two paths by the coupler (2), namely, a pump pulse light and a continuous detection light. The pump pulse light is modulated by the acousto-optic modulator (4), amplified by the first erbium-doped fiber amplifier (6), and enters the circulator (10) after the polarization is eliminated by the polarization scrambler (9); the continuous detection light is modulated by the microwave signal by the electro-optic modulator (5), and the output end of the electro-optic modulator (5) is connected to the isolator (8). The pump pulse light and the continuous detection light undergo a stimulated Brillouin scattering amplification effect in the test optical fiber. The local oscillator light generated by the piezoelectric oscillator (12) and the signal detected by the photodetector (11) are mixed in the mixer (13) to generate an intermediate frequency signal. The intermediate frequency signal is collected and the temperature information is extracted using a nuclear extreme learning machine to obtain the temperature information. The pseudo-Voigt curves serve as training samples, and after being processed by the kernel extreme learning machine algorithm, the input curves are converted into different temperature scales during the training phase; The training process is as follows: input the training data into the kernel extreme learning machine, the training data includes the Brillouin gain spectrum matrix [x i1 , x i2 ,...,x in ] T ∈R n And the label vector corresponding to the temperature information [c i1 , c i2 ,...,c im ] T ∈R m , where n represents the input feature dimension and m represents the output feature dimension; the gain spectrum data in the training data is mapped by the kernel function to generate a high-dimensional feature kernel matrix Ω ELM , use this feature matrix and the one-hot label matrix C to solve the output weight matrix B; The prediction process is as follows: the Brillouin gain spectrum data to be predicted is modulated and normalized, and then input into the kernel extreme learning machine; the high-dimensional features obtained by the kernel function operation are multiplied by the output weight matrix to obtain the classification evaluation value Y. For the kernel extreme learning machine, the temperature classification information is mapped to a 1*K vector y t In the end, y t The class with the highest score in the SoftMax distribution is taken as the predicted class.

2. The temperature extraction method based on BOTDA of kernel extreme learning machine according to claim 1, characterized in that: The output end of the distributed feedback laser (1) is connected to the input end of the coupler (2), the output end of the coupler (2) is connected to the input ends of the electro-optic modulator (5) and the acousto-optic modulator (4), respectively, a polarization controller (3) is connected between the coupler (2) and the electro-optic modulator (5), the output end of the acousto-optic modulator (4) is connected to the input end of the first erbium-doped fiber amplifier (6), the output end of the first erbium-doped fiber amplifier (6) is connected to the input end of the polarization scrambler (9), and the output end of the polarization scrambler (9) is connected to one input end of the circulator (10); the electro-optic modulator (5) The output end of the isolator (8) is connected to the input end of the isolator (8), the output end of the isolator (8) is connected to the other input end of the circulator (10) through a test optical fiber, the output end of the circulator (10) is connected to the input end of the photodetector (11) through a second erbium-doped optical fiber amplifier (7), the output end of the photodetector (11) is connected to the input end of the mixer (13), the output end of the mixer (13) is connected to the input end of the logarithmic detector (14), the piezoelectric oscillator (12) is also connected to the mixer (13), and the output end of the logarithmic detector (14) is connected to a data acquisition card (15).

3. The temperature extraction method based on BOTDA of kernel extreme learning machine according to claim 2, characterized in that: The data acquisition card (15) delivers the collected data to a computer, and the temperature parameters of the collected test optical fiber are extracted in the computer through a kernel extreme learning machine algorithm.

Images