Brillouin scattering double-peak frequency shift extraction method and device for training ANN model based on simulation data

Through the artificial neural network model trained based on simulation data, the problem of frequency shift extraction difficulties caused by Brillouin gain spectrum deformation is solved, and the rapid and accurate extraction of bimodal frequency shift is achieved, which improves the accuracy of Brillouin sensing technology and Real-time.

CN118965965APending Publication Date: 2024-11-15ZHEJIANG UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410964350.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

In temperature strain monitoring scenarios, the deformation of the Brillouin gain spectrum makes it difficult for the Lorentz fitting method to accurately extract the bimodal frequency shift, affecting the measurement accuracy and real-time performance.

Method used

Using an artificial neural network model based on simulation data training, the input layer, hidden layer and output layer are constructed, the model is trained using the simulation data set to generate a bimodal Brillouin gain spectrum, and the model effect is evaluated through the root mean square error and standard deviation. , and is ultimately used for frequency shift extraction in actual scenarios.

Benefits of technology

It realizes rapid and accurate extraction of Brillouin frequency shifts in bimodal cases, improves the accuracy and real-time performance of sensing, and overcomes the problems of long iteration processing time and insufficient accuracy in the existing methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118965965A_ABST
    Figure CN118965965A_ABST
Patent Text Reader

Abstract

The invention discloses a Brillouin scattering double-peak frequency shift extraction method and device for training an ANN model based on simulation data, and the method comprises the following steps: changing Brillouin gain spectrum parameters within a set parameter range according to a set step length, simulating to obtain a large number of double-peak Brillouin gain spectrums under different parameters in a set parameter range as a simulation data set; constructing an artificial neural network model which comprises an input layer, a hidden layer and an output layer; training an artificial neural network model by using the simulation data set, evaluating the training effect of the model by calculating performance indexes, and performing double-peak frequency shift extraction by using the trained artificial neural network model; and verifying the effect of the trained artificial neural network model by using the bimodal data of the actual pipeline settlement strain, and then applying the verified artificial neural network model to an actual scene. According to the method, the Brillouin frequency shift under the double-peak condition can be rapidly and accurately extracted, so that the sensing accuracy and real-time performance are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of Brillouin frequency shift extraction, and in particular relates to a method and device for extracting double-peak frequency shift of Brillouin scattering based on training an ANN model with simulation data. Background Art

[0002] Brillouin Scattering Distributed Optical Fiber Sensor (B-DOFS) is widely used in health monitoring and disaster warning of pipelines, cables, bridges, tunnels and other buildings due to its ability to sense temperature and strain, distributed, long sensing distance and high accuracy. B-DOFS measures the temperature and strain along the optical fiber by detecting the Brillouin Frequency Shift (BFS) of the Brillouin Gain Spectrum (BGS). The extraction of Brillouin Frequency Shift is an important step in temperature and strain detection.

[0003] The commonly used Brillouin frequency shift extraction method is the Lorentz fitting method, and the peak value of the fitting result is the Brillouin frequency shift value. This method requires the Brillouin gain spectrum to conform to the Lorentz curve. The Chinese patent application with publication number CN113447071A provides a method for extracting optical fiber Brillouin frequency shift based on an artificial neural network, using the actual single-peak Brillouin gain spectrum after noise reduction for training to extract the single-peak BFS. The Chinese patent application with publication number CN116678439A discloses a method for extracting single-peak and double-peak frequency shift information from the Brillouin scattering spectrum, using Lorentz fitting to directly fit the double peak to extract the double-peak frequency shift. However, in the temperature and strain monitoring scenario, when the temperature or strain distribution within the spatial resolution range of the system is uneven, the measured Brillouin gain spectrum will be deformed on the basis of the Lorentz curve. The Brillouin gain spectrum that conforms to the Lorentz curve presents a single peak, while the deformed Brillouin gain spectrum will have two or even more peaks, and the shapes of the peaks such as height, width and symmetry are different, which affects the frequency shift value obtained by the Lorentz fitting method, and further affects the measurement of temperature and strain.

[0004] Some methods are used to fit the double-peak Brillouin gain spectrum under the influence of inhomogeneous temperature strain, such as splitting the double peak into two single peaks and fitting them separately (see reference 1: L. Zhao, Z. Xu, and Y. Li, An accurate and rapid method for extracting parameters from multi-peak Brillouin scattering spectrum. Sens. Actuators, A. 232, 276-284 (2015)), and restoring the distorted double-peak Brillouin gain spectrum by zero-filling short-time Fourier transform (see reference 2: Y. Yu et al., Double peak-induced distance error in short-time-Fou rier-transform-Brillouin optical time domain reflectometers event detection and the recovery method, Appl. Opt. 54, E196 (2015)). However, these methods require iteration and take too long to process, making it difficult to extract the Brillouin frequency shift in the double-peak gain spectrum in real time. Therefore, in the field of Brillouin sensing, especially in the field of distributed optical fiber sensing technology, it is necessary to provide a method for quickly and accurately extracting the Brillouin frequency shift of a double-peak gain spectrum. Summary of the invention

[0005] In view of the above, the purpose of the present invention is to provide a method and device for extracting the double-peak frequency shift of Brillouin scattering based on an ANN model trained with simulation data, so as to realize the rapid and accurate extraction of the Brillouin frequency shift in the double-peak case, thereby improving the accuracy and real-time performance of sensing, and helping to expand the application scope of Brillouin sensing technology.

[0006] In order to achieve the above-mentioned invention object, the technical solution provided by the present invention is as follows:

[0007] In a first aspect, an embodiment of the present invention provides a method for extracting double-peak frequency shift of Brillouin scattering based on training ANN algorithm with simulation data, comprising the following steps:

[0008] The Brillouin gain spectrum parameters are changed according to a set step size within a set parameter range, and a large number of double-peak Brillouin gain spectra under different parameters within the set parameter range are obtained by simulation as a simulation data set;

[0009] Constructing an artificial neural network model, which includes an input layer for receiving a simulation data set, a hidden layer for extracting data features, and an output layer for converting the data features into a bimodal frequency shift;

[0010] The artificial neural network model is trained using a simulation data set, the training effect of the model is evaluated by calculating performance indicators, and the trained artificial neural network model is used to extract bimodal frequency shifts.

[0011] Preferably, the simulated bimodal Brillouin gain spectrum g(v) is generated according to the following formula:

[0012]

[0013] Among them, k is the ratio of the two peak amplitudes, g B1 and g B2 are the Brillouin gain coefficients of the left and right peaks, v is the sweep frequency, and v B1 and v B2 are the Brillouin frequency shifts of the left and right peaks, Δv B1 and Δv B2 are the half-peak widths of the two Brillouin gain spectra corresponding to the left peak and the right peak respectively.

[0014] Preferably, different parameters of the bimodal Brillouin gain spectrum are set during simulation including:

[0015] Set different Brillouin frequency shifts: The Brillouin frequency shift range of the left peak in the double peak is calculated by the base frequency and step size, and the Brillouin frequency shift range of the right peak in the double peak is calculated by the Brillouin frequency shift difference between the right peak and the left peak. The base frequency is set according to the type of optical fiber, and the step size is set according to the accuracy of the simulation data.

[0016] Set different half-peak widths: The half-peak widths of the double peaks are calculated based on the half-peak width range and step size;

[0017] Set different signal-to-noise ratios: The signal-to-noise ratios of the two peaks are calculated based on the signal-to-noise ratio range and step size;

[0018] Different ratios of the double peak amplitudes are set: the ratio of the double peak amplitudes is k=k2 / (k1+k2), wherein k1 and k2 are the peak amplitudes of the left peak and the right peak in the double peaks, respectively.

[0019] Preferably, the setting values ​​of each parameter include:

[0020] The Brillouin frequency shift is set as follows: the Brillouin frequency shift range of the left peak is the base frequency + (60MHz to 190MHz), and the Brillouin frequency shift difference range between the right peak and the left peak is 0 to 98MHz;

[0021] The half-peak width is set as follows: the half-peak width range is 30MHz to 64MHz;

[0022] The signal-to-noise ratio is set as follows: The signal-to-noise ratio range is 10 to 25 dB;

[0023] The ratio of the two peak amplitudes is set as follows: k<0.5 simulates the case where the left peak amplitude is larger, k=0.5 simulates the case where the two peak amplitudes are equal, and k>0.5 simulates the case where the right peak amplitude is larger.

[0024] Preferably, the method further comprises:

[0025] The effect of the trained artificial neural network model is verified using the bimodal data of actual pipeline settlement strain, and the verified artificial neural network model is used for bimodal gain spectrum frequency shift extraction and strain detection in actual settlement monitoring scenarios.

[0026] Preferably, when the artificial neural network model is trained using a simulation data set, the performance indicators used include:

[0027] The root mean square error RMSE is calculated as follows:

[0028]

[0029] Among them, y i and are the predicted value and actual value of the i-th Brillouin double-peak frequency shift among the N extracted;

[0030] The standard deviation SD is calculated as follows:

[0031]

[0032] Among them, y i is the extracted i-th Brillouin doublet frequency shift, is the average of all extracted Brillouin doublet frequency shifts.

[0033] In a second aspect, an embodiment of the present invention provides a Brillouin scattering double-peak frequency shift extraction device based on a simulation data training ANN model, comprising: a simulation data set construction module, an artificial neural network model construction module and a simulation data model training module;

[0034] The simulation data set construction module is used to change the Brillouin gain spectrum parameters according to the set step size within the set parameter range, and simulate a large number of bimodal Brillouin gain spectra under different parameters within the set parameter range as the simulation data set;

[0035] The artificial neural network model construction module is used to construct an artificial neural network model, which includes an input layer for receiving a simulation data set, a hidden layer for extracting data features, and an output layer for converting the data features into a bimodal frequency shift;

[0036] The simulation data model training module is used to train the artificial neural network model using the simulation data set, evaluate the training effect of the model by calculating the performance index, and perform bimodal frequency shift extraction using the trained artificial neural network model.

[0037] Preferably, the device further comprises a model verification and application module;

[0038] The model verification and application module is used to verify the effect of the trained artificial neural network model using the bimodal data collected through the simulated strain experiment and the bimodal data of the actual pipeline settlement strain, and to use the verified artificial neural network model for bimodal gain spectrum frequency shift extraction and strain detection in actual settlement monitoring scenarios.

[0039] In the third aspect, an embodiment of the present invention provides a Brillouin scattering double-peak frequency shift extraction device based on a simulated data trained ANN model, comprising a memory and one or more processors, wherein the memory is used to store a computer program, and the processor is used to implement the above-mentioned Brillouin scattering double-peak frequency shift extraction method based on a simulated data trained ANN model when executing the computer program.

[0040] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a computer, the above-mentioned Brillouin scattering double-peak frequency shift extraction method based on training an ANN model with simulation data is implemented.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] (1) The present invention uses a simulation method to obtain a large amount of bimodal gain spectrum data with multiple parameters for artificial neural network model training, which can overcome the problem of small training set due to the difficulty in collecting data for bimodal situations. After training, the model effect is verified by testing the simulated data and the actual collected data respectively, so that the trained and verified artificial neural network model can accurately extract the Brillouin frequency shift of the bimodal Brillouin gain spectrum caused by the uneven distribution of temperature strain, and has higher accuracy than the existing Lorentz fitting method.

[0043] (2) The artificial neural network model trained by the present invention through simulation data can realize rapid extraction of Brillouin frequency shift in the case of double peaks, has good real-time performance, and helps to expand the application scope of Brillouin sensing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0045] Figure 1 It is a flow chart of a method for extracting double-peak frequency shift of Brillouin scattering based on training ANN model with simulation data provided by an embodiment of the present invention;

[0046] Figure 2 Schematic diagrams of various simulated double-peak Brillouin gain spectra provided by an embodiment of the present invention, wherein FIG. (a) illustrates different double-peak amplitude ratios, FIG. (b) illustrates different half-peak widths, FIG. (c) illustrates different Brillouin frequency shifts and different Brillouin frequency shift differences, and FIG. (d) illustrates different signal-to-noise ratios;

[0047] Figure 3 It is a schematic diagram of a process of training an ANN model using a simulation data set provided by an embodiment of the present invention;

[0048] Figure 4 It is a schematic diagram of the process of using the ANN model trained by simulation for actual settlement monitoring provided by an embodiment of the present invention;

[0049] Figure 5 It is a structural schematic diagram of a Brillouin scattering double-peak frequency shift extraction device based on a simulation data training ANN model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] To make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation methods described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0051] The inventive concept of the present invention is: in view of the problem of insufficient efficiency and accuracy of the Brillouin frequency shift extraction method in the double-peak gain spectrum in the prior art, the embodiment of the present invention provides a Brillouin scattering double-peak frequency shift extraction method and device based on the training of the ANN model with simulation data, and obtains a large amount of double-peak gain spectrum data with multiple parameters by adopting the simulation method, and after training, the simulation data and the actual collected data are respectively tested on the artificial neural network (ANN) model to verify the effect of the ANN model. Compared with the existing methods such as the improved Lorentz fitting, split double-peak extraction, etc., each extraction takes a long time and the accuracy cannot be guaranteed, the present invention can achieve faster and more accurate double-peak frequency shift extraction.

[0052] Figure 1 FIG. 1 is a flow chart of a method for extracting double-peak frequency shift of Brillouin scattering based on training ANN model with simulation data provided by an embodiment of the present invention. Figure 1 As shown, the embodiment provides a method for extracting double-peak frequency shift of Brillouin scattering based on training ANN model with simulation data, comprising the following steps:

[0053] S1, changing the Brillouin gain spectrum parameters according to the set step size within the set parameter range, and simulating a large number of double-peak Brillouin gain spectra under different parameters within the set parameter range as a simulation data set.

[0054] Specifically, according to the Lorentz function, two single-peak Brillouin gain spectra are obtained through simulation, and a double-peak Brillouin gain spectrum is obtained after adding them. The simulated double-peak Brillouin gain spectrum g(v) is generated according to the following formula:

[0055]

[0056] Among them, k is the ratio of the two peak amplitudes, g B1 and g B2 are the Brillouin gain coefficients of the left peak and the right peak respectively. In the embodiment, g B1 and g B2 In the simulation, they are all set to 1 to simulate the normalized Brillouin gain spectrum. v is the sweep frequency, and v B1 and v B2 are the Brillouin frequency shifts of the left and right peaks, Δv B1 and Δv B2 are the half-peak widths of the two Brillouin gain spectra corresponding to the left peak and the right peak respectively.

[0057] During the simulation, different parameters of the double-peak Brillouin gain spectrum are set, including Brillouin frequency shift, half-peak width, signal-to-noise ratio, and ratio of double-peak amplitude, and the actual value of the Brillouin double-peak frequency shift difference is used as the training label. Figure 2Shown are schematic diagrams of various simulated double-peak Brillouin gain spectra.

[0058] Wherein, different Brillouin frequency shifts are set: the Brillouin frequency shift range of the left peak in the double peak is calculated by the base frequency (base) and the step size, the Brillouin frequency shift difference range between the right peak and the left peak can be set, and the Brillouin frequency shift range of the right peak in the double peak is calculated by the Brillouin frequency shift difference between the right peak and the left peak. Wherein, the base frequency is set according to the type of optical fiber, and the influence of the type of optical fiber on the model extraction effect is avoided by introducing the base value. The step size of the Brillouin frequency shift value is set according to the accuracy of the simulation data. In the embodiment, the Brillouin frequency shift is set as: the Brillouin frequency shift range of the left peak = base frequency + (60MHz~190MHz), the step size is any value greater than 0, the step size in the embodiment = 1MHz, and the Brillouin frequency shift difference range between the right peak and the left peak = 0~98MHz. The step size of the Brillouin frequency shift value is any value greater than 0. In the embodiment, the step size = 1MHz can improve the accuracy of the simulation data and the accuracy of the model output. The left peak Brillouin frequency shift parameter and the Brillouin frequency shift difference parameter can simulate different Brillouin frequency shifts, enhance the representativeness of the training data, and improve the robustness of the training model.

[0059] Among them, different half-peak widths are set: the half-peak widths of the double peaks are calculated according to the half-peak width range and step size. In the embodiment, the half-peak width is set to: half-width range = 30MHz ~ 64MHz, the step size is any value greater than 0, and the step size in the embodiment is 2MHz. The setting of the half-peak width can simulate different half-peak widths, enhance the representativeness of the training data, and improve the robustness of the training model.

[0060] Among them, different signal-to-noise ratios are set: the signal-to-noise ratios of the two peaks are calculated according to the signal-to-noise ratio range and step size. In the embodiment, the signal-to-noise ratio is set to: signal-to-noise ratio (snr) = 10-25dB, the step size is any value greater than 0, and the step size in the embodiment is 5dB. By simulating the Brillouin gain spectrum under different noises, the anti-noise ability of the model training is improved, and the influence of noise on the model output results is reduced.

[0061] Among them, different ratios of the double peak amplitudes are set: the ratio of the double peak amplitudes is k=k2 / (k1+k2), where k1 and k2 are the peak amplitudes of the left peak and the right peak in the double peak, respectively. The ratio of the double peak amplitudes is set to: k<0.5 simulates the case where the left peak amplitude is larger, k=0.5 simulates the case where the double peak amplitudes are equal, and k>0.5 simulates the case where the right peak amplitude is larger. In the embodiment, the setting range of the ratio of the double peak amplitudes is k is 0.1~0.9, the step size is 0.1, k=0.1~0.4 simulates the case where the left peak amplitude is larger, k=0.5 simulates the case where the double peak amplitudes are equal, and k=0.6~0.9 simulates the case where the right peak amplitude is larger. The range of k values ​​includes various situations of the relative heights of the double peaks, which enhances the representativeness of the data and improves the accuracy of the model results.

[0062] S2, constructing an artificial neural network model, which includes an input layer for receiving a simulation data set, a hidden layer for extracting data features, and an output layer for converting the data features into a bimodal frequency shift.

[0063] Specifically, the artificial neural network model includes an input layer, two hidden layers and an output layer. The activation function in the hidden layer selects the ReLu activation function, and the learning rate is set to gradually decrease with the training rounds.

[0064] S3, using the simulation data set to train the artificial neural network model, evaluating the training effect of the model by calculating the performance index, and using the trained artificial neural network model to extract the bimodal frequency shift.

[0065] Specifically, Figure 3 The process of training the ANN model using a simulation data set is shown. First, the one-dimensional simulated Brillouin gain spectrum arranged in time obtained by simulation in the simulation data set is input, the model weights are determined through training, the training is supervised by the training labels, and finally the normalized Brillouin bimodal frequency shift (ΔBFS) is output, and the model with the best performance on the simulation data set is selected for bimodal frequency shift extraction. In the embodiment, appropriate training parameters are selected, the training rounds are set to 40, and the training learning rate is halved every time a specific round is reached. In order to evaluate the model training effect, the performance indicators used include root mean square error (RMSE) and standard deviation (SD). The RMSE indicator measures the gap between the frequency shift extracted by the ANN model and the actual simulation frequency shift, and the SD indicator measures the discrete degree between the frequency shift extraction results.

[0066] The calculation formula of root mean square error RMSE is as follows:

[0067]

[0068] Among them, y i and are the predicted value and actual value of the i-th Brillouin double-peak frequency shift extracted from a total of N.

[0069] The calculation formula of standard deviation SD is as follows:

[0070]

[0071] Among them, y i is the extracted i-th Brillouin doublet frequency shift, is the average of all extracted Brillouin doublet frequency shifts.

[0072] S4, the effect of the trained artificial neural network model is verified using the bimodal data collected through the simulated strain experiment and the bimodal data of the actual pipeline settlement strain, and the verified artificial neural network model is used for bimodal gain spectrum frequency shift extraction and strain detection in actual settlement monitoring scenarios.

[0073] Specifically, Figure 4 The figure shows the process of using the ANN model trained by simulation for actual settlement monitoring. First, the ANN model weights obtained by simulation data training and testing are used to extract the double-peak frequency shift of actual strain; the strain data detected by pipeline settlement is input into the model to obtain the normalized Brillouin double-peak frequency shift ΔBFS, and the result is verified by the SD index; finally, the model verified by actual data is used for settlement monitoring. Since the frequency shift under actual settlement and strain conditions cannot be accurately calculated, the SD index of the extracted results under the same strain is calculated to measure the discreteness of the results. The smaller the discreteness, the better the effect. The model with a small discreteness of the output result is verified, and the verified ANN model is used for double-peak gain spectrum frequency shift extraction and strain detection in the actual settlement monitoring scenario.

[0074] In summary, a method for extracting double-peak frequency shift of Brillouin scattering based on training ANN model with simulation data can overcome the problem of small training set due to difficulty in collecting data for double-peak situation. The artificial neural network model after training and verification can accurately extract the Brillouin frequency shift of double-peak Brillouin gain spectrum caused by uneven temperature strain distribution, which has higher accuracy than the existing Lorentz fitting method. At the same time, the present invention can realize the rapid extraction of Brillouin frequency shift in double-peak situation, has good real-time performance, and helps to expand the application scope of Brillouin sensing technology.

[0075] Based on the same inventive concept, Figure 5 As shown, an embodiment of the present invention also provides a Brillouin scattering double-peak frequency shift extraction device 500 based on simulation data training ANN model, including: a simulation data set construction module 510, an artificial neural network model construction module 520, a simulation data model training module 530 and a model verification and application module 540.

[0076] The simulation data set construction module 510 is used to change the Brillouin gain spectrum parameters according to the set step size within the set parameter range, and simulate a large number of double-peak Brillouin gain spectra under different parameters within the set parameter range as simulation data sets.

[0077] The artificial neural network model construction module 520 is used to construct an artificial neural network model, which includes an input layer for receiving a simulation data set, a hidden layer for extracting data features, and an output layer for converting data features into a bimodal frequency shift.

[0078] The simulation data model training module 530 is used to train the artificial neural network model using the simulation data set, evaluate the training effect of the model by calculating the performance index, and perform bimodal frequency shift extraction using the trained artificial neural network model.

[0079] The model verification and application module 540 is used to verify the effect of the trained artificial neural network model using the bimodal data collected through the simulated strain experiment and the bimodal data of the actual pipeline settlement strain, and to use the verified artificial neural network model for bimodal gain spectrum frequency shift extraction and strain detection in actual settlement monitoring scenarios.

[0080] Based on the same inventive concept, an embodiment of the present invention provides a Brillouin scattering double-peak frequency shift extraction device based on a simulated data trained ANN model, comprising a memory and one or more processors, the memory being used to store a computer program, and the processor being used to implement the above-mentioned Brillouin scattering double-peak frequency shift extraction method based on a simulated data trained ANN model when executing the computer program.

[0081] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a computer, the above-mentioned Brillouin scattering double-peak frequency shift extraction method based on training an ANN model with simulation data is implemented.

[0082] It should be noted that the Brillouin scattering double-peak frequency shift extraction device based on simulation data training ANN model, the Brillouin scattering double-peak frequency shift extraction equipment based on simulation data training ANN model and the computer-readable storage medium provided in the above-mentioned embodiments all belong to the same inventive concept as the Brillouin scattering double-peak frequency shift extraction method based on simulation data training ANN model. The specific implementation process is detailed in the embodiment of the Brillouin scattering double-peak frequency shift extraction method based on simulation data training ANN model, which will not be repeated here.

[0083] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for extracting double-peak frequency shift of Brillouin scattering based on training ANN model with simulation data, characterized in that: The following steps are involved: The Brillouin gain spectrum parameters are changed according to a set step size within a set parameter range, and a large number of double-peak Brillouin gain spectra under different parameters within the set parameter range are obtained by simulation as a simulation data set; Constructing an artificial neural network model, which includes an input layer for receiving a simulation data set, a hidden layer for extracting data features, and an output layer for converting the data features into a bimodal frequency shift; The artificial neural network model is trained using a simulation data set, the training effect of the model is evaluated by calculating performance indicators, and the trained artificial neural network model is used to extract bimodal frequency shifts.

2. The method for extracting double-peak frequency shift of Brillouin scattering based on training ANN model with simulation data according to claim 1, characterized in that: The simulated bimodal Brillouin gain spectrum g(v) is generated according to the following formula: Among them, k is the ratio of the two peak amplitudes, g B1 and g B2 are the Brillouin gain coefficients of the left and right peaks, v is the sweep frequency, and v B1 and v B2 are the Brillouin frequency shifts of the left and right peaks, Δv B1 and Δv B2 are the half-peak widths of the two Brillouin gain spectra corresponding to the left peak and the right peak respectively.

3. The method for extracting double-peak frequency shift of Brillouin scattering based on training ANN model with simulation data according to claim 2, characterized in that: The different parameters for setting the bimodal Brillouin gain spectrum during simulation include: Set different Brillouin frequency shifts: The Brillouin frequency shift range of the left peak in the double peak is calculated by the base frequency and step size, and the Brillouin frequency shift range of the right peak in the double peak is calculated by the Brillouin frequency shift difference between the right peak and the left peak. The base frequency is set according to the type of optical fiber, and the step size is set according to the accuracy of the simulation data. Set different half-peak widths: The half-peak widths of the double peaks are calculated based on the half-peak width range and step size; Set different signal-to-noise ratios: The signal-to-noise ratios of the two peaks are calculated based on the signal-to-noise ratio range and step size; Different ratios of the double peak amplitudes are set: the ratio of the double peak amplitudes is k=k2 / (k1+k2), wherein k1 and k2 are the peak amplitudes of the left peak and the right peak in the double peaks, respectively.

4. The method for extracting double-peak frequency shift of Brillouin scattering based on training ANN model with simulation data according to claim 3, characterized in that: The setting values ​​of each parameter include: The Brillouin frequency shift is set as follows: the Brillouin frequency shift range of the left peak is the base frequency + (60MHz to 190MHz), and the Brillouin frequency shift difference range between the right peak and the left peak is 0 to 98MHz; The half-peak width is set as follows: the half-peak width range is 30MHz to 64MHz; The signal-to-noise ratio is set as follows: The signal-to-noise ratio range is 10 to 25 dB; The ratio of the two peak amplitudes is set as follows: k<0.5 simulates the case where the left peak amplitude is larger, k=0.5 simulates the case where the two peak amplitudes are equal, and k>0.5 simulates the case where the right peak amplitude is larger.

5. The method for extracting double-peak frequency shift of Brillouin scattering based on training ANN model with simulation data according to claim 1, characterized in that: The method further comprises: The effect of the trained artificial neural network model is verified using the bimodal data of actual pipeline settlement strain, and the verified artificial neural network model is used for bimodal gain spectrum frequency shift extraction and strain detection in actual settlement monitoring scenarios.

6. The method for extracting double-peak frequency shift of Brillouin scattering based on training ANN model with simulation data according to claim 1, characterized in that: When training artificial neural network models using simulation data sets, the performance indicators used include: The root mean square error RMSE is calculated as follows: Among them, y i and are the predicted value and actual value of the i-th Brillouin double-peak frequency shift among the N extracted; The standard deviation SD is calculated as follows: Among them, y i is the extracted i-th Brillouin doublet frequency shift, is the average of all extracted Brillouin doublet frequency shifts.

7. A Brillouin scattering double-peak frequency shift extraction device based on simulation data training ANN model, characterized in that: include: A simulation data set building module, an artificial neural network model building module and a simulation data model training module; The simulation data set construction module is used to change the Brillouin gain spectrum parameters according to the set step size within the set parameter range, and simulate a large number of bimodal Brillouin gain spectra under different parameters within the set parameter range as the simulation data set; The artificial neural network model construction module is used to construct an artificial neural network model, which includes an input layer for receiving a simulation data set, a hidden layer for extracting data features, and an output layer for converting the data features into a bimodal frequency shift; The simulation data model training module is used to train the artificial neural network model using the simulation data set, evaluate the training effect of the model by calculating the performance index, and perform bimodal frequency shift extraction using the trained artificial neural network model.

8. The device for extracting double-peak frequency shift of Brillouin scattering based on training ANN model with simulation data according to claim 7, characterized in that: The device also includes a model verification and application module; The model verification and application module is used to verify the effect of the trained artificial neural network model using the bimodal data collected through the simulated strain experiment and the bimodal data of the actual pipeline settlement strain, and to use the verified artificial neural network model for bimodal gain spectrum frequency shift extraction and strain detection in actual settlement monitoring scenarios.

9. A Brillouin scattering double-peak frequency shift extraction device based on simulation data training ANN model, comprising a memory and one or more processors, wherein the memory is used to store a computer program, characterized in that: The processor is used to implement the Brillouin scattering double-peak frequency shift extraction method based on simulation data training ANN model as described in any one of claims 1 to 6 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a computer, the method for extracting double-peak frequency shift of Brillouin scattering based on training ANN model with simulation data as described in any one of claims 1 to 6 is implemented.

Citation Information

Cited By

  • Brillouin temperature extraction method and system based on self-attention recurrent neural network

    CN119848440A