CN-121786794-B - PGC light intensity and phase mapping demodulation method and device based on machine learning

CN121786794BCN 121786794 BCN121786794 BCN 121786794BCN-121786794-B

Abstract

The application relates to a PGC light intensity and phase mapping demodulation method and device based on machine learning. The method comprises the steps of injecting a single-frequency laser into an interference type optical fiber sensor, applying PGC high-frequency carrier waves through internal/external modulation of a data acquisition card, outputting interference light carrying parameters to be measured by the sensor, converting the interference light into an electric signal by a photoelectric detector, and obtaining normalized long-time-domain light intensity digital signals by the acquisition card. The demodulation device divides signals according to a single PGC period and performs Fourier transformation, frequency domain features are input into a random forest regression model, modulation depth is output, and stable modulation voltage is fed back. And inputting the stable signals into a particle swarm optimization support vector machine regression model, solving the initial phase, splicing, removing the direct current elimination phase delay, and finally demodulating out the physical parameters to be detected, thereby reducing the system demodulation cost.

Inventors

HU XIAOYANG
WANG DONGYING
CHEN YUREN
MENG ZHOU
WANG JIANFEI
CHEN MO
CHEN YU
CHEN WEI
LU YANG
BIAN QIANG

Assignees

中国人民解放军国防科技大学

Dates

Publication Date: 20260508
Application Date: 20260304

Claims (10)

1. The PGC light intensity and phase mapping demodulation method based on machine learning is characterized by being applied to an interference type optical fiber sensing system consisting of a single-frequency laser, an interference type optical fiber sensor, a photoelectric detector, a data acquisition card and a signal processing demodulation device which are connected in sequence, wherein a random forest regression model, a particle swarm optimization support vector machine regression model and a feedback control module are arranged in the signal processing demodulation device, and the method comprises the following steps: the single-frequency laser outputs single-frequency light to the interference type optical fiber sensor, the data acquisition card applies PGC high-frequency carrier modulation to the single-frequency light or the interference type optical fiber sensor in an internal modulation or external modulation mode, and the interference type optical fiber sensor outputs interference light carrying physical parameters to be detected; the photoelectric detector converts the interference light into an electric signal, and the data acquisition card converts the electric signal into a normalized long time domain light intensity digital signal and transmits the normalized long time domain light intensity digital signal to the signal processing demodulation device; The signal processing demodulation device divides the normalized long time domain light intensity digital signal into a plurality of normalized short time domain light intensity signals with the length of a single PGC modulation period, and the normalized short time domain light intensity signals are used as input data of a random forest regression model and a particle swarm optimization support vector machine regression model; The feedback control module adjusts the PGC modulation voltage of the data acquisition card in real time according to the PGC modulation depth, stabilizes the system PGC modulation depth to a fixed value, and obtains a normalized short time domain light intensity signal after parameter stabilization; The signal processing demodulation device inputs all normalized short time domain light intensity signals with stable parameters into a trained particle swarm optimization support vector machine regression model, outputs initial phase values corresponding to all normalized short time domain light intensity signals, splices all the initial phase values according to time sequence to obtain phase signals containing direct current constant items, performs direct current removal operation on the phase signals containing the direct current constant items, eliminates the direct current constant items introduced by system phase delay, and demodulates the physical parameter signals to be detected.
2. The method of claim 1, wherein the inner modulation is that the data acquisition card directly applies PGC high-frequency carrier modulation to single-frequency light output by the single-frequency laser, and the outer modulation is that the data acquisition card applies PGC high-frequency carrier modulation to piezoelectric ceramics wound on an interference arm of the interference type optical fiber sensor, and phase modulation is realized by driving the interference arm through the piezoelectric ceramics.
3. The method of claim 1, wherein the physical parameter to be measured is one or more of sound, vibration, or temperature.
4. The method of claim 1, wherein the training process of the random forest regression model comprises the steps of collecting normalized short-time domain light intensity signals output by the interference type optical fiber sensor under different PGC modulation depths, obtaining frequency domain characteristic signals through Fourier transformation, constructing a sample data set of the frequency domain characteristic signals-PGC modulation depths, inputting the sample data set into the random forest regression model for training, and enabling the model to learn a nonlinear mapping relation between the frequency domain characteristic signals and the PGC modulation depths.
5. The method of claim 1, wherein the training process of the particle swarm optimization support vector machine regression model comprises the steps of collecting normalized short time domain light intensity signals output by an interference type optical fiber sensor and corresponding real initial phase values under a fixed PGC modulation depth, constructing a sample data set of light intensity signals-initial phases, carrying out global optimization on kernel function parameters and penalty factors of the support vector machine regression model by adopting a particle swarm optimization algorithm, determining an optimal super-parameter combination, inputting the sample data set into the support vector machine regression model of the optimal super-parameter combination, and training the support vector machine regression model to enable the model to learn a nonlinear mapping relation between the normalized short time domain light intensity signals and the initial phases.
6. The method according to claim 1, wherein the dc removal operation uses a mean value subtraction method, calculates a mean value of the phase signals containing the dc constant term after the splicing, and subtracts the mean value from each phase value to obtain the physical parameter signal to be measured with the dc constant term eliminated.
7. The method of claim 1, wherein the particle swarm optimization support vector machine regression model treats a normalized short time domain light intensity signal of a single PGC modulation period length as an independent phase prediction unit, and the single unit correspondingly outputs an initial phase value.
8. The method of claim 1, wherein the division mode of the normalized long-time domain light intensity digital signal is non-overlapping equal-length division, and the time length of each divided normalized short-time domain light intensity signal is completely consistent with the period of the PGC high-frequency carrier modulation, so as to ensure that each short-time domain signal contains a complete single-period PGC modulation feature.
9. The method according to claim 1, wherein the signal processing demodulation device is a data processing computer or an FPGA development board, and the random forest regression model and the particle swarm optimization support vector machine regression model are deployed in the signal processing demodulation device by means of software programming or a hardware logic circuit, so as to implement PGC modulation depth calculation and real-time parallel processing of light intensity-phase mapping.
10. The PGC light intensity and phase mapping demodulation device based on machine learning is characterized by comprising a single-frequency laser, an interference type optical fiber sensor, a photoelectric detector, a data acquisition card and a signal processing demodulation device which are sequentially connected, wherein a random forest regression model, a particle swarm optimization support vector machine regression model and a feedback control module are arranged in the signal processing demodulation device; the single-frequency laser outputs single-frequency light to the interference type optical fiber sensor, the data acquisition card applies PGC high-frequency carrier modulation to the single-frequency light or the interference type optical fiber sensor in an internal modulation or external modulation mode, and the interference type optical fiber sensor outputs interference light carrying physical parameters to be detected; the photoelectric detector converts the interference light into an electric signal, and the data acquisition card converts the electric signal into a normalized long time domain light intensity digital signal and transmits the normalized long time domain light intensity digital signal to the signal processing demodulation device; The signal processing demodulation device divides the normalized long time domain light intensity digital signal into a plurality of normalized short time domain light intensity signals with the length of a single PGC modulation period, and the normalized short time domain light intensity signals are used as input data of a random forest regression model and a particle swarm optimization support vector machine regression model; The feedback control module adjusts the PGC modulation voltage of the data acquisition card in real time according to the PGC modulation depth, stabilizes the system PGC modulation depth to a fixed value, and obtains a normalized short time domain light intensity signal after parameter stabilization; The signal processing demodulation device inputs all normalized short time domain light intensity signals with stable parameters into a trained particle swarm optimization support vector machine regression model, outputs initial phase values corresponding to all normalized short time domain light intensity signals, splices all the initial phase values according to time sequence to obtain phase signals containing direct current constant items, performs direct current removal operation on the phase signals containing the direct current constant items, eliminates the direct current constant items introduced by system phase delay, and demodulates the physical parameter signals to be detected.

Description

PGC light intensity and phase mapping demodulation method and device based on machine learning Technical Field The application relates to the technical field of interference type optical fiber sensing, in particular to a PGC light intensity and phase mapping demodulation method and device based on machine learning. Background The Phase Generating Carrier (PGC) method is a conventional demodulation method for demodulating an interferometric phase-change optical fiber sensor. In an interference type optical fiber sensing system, a traditional PGC method firstly carries out phase high-frequency carrier modulation on single-frequency light output by a single-frequency laser through external modulation or internal modulation, the modulated single-frequency light is input into an interference type optical fiber sensor, and the sensor outputs output light carrying a target signal to be detected. The output light is processed by the photoelectric detector and the data acquisition card to generate a digital signal, the digital signal is sent to the signal processing demodulation device, and finally the signal to be detected is demodulated through the steps of carrier signal mixing, low-pass filtering, arctangent or differential multiplication. The demodulation process of the traditional PGC demodulation method is complex, and particularly in a large-scale interference optical fiber sensor array system, the required calculation resources are large and the power consumption is high. Secondly, PGC demodulation parameters often change due to changes in operating environment parameters during operation of the interferometric fiber optic sensing system. And the PGC demodulation result is sensitive to the PGC demodulation parameters, and the accuracy of the demodulation result can be ensured by rapidly and accurately calculating the PGC demodulation parameters. However, the existing PGC demodulation parameter calculation method has the problems of large calculation amount, complex calculation method and the like. Breaks through the existing limitation of the traditional PGC demodulation method, and can improve the sensing performance of the interference type optical fiber sensing system. Disclosure of Invention In view of the foregoing, it is desirable to provide a method and apparatus for demodulating PGC light intensity and phase mapping based on machine learning, which can reduce demodulation overhead of an interferometric fiber sensing system. The PGC light intensity and phase mapping demodulation method based on machine learning is applied to an interference type optical fiber sensing system consisting of a single-frequency laser, an interference type optical fiber sensor, a photoelectric detector, a data acquisition card and a signal processing demodulation device which are connected in sequence, wherein the signal processing demodulation device is internally provided with a random forest regression model, a particle swarm optimization support vector machine regression model and a feedback control module, and the method comprises the following steps: the single-frequency laser outputs single-frequency light to the interference type optical fiber sensor, the data acquisition card applies PGC high-frequency carrier modulation to the single-frequency light or the interference type optical fiber sensor in an internal modulation or external modulation mode, and the interference type optical fiber sensor outputs interference light carrying physical parameters to be detected; The photoelectric detector converts interference light into an electric signal, the data acquisition card converts the electric signal into a normalized long time domain light intensity digital signal, and the normalized long time domain light intensity digital signal is transmitted to the signal processing demodulation device; The signal processing demodulation device divides the normalized long time domain light intensity digital signal into a plurality of normalized short time domain light intensity signals with the length of a single PGC modulation period, and the normalized short time domain light intensity signals are used as input data of a random forest regression model and a particle swarm optimization support vector machine regression model; The feedback control module adjusts the PGC modulation voltage of the data acquisition card in real time according to the PGC modulation depth, stabilizes the system PGC modulation depth to a fixed value, and obtains a normalized short time domain light intensity signal after parameter stabilization; The signal processing demodulation device inputs all normalized short time domain light intensity signals with stable parameters into a trained particle swarm optimization support vector machine regression model, outputs initial phase values corresponding to all normalized short time domain light intensity signals, splices all the initial phase values according to time sequence to obtain phase signals contain