Machine learning based pgc parameter calibration and error suppression method

Abstract

The application relates to a PGC parameter calibration and error suppression method based on machine learning. The method comprises the following steps: a single-frequency laser outputs input light, and interference type optical fiber sensors process the input light to generate interference light. The interference light is converted into an electric signal by a photoelectric detector, and then the electric signal is converted into a digital signal by a data acquisition card and transmitted to a signal processing demodulation device. The digital signal is mixed and low-pass filtered to obtain a sub-signal affected by a modulation depth and a phase delay. Meanwhile, a short time domain signal output by the sensor or a Fourier transform signal thereof is input into two trained machine learning models to quickly output the modulation depth and the phase delay. The sub-signal is compensated, and the ratio of the compensated sub-signal is subjected to an inverse tangent operation to finally obtain a to-be-measured signal in which PGC parameter offset distortion is eliminated. The method can break through the limitations of the existing PGC parameter offset demodulation error calibration method.

Description

Technical Field

[0001] This application relates to the field of interferometric fiber optic sensing technology, and in particular to a machine learning-based method for PGC parameter calibration and error suppression. Background Technology

[0002] Phase-generated carrier (PGC) technology is a demodulation method for interferometric phase-change fiber optic sensors. It extracts changes in the measured physical quantity by generating a phase carrier signal. In this method, the modulation depth C and phase delay are related... θ c The values ​​of the parameters affect the accuracy of demodulation. Accurate acquisition and calibration of the modulation depth and phase delay ensure the reliability of the demodulation effect. In 2017, a novel PGC demodulation scheme based on integral control feedback was proposed, involving phase modulation depth evaluation and calibration. However, the operating point of the sensor output signal affects the accuracy of the modulation depth calculation. In 2020, a PGC technology acquired a complete Lissajous figure through active laser wavelength scanning, overcoming the limitations of traditional ellipse fitting methods, and thus calibrating PGC parameter offsets. However, this method is overly complex and costly to implement. In 2024, a PGC demodulation technique was proposed that uses a nonlinear fitting algorithm to suppress PGC parameter offsets in interferometric measurement systems. This method uses a nonlinear fitting algorithm to fit the elliptical parameters of a Lissajous circle, improving fitting accuracy and ultimately achieving signal correction and PGC parameter offset error suppression. However, this method is also overly complex and computationally intensive.

[0003] Currently, many parameter offset error calibration methods for PGC technology have been proposed, but most of them face challenges such as large computational load, complex calibration methods, and failure of calibration methods at certain signal operating points. Summary of the Invention

[0004] Therefore, it is necessary to address the aforementioned technical problems by providing a method that can quickly obtain the modulation depth of PGC demodulation technology. C Phase delay θ、 A machine learning-based method for PGC parameter calibration and error suppression that overcomes the limitations of existing PGC parameter offset error calibration methods.

[0005] A machine learning-based method for PGC parameter calibration and error suppression, applied to an interferometric fiber optic sensing system including a single-frequency laser, an interferometric fiber optic sensor, a photodetector, a data acquisition card, and a signal processing and demodulation device, comprising:

[0006] Based on the output of the single-frequency laser, the input light is transmitted to an interferometric fiber optic sensor; the interferometric fiber optic sensor performs interference processing on the input light, outputs interference light, and transmits the interference light to a photodetector;

[0007] The photodetector converts the interference light signal into an electrical signal, which is then transmitted to the data acquisition card. The data acquisition card then converts the electrical signal into a digital signal, which is transmitted to the signal processing and demodulation device.

[0008] The signal processing and demodulation device performs mixing and low-pass filtering on the digital signal to obtain a sub-signal affected by modulation depth and phase delay. It then acquires the short-time domain signal or its Fourier transform from the output of the interferometric fiber optic sensor, inputs it into two pre-trained machine learning models, and outputs the modulation depth and phase delay respectively. Based on the modulation depth and phase delay, the sub-signal is compensated to obtain a compensated sub-signal. Finally, the arctangent of the ratio of the compensated sub-signals is calculated to obtain the test signal that suppresses the demodulation error of the PGC parameter offset.

[0009] The aforementioned machine learning-based PGC parameter calibration and error suppression method first completes the ordered conversion of optical signals to digital signals using a single-frequency laser, an interferometric fiber optic sensor, a photodetector, and a data acquisition card, providing a high-quality signal foundation for subsequent processing. Simultaneously, it directly extracts the short-time domain signal or its Fourier transform signal from the interferometric fiber optic sensor as input to the machine learning model, significantly reducing the computational load of data acquisition and preprocessing compared to the large number of long signal samples required by traditional methods. Secondly, it incorporates two pre-trained machine learning models to solve for modulation depth and phase delay, respectively. These models, through pre-training the mapping relationship between signal samples and parameter values ​​under different parameters, can directly output accurate parameters without the complex fitting or iterative calculations of traditional methods, achieving rapid parameter acquisition. Furthermore, the model training process covers samples at different signal operating points, breaking through the limitations of traditional methods on signal operating points and enabling stable parameter output at any operating point. Finally, the method utilizes the modulation depth and phase delay output by the machine learning model to specifically compensate the sub-signals obtained from mixing and low-pass filtering. The measured signal is then obtained through calculation, effectively eliminating distortion caused by variations in modulation depth and phase delay. This eliminates the need for complex calibration algorithms, further simplifying the operation and improving real-time performance. In summary, this application, through innovation in the parameter solving process using machine learning technology, fundamentally solves the pain points of traditional techniques, achieving efficient, accurate, and robust PGC demodulation parameter offset error calibration. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating a machine learning-based PGC parameter calibration and error suppression method in one embodiment.

[0011] Figure 2 This is a schematic diagram of the technical process of one specific embodiment in one example;

[0012] Figure 3 This is a schematic diagram of the technical process of another specific embodiment in one example;

[0013] Figure 4 One embodiment uses machine learning to quickly obtain the modulation depth in PGC. C Phase delay θ c And demodulation may result in PGC parameter offset distortion. I ( t ), Q ( t The Lissajous figures before and after compensation are shown in the figure; Figure 4 (a) is the cause C Changes that cause horizontal or vertical stretching I ( t ), Q ( t Lissard, as shown in the figure, Figure 4 (b) After compensation and correction I’ ( t ), Q’ ( t The orthogonal Lissajous figures are shown in the diagram. Figure 4 (c) is the cause θ c Changes that cause horizontal or vertical stretching I ( t ), Q ( t Lissard, as shown in the figure, Figure 4 (d) After compensation and correction I’ ( t ), Q’ ( t The orthogonal Lissajous figures are shown in the diagram. Figure 4 (e) is the cause θ c and C Simultaneous changes that cause horizontal or vertical stretching I ( t ), Q ( t Lissard, as shown in the figure, Figure 4 (f) After compensation and correction I’ ( t ), Q’ ( t The orthogonal Lissajous figures are shown in the diagram.

[0014] Figure 5 In another embodiment, the Lissajous figures of the two sub-signals before and after correcting the offset errors of the three PGC parameters are shown. Figure 5 (a) is a schematic diagram showing the distortion of the demodulated signal caused by changes in modulation depth. Figure 5(b) Schematic diagram for correcting PGC parameter offset error in demodulated signal. Figure 5 (c) is a schematic diagram showing the distortion of the demodulated signal caused by the phase delay. Figure 5 (d) is a schematic diagram for correcting the PGC parameter offset error of the demodulated signal. Figure 5 (e) is a schematic diagram showing the distortion of the demodulated signal caused by changes in modulation depth and phase delay. Figure 5 (f) is a schematic diagram of correcting the PGC parameter offset error of the demodulated signal;

[0015] Figure 6 In one embodiment, a random forest regression model is used to calculate the modulation depth under environments with no low-frequency noise, 10Hz noise, and 50Hz noise. C Performance graph; Figure 6 (a) Modulation depth in the absence of low-frequency noise C Calculation results graph Figure 6 (b) Modulation depth under 10Hz noise conditions C Calculation results graph Figure 6 (c) Modulation depth under 50Hz noise conditions C Calculation results graph;

[0016] Figure 7 This is a schematic diagram illustrating the modulation depth calculated using two methods in one embodiment; Figure 7 (a) is a schematic diagram of calculating sample modulation depth using traditional methods. Figure 7 (b) is a schematic diagram of calculating the sample modulation depth using the random forest regression model;

[0017] Figure 8 This is a schematic diagram illustrating the calculation of the phase delay of a sample using two methods in one embodiment; Figure 8 (a) is a schematic diagram of calculating the phase delay of a sample using traditional methods. Figure 8 (b) is k A schematic diagram illustrating the calculation of the phase delay of a sample using a nearest neighbor regression model. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] In one embodiment, such as Figure 1 As shown, a machine learning-based method for PGC parameter calibration and error suppression is provided. This method is applied to an interferometric fiber optic sensing system including a single-frequency laser, an interferometric fiber optic sensor, a photodetector, a data acquisition card, and a signal processing and demodulation device. The method includes:

[0020] Step 102: Based on the input light output from the single-frequency laser, the input light is transmitted to the interferometric fiber optic sensor; the interferometric fiber optic sensor is used to perform interference processing on the input light, outputting interference light, which is then transmitted to the photodetector.

[0021] A single-frequency laser refers to a laser whose output laser frequency is singular and stable, providing a stable input light for this sensing system. Interferometric fiber optic sensors can utilize Michelson interferometers or Mach-Zehnder interferometers. After the input light enters the sensor, it is split into two beams. The two beams interfere due to the optical path difference, forming interference light. This interference light is modulated by external environmental physical parameters, resulting in a phase difference. The photodetector converts the optical signal into an electrical signal, providing the foundation for subsequent digital processing.

[0022] Step 104: The photodetector converts the interference light signal into an electrical signal and transmits the electrical signal to the data acquisition card; the data acquisition card converts the electrical signal into a digital signal and transmits the digital signal to the signal processing and demodulation device.

[0023] The data acquisition card is a key device connecting analog electrical signals and digital signals. It converts the continuous electrical signals output by the photodetector into discrete digital signals. The signal processing and demodulation unit can be a data processing computer or an FPGA development board, etc. It can receive digital signals and perform subsequent mixing, filtering, parameter calculation, and signal compensation operations. This realizes the digital conversion of signals and provides a carrier for subsequent digital signal processing.

[0024] Step 106: The digital signal is mixed and low-pass filtered by the signal processing demodulation device to obtain a sub-signal affected by modulation depth and phase delay; the short-time domain signal or the Fourier transform signal of the short-time domain signal output by the interferometric fiber optic sensor is acquired, and the short-time domain signal or the Fourier transform signal of the short-time domain signal is input into two built-in trained machine learning models, which output the modulation depth and phase delay respectively; the sub-signal is compensated according to the modulation depth and phase delay to obtain the compensated sub-signal; the arctangent operation is performed on the ratio of the compensated sub-signal to obtain the test signal with PGC parameter offset demodulation error eliminated.

[0025] Mixing and low-pass filtering specifically refer to using the first harmonic term of the high-frequency carrier, cos( ω 0 t ) and the second harmonic of the high-frequency carrier cos(2 ω 0 t )right S ( t The signals are then mixed and low-pass filtered to obtain two sub-signals. Q ( t )and I ( tShort-time domain signals refer to short-duration signal samples output by interferometric fiber optic sensors. Compared to the long signal samples required by traditional methods, they have the advantages of small data volume and fast processing speed. Performing Fourier transform on short-time domain signals can convert the signal from the time domain to the frequency domain, which can better extract the frequency characteristics of the signal.

[0026] The signal processing and demodulation device has two built-in machine learning models for calculating phase delay. θ c of k Nearest neighbor regression model, calculating modulation depth C The random forest regression model. k Nearest neighbor regression models are trained by different systems with different phase delays. θ c Below, a large number of short-time domain signals or their Fourier transform samples output by interferometric fiber optic sensors and phase delays θ c The mapping relationship is obtained; the random forest regression model is obtained by training different systems to modulate the depth. C Below, a large number of short-time domain signals or their Fourier transform samples output by interferometric fiber optic sensors and modulation depth C The mapping relationship is obtained.

[0027] By inputting a short-time domain signal or its Fourier transform signal into the two models mentioned above, the modulation depth can be quickly output. C and phase delay θ c Then, these two parameters are used to analyze the sub-signals. I ( t ), Q ( t Compensation is performed to obtain the compensated sub-signal. Q’ ( t )and I’ ( t ).

[0028] Finally, by solving the compensated sub-signal, the test signal with PGC parameter offset demodulation error eliminated can be obtained.

[0029] The aforementioned machine learning-based PGC parameter calibration and error suppression method first completes the ordered conversion of optical signals to digital signals using a single-frequency laser, an interferometric fiber optic sensor, a photodetector, and a data acquisition card, providing a high-quality signal foundation for subsequent processing. Simultaneously, it directly extracts the short-time domain signal or its Fourier transform signal from the interferometric fiber optic sensor as input to the machine learning model, significantly reducing the computational load of data acquisition and preprocessing compared to the large number of long signal samples required by traditional methods. Secondly, it incorporates two pre-trained machine learning models to solve for modulation depth and phase delay, respectively. These models, through pre-training the mapping relationship between signal samples and parameter values ​​under different parameters, can directly output accurate parameters without the complex fitting or iterative calculations of traditional methods, achieving rapid parameter acquisition. Furthermore, the model training process covers samples at different signal operating points, breaking through the limitations of traditional methods on signal operating points and enabling stable parameter output at any operating point. Finally, the method utilizes the modulation depth and phase delay output by the machine learning model to specifically compensate the sub-signals obtained from mixing and low-pass filtering. The measured signal is then obtained through calculation, effectively eliminating the PGC parameter offset demodulation error caused by variations in modulation depth and phase delay. This eliminates the need for complex calibration algorithms, further simplifying the operation and improving real-time performance. In summary, this application, through innovation in the parameter solving process using machine learning technology, fundamentally solves the pain points of traditional techniques, achieving efficient, accurate, and robust suppression of PGC parameter offset demodulation errors.

[0030] In one embodiment, the single-frequency laser, interferometric fiber optic sensor, photodetector, data acquisition card, and signal processing demodulation device are connected sequentially in the interferometric fiber optic sensing system.

[0031] Specifically, Figure 2 This is a schematic diagram of the technical process of one specific embodiment in one example; Figure 3 This is a schematic diagram of the technical flow of another specific embodiment in one example; a single-frequency laser 1, an interferometric fiber optic sensor 2, a photodetector 3, a data acquisition card 4, and a signal processing and demodulation device 5 are connected in sequence. The output end of the single-frequency laser 1 is connected to the input end of the interferometric fiber optic sensor 2, and the single-frequency laser provides input light to the interferometric fiber optic sensor. The input light input to the fiber optic sensor is split into two beams, and the two beams interfere with each other due to the optical path difference. The interference light can be influenced by external environmental physical parameter signals. φ ( tThe phase difference is generated by modulation of signals (such as temperature, sound, vibration, etc.). Interferometric sensing systems can be divided into internal modulation schemes and external modulation schemes. In the internal modulation scheme, the data acquisition card 4 is connected to the single-frequency laser 1 and directly modulates the output light of the single-frequency laser 1 with a high-frequency carrier. In the external modulation scheme, the interferometric fiber optic sensor 2 is wound with a piezoelectric ceramic sheet, and the data acquisition card 4 is connected to the interferometric fiber optic sensor 2 to modulate the piezoelectric ceramic sheet with a high-frequency carrier. After internal or external modulation, the interference light output by the interferometric fiber optic sensor 2 is modulated by the high-frequency carrier. The output end of the interferometric fiber optic sensor 2 is connected to the input end of the photodetector 3, which converts the interference light signal into an electrical signal. The output end of the photodetector 3 is connected to the input end of the data acquisition card 4, which converts the electrical signal into a long digital signal. S ( t The signal is transmitted to the signal processing and demodulation device 5.

[0032] In one embodiment, the conversion of electrical signals into digital signals according to the data acquisition card includes:

[0033] The data acquisition card converts the electrical signal into a digital signal as follows:

[0034] ;

[0035] in, A The DC term of a digital signal. B The amplitude of the AC term representing a digital signal. C The high-frequency carrier modulation depth for PGC demodulation technology. ω 0 represents the high-frequency carrier modulation frequency of the PGC demodulation technology. θ c For system phase delay, φ ( t ) represents the environmental signal to be measured.

[0036] Specifically, this digital signal fully characterizes the signal features of the input light after interference modulation, including key information such as DC component, AC component, carrier frequency, modulation depth, phase delay, and the signal under test. Through this expression, the influence of each parameter on the signal can be clearly analyzed, providing a theoretical basis for subsequent mixing and filtering, parameter calculation, and signal compensation, and possessing clear physical significance and practical value.

[0037] In one embodiment, the digital signal is mixed and low-pass filtered by the signal processing demodulation device to obtain a sub-signal affected by modulation depth and phase delay, including:

[0038] The signal processing demodulation device utilizes the first harmonic term of the high-frequency carrier, cos( ω 0 t) and the second harmonic term of the high-frequency carrier cos(2 ω 0 t )right S ( t The signal is mixed and low-pass filtered to obtain two sub-signals affected by modulation depth and phase delay. Q ( t )and I ( t They are respectively:

[0039] ;

[0040] in, C The high-frequency carrier modulation depth for PGC demodulation technology. θ c For system phase delay, φ ( t () represents the environmental signal to be measured. B The amplitude of the AC term representing a digital signal. J 1( C )and J 2( C ) are the first-order and second-order Bessel functions of the first kind, respectively.

[0041] The purpose of mixing is to extract the signal components related to the carrier frequency, while low-pass filtering removes high-frequency noise generated during mixing, thereby obtaining a sub-signal containing information about the signal under test. I (t) and Q (t). These two sub-signals are directly affected by the modulation depth. C and phase delay θ c If these two parameters are not calibrated and compensated, the orthogonality of the sub-signals will be destroyed, leading to errors in the demodulated signal. This provides the processing object for subsequent parameter calibration and signal compensation, and is a crucial preliminary step in achieving PGC parameter offset error suppression.

[0042] In one embodiment, the two trained machine learning models are respectively for calculating phase delay. k Nearest neighbor regression model, random forest regression model for calculating modulation depth; the aforementioned k The nearest neighbor regression model is obtained by training the mapping relationship between the short-time domain signals or their Fourier transform samples and the phase delay of a large number of interferometric fiber optic sensors under different system phase delays; the random forest regression model is obtained by training the mapping relationship between the short-time domain signals or their Fourier transform samples and the modulation depth of a large number of interferometric fiber optic sensors under different system modulation depths.

[0043] Specifically, kThe core principle of the nearest neighbor regression model is prediction based on the similarity between samples. By training a large number of signal samples with different phase delays, the model can learn the mapping relationship between signal features and phase delay. After inputting a new short-time domain signal or its Fourier transform signal, the model can quickly match the most similar training sample and then output the corresponding phase delay value. The random forest regression model is an ensemble learning model composed of multiple decision trees. By training a large number of signal samples with different modulation depths, the model can learn the complex mapping relationship between signal features and modulation depth, exhibiting high prediction accuracy and resistance to overfitting. Both models only require a single short-time domain signal sample to complete parameter calculation, offering advantages over traditional methods in terms of low computational cost and fast processing speed. Furthermore, after training, they can be applied to interferometer fiber optic sensors with different arm differences, demonstrating strong generalization ability.

[0044] In one embodiment, the sub-signal is compensated based on the modulation depth and phase delay to obtain a compensated sub-signal, including:

[0045] Based on the modulation depth and phase delay, the sub-signals are compensated to obtain the compensated sub-signals as follows:

[0046] ;

[0047] in, C The high-frequency carrier modulation depth for PGC demodulation technology. θ c For system phase delay, φ ( t () represents the environmental signal to be measured. B The amplitude of the AC term representing a digital signal. J 1( C )and J 2( C ) are the first-order and second-order Bessel functions of the first kind, respectively.

[0048] In one embodiment, the environmental signal to be measured is solved as follows:

[0049] .

[0050] Specifically, such as Figure 4 As shown, Figure 4 (a) Figure 4 (c) Figure 4 (e) are due to modulation depth C Changes, phase delay θ c Variation, modulation depth C Phase delay θc Simultaneous changes that cause horizontal or vertical stretching I ( t ), Q ( t As shown in the Lissajous figure, the Lissajous figure presents a non-orthogonal elliptical shape, reflecting the existence of PGC parameter offset demodulation error in the demodulated signal. Figure 4 (b) Figure 4 (d) Figure 4 (f) represents the compensation and correction results under the corresponding conditions. Q’ ( t )and I’ ( t The orthogonal Lissajous figure is shown in the figure. At this time, the Lissajous figure presents a standard circle, indicating that the PGC parameter offset demodulation error of the sub-signal is effectively suppressed.

[0051] like Figure 5 As shown, Figure 5 (a) Figure 5 (c) Figure 5 (e) are schematic diagrams of demodulated signal distortion caused by changes in modulation depth, phase delay, and simultaneous changes in modulation depth and phase delay, respectively. The demodulated signal deviates significantly from the standard signal. Figure 5 (b) Figure 5 (d) Figure 5 (f) are schematic diagrams after correcting the demodulated signal error under the corresponding conditions. The demodulated signal basically overlaps with the standard signal, which verifies the PGC parameter demodulation error calibration capability of this method.

[0052] like Figure 6 As shown, Figure 6 (a) Figure 6 (b) Figure 6 (c) The modulation depth calculated by the random forest regression model under environments with no low-frequency noise, 10Hz noise, and 50Hz noise, respectively. C The performance graphs and experimental results show that, under these three environments, the model supports nine different modulation depths. C The calculations on a sample of approximately 5000 samples all achieved an absolute average error of modulation depth of less than 0.015 rad, demonstrating that the method has good noise robustness.

[0053] like Figure 7 As shown, Figure 7 (a) is a schematic diagram of the traditional method for calculating the sample modulation depth. The traditional method requires the operating point of the signal sample to be far away from 0 rad and π / 2 rad. Within this operating point range, the calculation results have a large deviation. Figure 7(b) is a schematic diagram of the random forest regression model calculating the modulation depth of the sample. Under any signal operating point from 0 rad to 1.6 rad, the modulation depth calculated by the model is stable at around 2.63 rad with minimal deviation, proving that the method is not limited by the signal sample operating point.

[0054] like Figure 8 As shown, Figure 8 (a) is a schematic diagram of the traditional method for calculating the sample phase delay. The traditional method requires the operating point of the signal sample to be far from 0 rad and close to π / 2 rad. The calculation results are significantly biased at other operating points. Figure 8 (b) is a schematic diagram of the k-nearest neighbor regression model calculating the sample phase delay. Under any signal operating point from 0 rad to 1.6 rad, the phase delay calculated by the model is stable at around 0.858 rad with minimal deviation, further verifying that the method is not limited by the signal sample operating point.

[0055] This application has the following advantages over other PGC parameter demodulation error calibration methods: (1) This application only requires a short signal sample from a single interferometric fiber optic sensor to calculate the modulation depth and phase delay, which has low computational requirements. Traditional methods require a large number of signal samples and a large number of sample data points. (2) This application uses a machine learning model to quickly calculate the modulation depth and phase delay. After the machine learning model is trained, it can be applied to interferometric fiber optic sensors with different arm differences, and has strong generalization. (3) Compared with other PGC demodulation parameter offset distortion calibration methods, this application has the ability to quickly calculate the modulation depth and phase delay, so the obtained modulation depth and phase delay can be used to perform real-time and fast calibration of the demodulation results of PGC technology. (4) Compared with other PGC parameter offset demodulation error calibration methods, this application does not require the data operating point of the output signal of the interferometric fiber optic sensor. For signal samples at any operating point, the modulation depth and phase delay can be quickly calculated.

[0056] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0057] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0058] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for PGC parameter calibration and error suppression based on machine learning, characterized in that, The method is applied to an interferometric fiber optic sensing system including a single-frequency laser, an interferometric fiber optic sensor, a photodetector, a data acquisition card, and a signal processing and demodulation device, including: The input light from the single-frequency laser is transmitted to an interferometric fiber optic sensor; the interferometric fiber optic sensor performs interference processing on the input light, outputs interferometric light, and transmits the interferometric light to a photodetector. The photodetector converts the interference light signal into an electrical signal, which is then transmitted to a data acquisition card. The data acquisition card then converts the electrical signal into a digital signal, which is transmitted to a signal processing and demodulation device. The signal processing and demodulation device performs mixing and low-pass filtering on the digital signal to obtain a sub-signal affected by modulation depth and phase delay. It then acquires a short-time domain signal or its Fourier transform from the output of an interferometric fiber optic sensor, inputs it into two pre-trained machine learning models, and outputs the modulation depth and phase delay respectively. Based on the modulation depth and phase delay, the sub-signal is compensated to obtain a compensated sub-signal. Finally, an arctangent operation is performed on the ratio of the compensated sub-signals to obtain the test signal with PGC parameter offset demodulation error eliminated. The two trained machine learning models are used to calculate phase delay. k Nearest neighbor regression model, random forest regression model for calculating modulation depth; the aforementioned k The nearest neighbor regression model is obtained by training the mapping relationship between the short-time domain signals or their Fourier transform samples and the phase delay of a large number of interferometric fiber optic sensors under different system phase delays; the random forest regression model is obtained by training the mapping relationship between the short-time domain signals or their Fourier transform samples and the modulation depth of a large number of interferometric fiber optic sensors under different system modulation depths. Based on the modulation depth and phase delay, the sub-signal is compensated to obtain the compensated sub-signal, including: Based on the modulation depth and phase delay, the sub-signals are compensated to obtain the compensated sub-signals as follows: in, C The high-frequency carrier modulation depth for PGC demodulation technology. θ c For system phase delay, φ ( t () represents the environmental signal to be measured. B The amplitude of the AC term representing a digital signal. J 1( C )and J 2( C ) are the first-order and second-order Bessel functions of the first kind, respectively; The solution method for the environmental signal to be measured is as follows: 。 2. The method according to claim 1, characterized in that, The single-frequency laser, interferometric fiber optic sensor, photodetector, data acquisition card, and signal processing and demodulation device are connected sequentially in the interferometric fiber optic sensing system.

3. The method according to claim 1, characterized in that, The data acquisition card converts electrical signals into digital signals, including: The data acquisition card converts the electrical signal into a digital signal. in, A The DC term of a digital signal. B The amplitude of the AC term representing a digital signal. C The high-frequency carrier modulation depth for PGC demodulation technology. The high-frequency carrier modulation frequency for PGC demodulation technology. θ c For system phase delay, φ ( t ) represents the environmental signal to be measured.

4. The method according to claim 1, characterized in that, The digital signal is mixed and low-pass filtered by the signal processing and demodulation device to obtain a sub-signal affected by modulation depth and phase delay, including: The signal processing demodulation device utilizes the first harmonic term of the high-frequency carrier, cos( ω 0 t ) and the second harmonic term of the high-frequency carrier cos(2 ω 0 t )right S ( t The signal is mixed and low-pass filtered to obtain two sub-signals affected by modulation depth and phase delay. Q ( t )and I ( t ) are respectively in, C The high-frequency carrier modulation depth for PGC demodulation technology. θ c For system phase delay, φ ( t () represents the environmental signal to be measured. B The amplitude of the AC term representing a digital signal. J 1( C )and J 2( C ) are the first-order and second-order Bessel functions of the first kind, respectively.

Images