CN116539018A

CN116539018A - Machine learning-assisted optical fiber gyroscope polarization locking and slow drift compensation system and method

Info

Publication number: CN116539018A
Application number: CN202310502448.2A
Authority: CN
Inventors: 陈丽清; 赵杰; 吴媛; 黄文峰; 梁馨云
Original assignee: East China Normal University
Current assignee: East China Normal University
Priority date: 2023-04-24
Filing date: 2023-05-06
Publication date: 2023-08-04

Abstract

The invention discloses a machine learning-assisted optical fiber gyroscope polarization locking and slow drift compensation system. The system can intelligently search and predict the relation between the interference error signal of the optical fiber gyroscope and the polarization direction of the incident light by introducing a neural network in machine learning, and feeds the predicted optimal value back to a hardware system for compensation, thereby realizing the functions of locking the optical fiber polarization and inhibiting the intensity drift of an output signal. The system comprises a coherent light source, a gram laser prism, a first polarization beam splitter, a first reflecting mirror, a first Faraday rotator, a first half wave plate, a liquid crystal modulator, a polarization maintaining optical fiber, a phase modulator, a second Faraday rotator, a second half wave plate, a second reflecting mirror, a third half wave plate, a second polarization beam splitter, a photoelectric detector, a computer and a signal generator. The invention also discloses a machine learning-assisted optical fiber gyroscope polarization locking and slow drift compensation method realized by the system.

Description

Machine learning-assisted optical fiber gyroscope polarization locking and slow drift compensation system and method

Technical Field

The invention belongs to the fields of artificial intelligence, electronic control, quantum optics and precise measurement, and mainly relates to a machine learning-assisted optical fiber gyroscope polarization locking and slow drift compensation system and method, which are used for measuring the angular velocity and the phase of an optical fiber gyroscope.

Background

The fiber optic gyroscope has important significance in the field of precision measurement, and has wide application prospect in the fields of gravitational wave detection, distributed sensing, angular velocity measurement, space navigation, temperature sensing, deformation detection and the like. The laser means can overcome the defects of the traditional mechanical gyroscopes, suspension gyroscopes and piezoelectric gyroscopes. In addition, due to the maturity of the optical fiber micro-nano processing technology, the loop length can be obviously increased by utilizing the multi-turn polarization maintaining optical fiber ring, and the angular velocity measurement precision is further improved. However, due to the introduction of the optical fiber, errors such as optical fiber drift, zero drift, noise and the like are inevitably brought to the gyroscope. Particularly for interferometric fiber-optic gyroscopes, the slow-drift and noise can be derived from a number of factors such as temperature, mechanical disturbances, stress deformation, polarization coupling, brillouin scattering, rayleigh scattering, kerr effects, and the like. The temperature, mechanical disturbance and polarization mode coupling are important sources of noise and also are main sources of slow drift of the fiber-optic gyroscope signal. Under the experimental temperature condition and the platform vibration isolation condition of relatively constant control, the polarization factor is the object to be considered seriously.

The measurement accuracy of angular velocity and the robustness of the gyroscope are key to the spatial navigation and positioning accuracy, and therefore, it is necessary to eliminate unnecessary noise and slow drift as much as possible to improve the angular velocity measurement accuracy and robustness of the optical fiber gyroscope. The conventional methods can be mainly classified into two types: the gyroscope is characterized in that from the light path structure, for example, a minimum reciprocity light path, an integrated Y waveguide or a quadrupole symmetric surrounding structure is adopted to reduce slow drift and noise, so that the accuracy of the gyroscope is improved; yet another method starts from the control algorithm, for example, a proportional-integral-derivative algorithm is adopted to remove loop feedback and phase lock, lock polarization or a kalman filter method is adopted to post-process the acquired data. The traditional proportional-integral-derivative algorithm relies on multiple pre-experiments to fudge the empirical parameters of the controller, and the selection of the parameters has a great influence on the locking effect. The method of data post-processing cannot directly act on the experimental measurement system and cannot rapidly process in real time, so that the applicability and operability of the method in a practical complex and changeable environment are greatly reduced.

Disclosure of Invention

In order to solve the defects in the prior art, the invention aims to provide a novel polarization locking and slow-drift compensation system and a corresponding method for a machine learning-assisted optical fiber gyroscope, and aims to solve the problems of polarization mode coupling, slow-drift of output signals and low measurement sensitivity in the traditional optical fiber gyroscope. The method is inspired by the traditional interferometer phase locking technology, combines machine learning and a proportional-integral-derivative control algorithm, is adaptive to a fiber-optic gyroscope hardware system, can intelligently search the global minimum of an error signal of the fiber-optic gyroscope, and can simultaneously respond quickly in real time to avoid additional interference of an angular velocity signal to be detected. The system of the invention can better solve the influence of polarization factors on fiber optic gyroscope signals on the cross control of programs and hardware. Compared with the traditional fiber optic gyroscope, the invention obviously inhibits slow drift of signals, greatly improves long-time robustness and measurement sensitivity of the system, and improves anti-interference capability in a universal environment. The invention further improves the application value of the optical fiber gyroscope in the precision measurement fields of optical fiber sensing, space navigation, geographical survey and the like.

In order to achieve the above purpose, the present invention designs and implements a polarization locking and slow drift compensation system of a fiber optic gyroscope assisted by machine learning, which is specifically as follows:

the coherent light source is linearly polarized light, is generated by a continuous laser, and can sense the phase difference between a clockwise loop and a counterclockwise loop caused by a sagnac effect after entering an optical fiber loop of the gyroscope;

the gram laser prism has high extinction ratio and is used for further improving the linear polarization purity of an incident coherent light source and ensuring that the extinction ratio of light incident into a loop is larger than 1:10000 linearly polarized light and the polarization direction is adjustable.

The first polarizing beam splitter is used to polarize the linear polarization in the horizontal and vertical polarization directions 50:50 splitting;

the first reflector is used for changing the propagation direction of the light field to enable the light path to coincide with the central line of a subsequent device;

the first Faraday rotator is a non-reciprocal device, and utilizes Faraday rotation effect to enable the polarization direction of forward passing light to rotate 45 degrees, and the polarization direction of reverse passing light rotates 45 degrees along the rotation direction when the forward passing light passes, so that the polarization of forward incident light and reverse emergent light are mutually orthogonal;

the first half-wave plate can be used for adjusting the polarization direction of linearly polarized light, in particular aligning the polarization direction of an optical field with the crystal optical axis of a rear liquid crystal modulator;

The liquid crystal modulator is a core device for hardware compensation feedback, and the inside of the liquid crystal modulator is mainly formed by combining a liquid crystal delay wave plate and a 1/4 wave plate. The slow axis of the two forms an included angle of 45 degrees. In the liquid crystal modulator, liquid crystal molecules can be aligned under the drive of external voltage, so that the refractive index is changed, and the polarization direction of emergent light is rotated. Under the influence of factors such as environmental temperature change, vibration, etc., the polarization maintaining fiber can generate phenomena such as deformation, optical axis distortion, polarization coupling, etc. The liquid crystal modulator is used for correcting the polarization direction of the linear polarization in real time under the control of feedback voltage, so that the linear polarization is aligned with the optical axis of the polarization maintaining fiber for a long time. The specific structure of the liquid crystal modulator is shown in fig. 3. In use, incident light is polarized to align with a calibration axis on the housing of the liquid crystal modulator and is incident on the liquid crystal. A voltage is applied along the fast or slow axis of the liquid crystal to cause a change in refractive index between the fast and slow axes. After passing through the 1/4 wave plate, the polarization angle is rotated.

The laser sequentially passes through the first half-wave plate and the liquid crystal modulator and then enters the polarization maintaining optical fiber.

The polarization-maintaining optical fiber forms a sagnac ring, so that a sagnac effect is generated when linearly polarized incident light propagates in an optical fiber medium, and meanwhile, the polarization-maintaining optical fiber can ensure that linearly polarized light incident along an optical axis can be stable as much as possible, and the polarization of the emitted light can be kept as much as possible;

The phase modulator can change the refractive index difference between a fast axis and a slow axis under the drive of external voltage so as to generate a phase difference, and is used for phase shifting of interference signals of the optical fiber gyroscope system;

a second Faraday rotator for making the polarization directions of the forward incident light and the reverse emergent light mutually perpendicular, similar to the first Faraday rotator;

the second half wave plate is used for adjusting the polarization direction of the linear polarization and is primarily aligned with the optical axis of the polarization maintaining optical fiber;

and a head-tail coupling interface of the polarization maintaining optical fiber is arranged between the liquid crystal regulator and the polarization maintaining optical fiber respectively between the second half wave plate and the polarization maintaining optical fiber. The coupling interface plays a role of converging and receiving incident light rays, so that energy of laser is utilized as much as possible and transmitted along the optical fiber.

The second reflector can change the propagation direction of light and enable the light to be collimated and incident on a subsequent device;

the third half wave plate is used for changing the polarization state of light emitted from the optical fiber loop, so that the polarization directions of light emitted clockwise and light emitted anticlockwise coincide, and the interference condition is met;

the second polarization beam splitter is matched with the front third half-wave plate, so that the light intensity transmitted by the second polarization beam splitter reaches the maximum, which is equivalent to a polarization analyzer;

The photoelectric detector comprises a photodiode and is used for converting an optical signal into an electric signal through a photoelectric effect;

the computer is used for running a photoelectric signal acquisition program and a machine learning optimization program, controlling data transmission and playing a role in communication with the hardware system;

the signal generator is an arbitrary waveform pulse generator, and can be controlled by a computer to combine input pulse digital signals into analog electric signals and transmit the analog electric signals to the liquid crystal modulator through BNC cables.

The key points are as follows:

the coherent light source is produced by a continuous laser with a linewidth of 1.5kHz, an extremely stable optical power, and linearly polarized light.

The first polarizing beam splitter described above directs incident light at a ratio of 50:50 beam splitting is carried out, and the polarization of the two beams of light after beam splitting is perpendicular to each other.

The liquid crystal modulator receives the electric signal from the signal generator and responds to the light path, so that the polarization of the light field is quickly corrected in real time, and the output signal slow drift caused by polarization factors is restrained. And the amplitude of the received electric signal is set within a limited range according to the response limit of the liquid crystal modulation device. The operating voltage of the liquid crystal modulator has a defined interval of about 0-5V. In the range of 0-5V, the response of the liquid crystal modulator is relatively sensitive. Beyond a defined interval, the modulation response of the liquid crystal is very slow and there is a risk of breakdown by the voltage, which affects the lifetime. In order to respond to feedback quickly, the amplitude of the operating voltage of the liquid crystal modulator must be set within a defined interval in the experiment.

The clockwise incident light and the anticlockwise incident light can sense the phase difference caused by the rotation of the external angular velocity when passing through the polarization maintaining optical fiber, and the angular velocity data of the optical fiber gyroscope can be obtained through conversion according to a formula of the sagnac effect.

The phase modulator can change the relative phase difference of clockwise emergent light and anticlockwise emergent light in the optical fiber gyroscope under the power supply of a high-voltage power supply, and fix the working point of the optical fiber gyroscope near pi/2 phase point, namely near the balance position of an interference signal.

The computer acquires the signals detected by the photoelectric detector, transmits the signals to an optimizing program of machine learning, calculates the signals through a proportional-integral-derivative operation module and outputs the signals to the signal generator. The machine learning program and the proportional-integral-derivative module are mixed and circulated, and the difference value between the gyroscope interference output signal and the set value is used as an optimization criterion. The feedback system can stably operate for a long time.

As shown in fig. 2, the portion within the right-hand box is the specific loop of the hybrid algorithm. When the target value (the difference between the gyroscope interference output signal and the set value) acquired by the experimental system is larger than the set threshold value, the gyroscope is indicated to have deviated from the global optimal value of the working operation, and the neural network algorithm part (the machine learning program) in the hybrid algorithm starts to search rapidly within a given range, so that the target value is made to approach the optimal working point (the global optimal value) again. When the optimal target value is found (the ending condition of the neural network algorithm is met), the proportional integral derivative PID part in the hybrid algorithm works, and the target quantity acquired by the experimental device is subjected to real-time operation to continuously correct the parameters applied to the liquid crystal modulator; generally, the number of training samples collected from the initial several cycles is insufficient to allow the neural network algorithm to search for the optimal target value, and the old parameters are kept, and new target values and parameters are obtained from the experimental system to add training samples. Generally, after about 20 cycles, the search result of the neural network begins to gradually converge, and finally the search is ended and the optimal value is predicted. After this, the algorithm is not turned off and it is always monitored whether the target value (the difference between the gyroscope interference output signal and the set value) is greater than the set threshold. It can be seen from fig. 2 that the pid module will always be in real time based on the corrected control parameters in a small range when the neural network is not operating. Thus, the experimental system is locked around the optimal operating point for a long time, i.e. the point where the polarization coupling error of the experimental system is minimal. This is a specific process of the mixing cycle.

The specific function of the proportional-integral-derivative algorithm is that the optical fiber gyroscope is predicted by the neural network algorithmThe method is used for correcting the polarization error and the intensity drift of the incident light in a long time process by continuing the rapid operation in a small range near the polarization locking optimal point, and is used for overcoming the defects of low feedback speed and long training time of a single neural network algorithm. Which is integrated with the neural network algorithm in the python code, is part of the procedure in the present invention. As shown in fig. 2, in the present invention, the classification conditions are set: (I) _s -Q)/I _s >1%. Wherein the gyroscope interferes with the output signal I _s Difference I from the reference light intensity Q _s -Q. When (I) _s -Q)/I _s If the value is more than 1%, the neural network algorithm starts to operate, and an optimal value is found according to the loss function criterion, when (I _s -Q)/I _s And when the ratio is less than 1%, the polarization locking is directly carried out through a proportional-integral-derivative algorithm. In the present invention, the loss function is set to bet ₀ Is the starting time t of the neural network ₁ By the current time, I _s And (t) represents the gyroscope interference output signal at the time t. When the loss function starts to continuously converge and reaches a stable minimum value, the neural network algorithm ends the search, and can predict the optimal polarization locking point of the experimental system and transmit the optimal parameters to the proportional-integral-derivative algorithm. The latter will perform a small range of fast polarization locking around the optimal parameters. The end condition in fig. 2: the loss function L reaches a stable minimum (no smaller value of L can be found for 20 consecutive iterations). Thanks to the setting of the experimental hardware system in the invention, the time from the acquisition of one data to the algorithm operation, the judgment of the correction parameters, the response of the experimental system and the new acquisition … … round of complete iteration process is only about 0.1 seconds.

The machine learning procedure used in the present invention is the feed forward neural network FNN in the Tensorflow open source package of python. The hybrid machine learning algorithm is based on the FNN neural network, an initial parameter set is generated by using a built-in differential evolution algorithm, a mean square error loss function is used as an optimization criterion, and the hybrid machine learning algorithm is formed by combining a proportional-integral-derivative PID program according to the process shown in figure 2. Such a kind ofThe unique hybrid structure plays a central role in the actual polarization locking process of the fiber optic gyroscope. The single neural network algorithm has practical application in the physical front fields of NV spin reading, light pulse prediction, topology invariant searching, quantum gas preparation, multi-parameter estimation and the like ^[1-6] More particularly, the technology has been developed and matured in the fields of biology, chemistry and computer science. At the same time, some hybrid algorithms also begin to appear in succession ^[7-9] . The mixing algorithm in the invention adopts an MLOOP open source program package ^[10] Is provided. The hybrid framework is used in the fields of cold atom BEC experiments, magnetometer coil design, quantum state auxiliary identification and the like ^[10-12] In the aspect of locking the fiber-optic gyroscope, the use of a similar frame is the first time. The invention designs the optical fiber gyroscope polarization locking experimental system shown in figure 1 independently, and designs a hybrid algorithm program according to specific experimental parameters and optimization targets on the basis of a frame interface of MLOOP. A corresponding algorithm flow diagram is shown in fig. 2.

The invention also provides a fiber optic gyroscope polarization locking and slow drift compensation method based on machine learning, and the fiber optic gyroscope polarization locking and slow drift compensation system is utilized. In the method, a beam of linearly polarized laser with stable power is prepared according to the following ratio of 50:50 beams enter a loop of the optical fiber gyroscope in a splitting way, and interference occurs after the beams exit. Then, the phase modulator shifts the phase to enable the working point of the gyroscope to move near pi/2 phase point, and then the signal measured by the photoelectric detector is converted, so that the angular velocity information loaded on the optical fiber loop can be obtained. The locking method based on machine learning is to input the error value between the signal measured by the photoelectric detector and the set value into a neural network, search and predict the error value, and rapidly feed back the error value to the signal generator and the liquid crystal modulator after the operation of a proportional-integral-derivative algorithm to complete one cycle and continue the next cycle. And then cycles through the gyroscope during operation. The method is not only suitable for locking the polarization of the incident light of the optical fiber gyroscope, but also can be used for loop loss compensation and phase quick locking of the optical fiber gyroscope, has good expansibility, and only needs to change the optimization target amount of machine learning.

The hybrid machine learning method adopted by the invention is combined with the fiber-optic gyroscope experimental system, and the polarization of the light field in the fiber-optic gyroscope is corrected in real time according to the continuous circulation of the sequence of parameter initialization, data acquisition of the fiber-optic gyroscope, error signal extraction, neural network training, operation of a proportional-integral-differential device, feedback voltage transmission and response of a liquid crystal modulator, so that the long-time slow drift of signals caused by polarization coupling in a fiber-optic loop is reduced, and the long-term stability of the fiber-optic gyroscope near a working point is improved.

The involvement of the sub-algorithms is not an innovation of the present invention. The invention only integrates and uses the neural network algorithm and the traditional proportional-integral-derivative method, and specifically realizes and solves the problems of polarization locking and slow drift compensation of the fiber optic gyroscope by using a hybrid algorithm. The hybrid algorithm is different from the existing algorithm: the combined use of the neural network algorithm and the traditional proportional-integral-derivative algorithm is realized, and the defect of a single algorithm is overcome. The single neural network algorithm needs a certain sample as a training set, and can not quickly compensate and correct the result at the next moment according to the data result at the last moment like a proportional-integral-derivative algorithm; whereas a single conventional pid algorithm relies on multiple pre-experiments to obtain empirical parameters for stable operating points of the fiber optic gyroscope. If the initial parameters are not good, the locking effect of the algorithm is greatly compromised. As shown in fig. 2, the mixed algorithm is combined with the experimental system shown in fig. 1, so that the polarization locking of the optical fiber gyroscope is realized, and the long-term stability and the measurement sensitivity of the optical fiber gyroscope are improved. The invention does not simply splice the series connection when implementing the hybrid algorithm. The neural network algorithm and the proportional-integral-derivative method are circulated in parallel to meet the actual experimental requirements of the fiber-optic gyroscope. The difficulties are mainly two: first, it is necessary to select an appropriate classification condition. The classification conditions determine the actual experimental effect of the neural network algorithm and the proportional-integral-derivative method. Too wide a classification condition setting may make the search training time of the neural network algorithm too long, so that the experiment cannot realize real-time locking. The classification conditions are too accurate and may lead to a situation once exposed to the outside world After the sudden disturbance jumps out of the equilibrium position locking point, the feedback of the hybrid algorithm cannot pull the working point of the gyroscope back to the locking position. Thus, the classification condition (I _s -Q)/I _s The more than 1% is reasonably set according to the actual polarization locking effect of the experimental device in the invention, so that the training time of the neural network can be shortened, the interference of the environment can be resisted, and the robustness of the fiber-optic gyroscope in the operation near the working point can be improved. And secondly, the neural network algorithm needs to give the optimized optimal parameters to the PID algorithm for operation. In order to ensure that the optimal parameters can reflect the optimal position of the experimental system, parameters which meet experimental conditions and are used for optimization need to be selected according to experimental actual conditions. For example, there is a mapping relationship between the output light intensity of the fiber optic gyroscope and the voltage of the liquid crystal modulator. The voltage, frequency of the liquid crystal modulator is chosen as the optimization parameter to ensure this.

The specific method comprises the following steps:

step one: operating a laser, adjusting a light path, scanning a phase modulator to stabilize an interference signal of the optical fiber gyroscope, acquiring the stabilized interference signal from a photoelectric detector, closing a scanning voltage, and driving the phase modulator by using a constant direct current voltage to fix the phase modulator at a working phase point of the gyroscope;

Step two: machine learning parameters are initialized. Setting the amplitude value and the frequency of the driving voltage of the liquid crystal modulator as optimized parameters, setting the maximum iteration times and parameter boundary values, generating initial training set data by adopting a differential evolution algorithm according to the response curve of the liquid crystal modulator, responding by a hardware system every time the parameters are generated, and starting to acquire the detection data of the photoelectric detector at the next moment.

Step three: on the basis of an initial training set, a neural network algorithm in machine learning starts to predict and construct the relation between parameters (parameters such as voltage amplitude value and frequency of a liquid crystal modulator) and an optimization target (the difference between a gyroscope interference output signal and a set value), and the convergence process is accelerated through a built-in quasi-Newton iterative algorithm such as L-BFGS, and the optimal value and the corresponding parameters are obtained through prediction. The corresponding parameters are rapidly calculated by the proportional integral derivative module and then transmitted to the liquid crystal modulator through the signal generator, and the previous parameters are corrected.

Step four: the neural network does not end the search process after predicting the optimal parameters. When the system meets the corresponding classification condition, the traditional proportional-integral-derivative algorithm compensates and locks polarization according to the optimal parameters, and the system and the hardware system continuously repeat the acquisition and feedback processes, and the continuous loop … … can still acquire the data of the optical fiber gyroscope acquired by the photoelectric detector during the period. The algorithm judges whether the interference signal has greatly deviated from the optimal value of the previous optical fiber polarization locking under the interference of the factors such as external environment temperature, vibration, stress torsion and the like, and determines whether a new round of machine learning search and iteration are needed; if the optimum value of the previous fiber polarization lock is greatly deviated, a new round of machine learning search and iteration is performed.

The machine learning neural network algorithm still uses the architecture of the feedforward neural network algorithm, but only fuses the PID proportional integral derivative algorithm into the iterative process of the neural network search according to the actual long-time continuous polarization locking requirement of the fiber optic gyroscope, and performs iterative loop (the prior mentioned). The hybrid approach has better polarization locking and slow-drift suppression than the single algorithm. When the fiber optic gyroscope measures a small angular velocity, long-time accumulation and Allan variance noise analysis are needed to be carried out on the data so as to improve the angular velocity measurement accuracy. Therefore, the better the polarization locking effect of the fiber-optic gyroscope is, the better the stability of the fiber-optic gyroscope is over a long time, and the sensitivity of the fiber-optic gyroscope is improved.

Step five: and converting the interference data of the optical fiber gyroscope acquired in continuous time into phase variation, and converting the phase variation into angular velocity variation according to a corresponding formula of the sagnac effect. And the Allan variance is used for carrying out the noise analysis at intervals, so that the working indexes such as corresponding angle random walk coefficient, zero offset stability, quantization noise and the like of the gyroscope can be obtained. And comparing the index with the index of the optical fiber gyroscope locked by the traditional proportional-integral-derivative method and the index of the optical fiber gyroscope unlocked.

The Sagnac effect is a phenomenon that describes the phase difference that occurs when light propagates along a closed track (referred to herein as a fiber loop) on a rotating platform, and can be expressed as:

wherein, the liquid crystal display device comprises a liquid crystal display device,is the phase difference of light propagation, a is the area enclosed by the optical fiber loop, Ω is the angular velocity of the rotating platform, λ is the wavelength of the light, and c is the retransmission speed of the light in vacuum.

The Allan variance (Allan variance) is a statistical method commonly used to evaluate frequency stability. The method can obtain a numerical index describing frequency stability by sampling and processing signals in a time domain.

The calculation formula of the Allan variance is as follows:

σ(τ)2＝1/(2(N-1))*∑[y(n+1)-y(n)-<y>]2

where σ (τ) 2 is the Allan variance, τ is the sampling interval, N is the total number of sampling points, y (N) is the value of the nth sampling point, and < y > is the average of the entire signal.

Compared with the prior art, the invention has the following beneficial effects:

the invention uses machine learning, proportional integral derivative and other operation methods to calibrate polarization rapidly, so that noise and slow drift of interference output signals of the optical fiber gyroscope are obviously reduced compared with those before unlocking. In the experiment, the loop length of the polarization maintaining fiber of the fiber-optic gyroscope is 300m, and the emergent light power is 40 mu W. The intensity of the output signal is reduced from + -2.4 mu W to + -0.3 mu W by the peak-to-peak value of the intensity of the output signal before polarization locking.

By introducing a machine learning technology, the searching capability and the environment interference resistance capability of the optical fiber gyroscope are greatly enhanced, and the optimal lock can be adaptively searched and retrieved after the optical fiber gyroscope is separated from the original optimal locking position of the gyroscope under the external influenceAnd (5) determining parameters. The robustness of the system is greatly improved; under the condition of long-time continuous acquisition, the adaptive gyroscope locking method is beneficial to optimizing corresponding working indexes such as angle random walk, zero offset stability and the like, and the measuring sensitivity of the system near the working phase point of the optical fiber gyroscope is improved. In the present invention, data collected continuously for 3 hours were analyzed by Allan variance, and the results are shown in FIGS. 4 and 5: wherein ARW represents the angle random walk noise of the fiber optic gyroscope and is also an index reflecting the minimum angular velocity sensitivity thereof. And BI is an abbreviation for zero offset instability. Under the condition of the same optical power and the same optical fiber loop length, the index locked by the traditional PID algorithm is obtained through testing: angle random walk:zero offset instability: 0.133 °/h; and after feedback locking by adopting a hybrid machine learning method: angle random walk: />Zero offset instability: 0.065 DEG/h. By comparison, the zero offset instability is found to be reduced by 0.068 °/h, which reflects a significant decrease in the intensity slow-drift, i.e. a significant increase in the long-term stability of the experimental system. In addition, the angle random walk coefficient is slightly reduced, which reflects that random white noise mixed in the output signal of the optical fiber gyroscope is improved after a hybrid machine learning method is used, so that the sensitivity of the angular velocity component under long-time accumulated measurement of an experimental system is improved >More importantly, the zero offset instability index of the system is reduced by 0.068 degrees/h, which means that the resistance of the optical fiber gyroscope to environmental interference and noise is improved under the continuous long-time operation condition, and the slow drift and zero drift errors in the long-time operation process are reduced. This is obviously of practical significance for gyroscope measurements.

Drawings

FIG. 1 is a block diagram of a machine-learning assisted fiber optic gyroscope polarization locking and slow drift compensation system of the present invention.

FIG. 2 is a schematic diagram of the workflow of the method for polarization locking and slow drift suppression of a fiber optic gyroscope based on machine learning in the present invention.

Fig. 3 is an internal structural diagram of a liquid crystal modulator used in the experimental system of the present invention.

FIG. 4 is a graph of the Allan variance results of a fiber optic gyroscope angular velocity signal locked using a conventional PID algorithm.

FIG. 5 is a graph of the Allan variance results of a fiber optic gyroscope angular velocity signal locked using a hybrid machine learning method in accordance with the present invention.

FIG. 6 is a graph of the interference signal output by the fiber-optic gyroscope system of the present invention at 200Hz alternating angular rotation.

Detailed Description

The invention will be described in further detail with reference to the following specific examples and drawings. The procedures, conditions, experimental methods, etc. for carrying out the present invention are common knowledge and common knowledge in the art, except for the following specific references, and the present invention is not particularly limited.

The invention provides a machine learning-assisted optical fiber gyroscope polarization locking and slow drift compensation system. The system can intelligently search and predict the relation between the interference error signal of the optical fiber gyroscope and the polarization direction of the incident light by introducing a neural network in machine learning, and feeds the predicted optimal value back to a hardware system for compensation, thereby realizing the functions of locking the optical fiber polarization and inhibiting the intensity drift of an output signal. Meanwhile, the system organically combines a machine learning algorithm with a traditional proportional-integral-derivative feedback method, has the advantages of the two methods, can adaptively search the global minimum of the error signal of the optical fiber gyroscope, and simultaneously can respond quickly in real time to avoid additional interference to the angular velocity signal to be detected. Compared with the traditional fiber optic gyroscope, the invention obviously inhibits slow drift of signals, thereby improving long-time robustness and measurement sensitivity of the system. The specific experimental device comprises a coherent light source, a gram laser prism, a first polarization beam splitter, a first reflecting mirror, a first Faraday rotator, a first half wave plate, a liquid crystal modulator, a polarization maintaining optical fiber, a phase modulator, a second Faraday rotator, a second half wave plate, a second reflecting mirror, a third half wave plate, a second polarization beam splitter, a photoelectric detector, a computer and a signal generator. The invention has good application background in the aspects of angular velocity measurement, temperature sensing, space navigation, distributed sensing, deformation detection and the like, and has important significance in the fields of precision measurement and national defense science and technology.

Referring to fig. 1, the structure of the polarization locking and slow drift compensation system of the machine learning-assisted optical fiber gyroscope according to the invention comprises: a coherent light source 1, a gram laser prism 2, a first polarization beam splitter 3, a first reflecting mirror 4, a first Faraday rotator 5, a first half-wave plate 6, a liquid crystal modulator 7, a polarization maintaining optical fiber 8, a phase modulator 9, a second Faraday rotator 10, a second half-wave plate 11, a second reflecting mirror 12, a third half-wave plate 13, a second polarization beam splitter 14, a photoelectric detector 15, a computer 16 and a signal generator 17;

the coherent light source 1 is linear polarized light with stable power, and the linear polarized light purity is further purified after being filtered by the gram laser prism 2. The incident light is split into a beam of light traveling clockwise around the loop and a beam of light traveling counterclockwise around the loop in equal proportion after passing through the first polarizing beam splitter 3. In the case where the optical fiber loop does not rotate, since the two light beams travel the same distance along the same optical axis in the optical fiber loop, the phase difference is constant. However, when an angular velocity is applied to the fiber loop, the phase difference between the two beams changes. Angular velocity information can be converted from the change in the interference signal. The first faraday rotator 5 and the second faraday rotator 10 are used to ensure that two beams of light are combined and emitted at the port of the first polarization beam splitter 3. After the two beams are combined, the interference can be generated by ensuring that the polarization is the same through the third wave plate 13 and the second polarization beam splitter 14. When the interference signal is observed from the photodetector 15, the phase difference between the two beams needs to be shifted by the phase modulator 9, so that the fiber-optic gyroscope operates near pi/2 phase point to measure the interference signal. And converting the error amount of the interference signal at different moments into phase error, and finally converting into angular velocity information. Dry optical fiber gyroscope caused by sagnac effect The wading signal is changed into By means of the inverse trigonometric function +.>Wherein (1)>Is the phase difference of light propagation, I ₀ Is the amplitude of the interference signal, a is the area enclosed by the fiber loop, Ω is the angular velocity of the rotating platform, λ is the wavelength of the light, and c is the retransmission speed of the light in vacuum. Therefore, the magnitude of the angular velocity can be extracted from the interference signal by the above formula.

When the system is in an operation state, if the error (variance) of the angular velocity deviates from a set threshold value, the machine learning algorithm searches and predicts the optimal parameter in a response range by taking the voltage amplitude and the frequency of the liquid crystal modulator as parameters, and then transmits the parameter to the signal generator 17 after being processed by the proportional-integral-derivative algorithm. The signal generator 17 is an arbitrary waveform pulse generator, and can reconstruct digital parameters into analog electrical signals to drive the liquid crystal modulator 7 for polarization angle compensation feedback.

The proportional-integral-derivative algorithm (PID algorithm) is a commonly used control algorithm for implementing automatic control. The PID algorithm is based on a feedback control principle, and the output signal of the controlled object gradually tends to the target value by continuously adjusting the output signal of the controller.

The PID algorithm consists of three parts, a Proportional (Proportional), integral (Integral), and Derivative (Derivative) controller, respectively. The proportion controller calculates an output signal according to the difference between the current output value and the set value of the controlled object; the integral controller calculates an output signal according to the difference between the historical output value of the controlled object and the set value; the differential controller calculates an output signal according to a difference between a current output value of the controlled object and an output value at a previous time. The output signals of the three controllers are weighted and summed to obtain the final controller output signal.

The output signal of the proportional controller is proportional to the deviation of the controlled object, i.e. the larger the deviation is, the larger the output signal is. The output signal of the integral controller is in direct proportion to the sum of the history deviation of the controlled object, namely, the larger the sum of the history deviation is, the larger the output signal is. The output signal of the differential controller is in direct proportion to the difference between the current deviation of the controlled object and the deviation at the previous moment, namely, the larger the deviation change rate is, the larger the output signal is. By comprehensively considering the output signals of the three controllers, the PID algorithm can realize the rapid and accurate adjustment of the output value of the controlled object, thereby realizing the purpose of automatic control.

In the experimental system, the controlled object is a fiber optic gyroscope hardware system. The current output value of the controlled object is the interference signal intensity value of the optical fiber gyroscope. The set value is a signal intensity value of the optical fiber gyroscope in the vicinity of the interference balance position (static operating point). The output signal of the controller obtained by the PID algorithm after calculation and weighting is the corrected value of the voltage and frequency parameters applied to the liquid crystal modulator.

The complete flow of the polarization locking and slow drift compensation method of the optical fiber gyroscope based on machine learning is shown in figure 2, and firstly, an optical path is adjusted and the stationary phase position of the phase modulator is used. The computer program will then generate initialization parameters within the given parameters by differential evolution algorithm and transmit the initial values to the fiber optic gyroscope hardware system for initialization. The photoelectric signal of the gyroscope can be collected by a computer in the continuous running process under the stationary phase point. The program calculates the deviation between the measured value and the set value and compares the deviation with the set classification threshold. When the deviation is smaller than the threshold value, the deviation is directly transmitted to the traditional proportional-integral-derivative algorithm for processing and is rapidly fed back to the signal generator and the liquid crystal modulation device, the polarization angle of the incident light before the polarization maintaining fiber port is compensated and locked in real time, so that the polarization direction is corrected, and the next acquisition and feedback cycle is continued; when the deviation is larger than the threshold value, the polarization direction of the incident light field is completely deviated from the transmission optical axis of the polarization maintaining optical fiber, and obvious noise and slow-drift of intensity corresponding to polarization coupling can be caused. At this time, the deviation data is transmitted into the neural network. The neural network searches and predicts parameters such as voltage amplitude value, frequency, duty ratio and the like of the adjustable liquid crystal modulator, gives the optimized predicted optimal parameter and optimal value to a proportional-integral-differential algorithm to be rapidly processed, feeds back the optimal parameter and the optimal value to a hardware system such as the liquid crystal modulator and the like to dynamically compensate to complete a complete cycle, and continues the next cycle until the set end condition is reached. The one-time cycle time can be lower than 0.1 second, so that the method can quickly compensate and inhibit the slow drift error, thereby improving the measurement sensitivity. For some high-frequency errors and noise, the method for polarization locking and slow drift compensation of the fiber optic gyroscope based on machine learning is still applicable, but hardware with shorter response time, such as an electro-optic modulation crystal, is replaced. In a word, the machine learning-based method can remarkably improve the stability and measurement sensitivity of the optical fiber gyroscope in a long-time continuous operation process, has universality developed in a complex environment, and can be beneficial to improving the application value of the optical fiber gyroscope in an actual measurement environment.

Differential evolution (Differential Evolution, DE for short) is an optimization method based on evolutionary algorithm, which is used for solving the optimal solution of the objective function.

The differential evolution algorithm generates a group of initial populations randomly, and continuously and iteratively searches for a better individual solution to finally obtain a global optimal solution or a local optimal solution of the objective function. The main idea is to generate new individual solutions through two operations of selection and mutation, compare and screen the new individual solutions with the existing individual solutions, and further find out better solutions.

In the differential evolution algorithm, each individual solution is represented by a vector, and each dimension represents an independent variable. In each iteration, the differential evolution algorithm generates a new individual solution by randomly selecting three different individual solutions, linearly combining them, and comparing the new individual solution with the current individual solution to select a better individual solution. This process is called a mutation operation. Then, the new individual solution is subjected to differential operation with the current individual solution to obtain a new individual solution, the new individual solution is compared with the current individual solution, a better individual solution is selected, and the population is updated. This process is referred to as a select operation.

The differential evolution algorithm used in the hybrid machine learning method is mainly used for generating initial parameters for the liquid crystal modulator of the experimental system, and the initial parameters and corresponding experimental data can be used as an initial training set of the neural network algorithm. The differential evolution algorithm is built into the neural network program and can be obtained in the open source TensorFlow package of python.

The neural network in the hybrid machine learning method adopts a FNN feedforward neural network which iterates parameters according to the principle of back propagation BP and takes a mean square error loss function as an optimization criterion. According to the actual operation condition of the optical fiber gyroscope, specific ending conditions are set as follows: no smaller value of L could be found for the 20 consecutive iterations. The training set data of the neural network are provided by the output signals of the fiber-optic gyroscope and the voltage parameters of the liquid crystal modulator, which are acquired at different moments. The initial training set is provided by the differential evolution algorithm described above.

The core part of the machine learning hybrid algorithm adopted by the invention is provided by the following python calculation package:

M-LOOP is a machine learning online optimization package, is an open source code package of Git-hub, and provides a basic framework and interface for combining a neural network algorithm with a traditional proportional-integral-derivative algorithm.

TensorFlow: the machine learning package of google is used for constructing a neural network part of the hybrid algorithm in the invention;

numpy, a scientific calculation package of python, is used for carrying out matrix operation on data collected by a hardware system.

Examples

According to the optical path structure shown in fig. 1, the method in the present embodiment is implemented in a fiber optic gyroscope. In experiments, the fiber optic gyroscope was first placed in an experimental environment at a temperature of about 20 ℃. The entire experimental light path is fixed on an optical platform. As shown in FIG. 6, the interference signal output by the fiber optic gyroscope when the 300 m polarization maintaining fiber optic ring rotates at an externally applied alternating angular velocity of 200Hz was tested in this example. The contrast of the interference pattern reaches over 96.0 percent.

The rotation of the fiber optic ring is then stopped and secured to the platform. In this embodiment, the phase operating point of the fiber optic gyroscope is fixed at the equilibrium position of the interference signal by applying 182.3V dc steady voltage to the phase modulator, as shown by the broken line in fig. 6, and the intensity is hardly changed with time. Near this phase point, the fiber loop is most sensitive to small angular velocity sensing. Therefore, the present embodiment selects the fiber optic gyroscope operating point at the equilibrium location of the interference signal. In order to avoid interference caused by power change of laser, the emergent light intensity of the optical fiber gyroscope is controlled to be 40 microwatts in the whole experimental process. However, due to unavoidable effects of fiber polarization coupling and environmental temperature changes, the actual output light intensity of the fiber-optic gyroscope can slowly drift around 40 microwatts. Therefore, the fiber optic gyroscope must be locked when it actually measures a minute angular velocity (e.g., the earth's self-rotation angular velocity). In this embodiment, a hybrid machine learning algorithm is used to polarization lock the experimental system. During the locking process, the voltage parameters and frequency parameters fed back to the liquid crystal modulator shown in fig. 3 are set at 0-5V and 1-10kHz. The intensity of the output signal is reduced from + -2.4 mu W to + -0.3 mu W by the peak-to-peak value of the intensity of the output signal before polarization locking.

Comparative examples

And (3) performing polarization locking near the optimal working point of the optical fiber gyroscope by using the conventional PID locking method and the hybrid machine learning method, so that the signal intensity value of the optical fiber gyroscope is locked at 40 microwatts for a long time, and the slow drift of the signal is compensated. The principle of polarization locking has been described in detail in the foregoing.

As shown in fig. 4 and fig. 5, polarization locking was performed by using the conventional PID locking method and the hybrid machine learning method of the present invention, respectively, and the continuously locked gyro data for 3 hours was subjected to an alan analysis and compared. Under the condition of equal optical power and equal optical fiber loop length, the locked traditional PID algorithm is obtained through testingThe index is as follows: angle random walk:zero offset instability: 0.133 °/h; and after locking by adopting a hybrid machine learning method: angle random walk:zero offset instability: 0.065 DEG/h. It can be found that: the zero offset instability is reduced by 0.068 degrees/h, and the sensitivity of the angular velocity component under long-time accumulated measurement of an experimental system is improved>Therefore, the slow drift compensation method and the experimental system based on machine learning can effectively help some bottlenecks of the fiber optic gyroscope in the prior art, so that the long-term stability and the angular velocity sensitivity of the fiber optic gyroscope in practical application are improved.

Reference is made to:

1.Qian P,Lin X,Zhou F,et al.Machine-learning-assisted electron-spin readout of nitrogen-vacancy center in diamond[J].Applied Physics Letters,2021,118(8):084001.

2.Lohani S,Knutson E M,Zhang W,et al.Dispersion characterization and pulse prediction with machine learning[J].OSAContinuum,2019,2(12):3438-3445.

3.Zhang P,Shen H,Zhai H.Machine learning topological invariants with neural networks[J].Physical review letters,2018,120(6):066401.

4.Barker A J,Style H,Luksch K,et al.Applying machine learning optimization methods to the production of a quantum gas[J].Machine Learning:Science andTechnology,2020,1(1):015007.

5.Tranter A D,Slatyer H J,Hush M R,et al.Multiparameter optimisation of a magneto-optical trap using deep learning[J].Nature communications,2018,9(1):4360.

6.Carleo G,Cirac I,Cranmer K,et al.Machine learning andthe physical sciences[J].Reviews of Modern Physics,2019,91(4):045002.

7.Psichogios D C,Ungar L H.A hybrid neural network-first principles approach to process modeling[J].AIChE Journal,1992,38(10):1499-1511.

8.Shon T,Moon J.A hybrid machine learning approach to network anomaly detection[J].Information Sciences,2007,177(18):3799-3821.

9.O’Driscoll L,Nichols R,Knott P A.A hybrid machine learning algorithm for designing quantum experiments[J].Quantum Machine Intelligence,2019,1:5-15.

10.Wigley P B,Everitt P J,van den Hengel A,et al.Fast machine-learning online optimization of ultra-cold-atom experiments[J].Scientific reports,2016,6(1):25890.

11.O’Driscoll L,Nichols R,Knott P A.A hybrid machine learning algorithm for designing quantum experiments[J].Quantum Machine Intelligence,2019,1:5-15.

12.Chen J,Wu Z,Bao G,et al.Design of coaxial coils using hybrid machine learning[J].Review of Scientific Instruments,2021,92(4):045103.

the protection of the present invention is not limited to the above embodiments. Variations and advantages that would occur to one skilled in the art are included in the invention without departing from the spirit and scope of the inventive concept, and the scope of the invention is defined by the appended claims.

Claims

1. A machine-learning assisted fiber optic gyroscope polarization locking and slow drift compensation system, the system comprising: the device comprises a coherent light source (1), a gram laser prism (2), a first polarization beam splitter (3), a first reflecting mirror (4), a first Faraday rotator (5), a first half-wave plate (6), a liquid crystal modulator (7), a polarization maintaining fiber (8), a phase modulator (9), a second Faraday rotator (10), a second half-wave plate (11), a second reflecting mirror (12), a third half-wave plate (13), a second polarization beam splitter (14), a photoelectric detector (15), a computer (16) and a signal generator (17); wherein:

the coherent light source (1) is linearly polarized light generated by a continuous laser and is used for being injected into an interference loop of a gyroscope to sense the phase difference between a clockwise loop and a counterclockwise loop caused by a sagnac effect;

the gram laser prism (2) has high extinction ratio, is used for improving the linear polarization purity of the incident laser and ensuring that the polarization direction of the incident laser is adjustable; the high extinction ratio means that the extinction ratio is greater than 1:10000;

The first polarizing beam splitter (3) is configured to polarize linear polarization in a horizontal polarization direction and a vertical polarization direction 50:50 splitting;

the first reflecting mirror (4) is used for changing the propagation direction of the light field so that the light path coincides with the central line of the subsequent device;

the first Faraday rotator (5) is a non-reciprocal device and is used for changing the polarization directions of forward and backward passing light and finally enabling the polarization directions of the forward incident light and the backward emergent light to be mutually orthogonal;

the first half-wave plate (6) is used for adjusting the polarization direction of linearly polarized light;

the liquid crystal modulator (7) is internally formed by combining a liquid crystal retardation wave plate and a 1/4 wave plate, and the slow axis of the liquid crystal retardation wave plate and the slow axis of the 1/4 wave plate form an included angle of 45 degrees;

the polarization maintaining optical fiber (8) forms a sagnac loop, so that a sagnac effect is generated when linearly polarized incident light propagates in an optical fiber medium, the linearly polarized light incident along an optical axis is ensured to be stable, and the polarization of the emergent light is kept to be original state;

the phase modulator (9) can change the refractive index difference between the fast axis and the slow axis under the drive of external voltage so as to generate a phase difference, and is used for phase shifting of interference signals of the optical fiber gyroscope system;

the second Faraday rotator (10) makes the polarization directions of the forward incident light and the reverse emergent light mutually perpendicular;

The second half-wave plate (11) is used for adjusting the polarization direction of the linear polarization and is primarily aligned with the optical axis of the polarization maintaining optical fiber;

the second reflecting mirror (12) can change the propagation direction of light and enable the light to be collimated and incident on a subsequent device;

the third half-wave plate (13) is used for changing the polarization state of light emitted from the optical fiber loop, so that the polarization directions of light emitted clockwise and light emitted anticlockwise coincide, and the interference condition is met;

the second polarization beam splitter (14) is used as an analyzer and is matched with the third half-wave plate (13) so as to maximize the light intensity transmitted from the second polarization beam splitter (14);

the photodetector (15) comprises a photodiode which converts an optical signal into an electrical signal by utilizing a photoelectric effect;

the computer (16) is used for running an acquisition program of photoelectric signals and program codes of machine learning and controlling the transmission of data;

the signal generator (17) is an arbitrary waveform pulse generator, and is controlled by a computer to combine the input pulse digital signals into analog electric signals and transmit the analog electric signals to the liquid crystal modulator (7) through BNC cables.

2. The system according to claim 1, characterized in that the coherent light source (1) has a linewidth of 1.5kHz and a stable optical power;

The first polarizing beam splitter (3) directs incident light at a ratio of 50:50 beam splitting is carried out, and the polarization of the two beams of light after beam splitting is perpendicular to each other.

3. A system as claimed in claim 1, characterized in that the liquid crystal modulator (7) receives the electrical signal from the signal generator (17) and is responsive to the light path such that the polarization of the light field is corrected in rapid real time, the output signal drift caused by polarization factors is suppressed, and the amplitude of the electrical signal received must be controlled within the voltage range of the liquid crystal modulator.

4. The system according to claim 1, wherein the phase difference caused by the rotation of the external angular velocity can be perceived when the clockwise incident light and the counterclockwise incident light pass through the polarization maintaining fiber (8), and the angular velocity data of the optical fiber gyroscope is calculated by a formula of the sagnac effect.

5. A system according to claim 1, characterized in that the phase modulator (9) is capable of changing the relative phase difference between the clockwise and counter-clockwise light emitted from the fiber optic gyroscope under the supply of an external voltage source, to fix the operating point of the fiber optic gyroscope near the pi/2 phase point, i.e. near the equilibrium position of the interference signal.

6. The system according to claim 1, wherein the computer (16) is configured to collect signals collected by the photodetector (15), run a machine learning optimization program, and output the signals to the signal generator (17) after calculation by the pid operation module; the machine learning program and the proportional-integral-derivative module are mixed and circulated, and the magnitude of the difference between the gyroscope interference output signal and a set value is used as an optimization criterion; the optimization criterion is a gyroscope interference output signal I _s Whether the difference from the reference light intensity Q is greater than 1%.

7. The system of claim 6, wherein the machine learning program and the pid module are in a mixed cycle, when the difference between the gyroscope interference output signal collected by the experimental system and the set value is greater than a set threshold, the gyroscope is shifted by a global optimum value of working operation, and the machine learning program in the mixed algorithm searches rapidly within a given range to make the target value approach to an optimal working point again; the proportional integral derivative PID part carries out real-time operation on the target quantity acquired by the experimental device so as to continuously correct the parameters applied to the liquid crystal modulator; the experimental system is locked around the optimal operating point for a long time.

8. The polarization locking and slow-drift inhibiting method for the fiber optic gyroscope assisted by machine learning is characterized by comprising the following steps of:

step one: operating a laser, adjusting an optical path, scanning a phase modulator (9) to stabilize an interference signal of the optical fiber gyroscope, acquiring the stabilized interference signal from a photoelectric detector (15), closing a scanning voltage, and driving the phase modulator (9) by using a constant direct current voltage to fix the phase modulator at a working phase point of the gyroscope;

Step two: initializing machine learning parameters, namely setting the amplitude value and the frequency of a driving voltage of a liquid crystal modulator (7) as optimization parameters, setting the maximum iteration times and parameter boundary values, generating initial training set data by adopting a differential evolution algorithm according to a response curve of the liquid crystal modulator, responding by a hardware system every time the initial training set data is generated, and starting to acquire detection data of a photoelectric detector (15) at the next moment;

step three: on the basis of an initial training set, a neural network algorithm in machine learning starts to predict and construct the relation between parameters and an optimization target, and accelerates a convergence process through a built-in quasi-Newton iterative algorithm, so as to predict and obtain an optimal value and an optimal parameter; the optimal parameters are rapidly calculated by a proportional-integral-derivative module and then are transmitted to a liquid crystal modulator (7) through the signal generator (17);

step four: after the machine learning predicts the optimal parameters, continuously acquiring the data of the optical fiber gyroscope acquired by the photoelectric detector (15), judging whether the interference signal is greatly deviated from the previous optimal value, and determining whether a new round of parameter searching and model predicting is needed;

step five: and converting the interference data of the optical fiber gyroscope acquired under continuous time into phase variation, converting into angular velocity variation according to a sagnac effect formula, and carrying out interval noise analysis by an Allan variance to obtain working indexes of the gyroscope, including corresponding angle random walk coefficients, zero offset stability and quantization noise.

9. The method of claim 8, wherein the Sagnac effect formulation is expressed as follows:

wherein, the liquid crystal display device comprises a liquid crystal display device,is the phase difference of light propagation, A is the area of the optical fiber loop, Ω is the angular velocity of the rotating platform, λ is the wavelength of light, and c is the velocity of light propagation in vacuum;

the calculation formula of the Allan variance is as follows:

σ(τ)2＝1/(2(N-1))*∑[y(n+1)-y(n)-<y>]2，

where σ (τ) 2 is the Allan variance, τ is the sampling interval, N is the total sampling point number, y (N) is the value of the nth sampling point, < y > is the average value of the entire angular velocity signal.

10. The method of claim 8 or 9, wherein the method is implemented in dependence on a machine-learning-assisted fiber-optic gyroscope polarization locking and slow-drift suppression system of any of claims 1-7.