Two-stage progressive fiber-optic gyroscope temperature compensation method based on weak spatial temperature gradient field

Abstract

The application discloses a two-stage progressive optical fiber gyroscope temperature compensation method based on a weak spatial temperature gradient field, and relates to the technical field of optical fiber gyroscope temperature compensation. The method comprises the following steps: acquiring distributed temperature data in a three-axis integrated optical fiber gyroscope inertial navigation system; temperature sensor arrays are distributed on the surface of a heat generating device and an optical fiber gyroscope package shell according to the rule of covering heat sources and heat conduction paths; according to the distributed temperature data, a static heat-induced drift error prediction value is determined by using an offline trained first-stage model; an online dynamic compensation second-stage model is used to calibrate the static heat-induced drift error prediction value in combination with external reference information, so that error deviation caused by time-varying factors is corrected, and a final error accurate prediction value is obtained. The application can realize high-precision error compensation for the optical fiber gyroscope error in the three-axis integrated optical fiber gyroscope inertial navigation system throughout the whole life cycle.

Description

Technical Field

[0001] This application relates to the field of fiber optic gyroscope temperature compensation technology, and in particular to a two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field. Background Technology

[0002] Based on the different error generation mechanisms of fiber optic gyroscopes in three-axis integrated fiber optic gyroscope inertial navigation systems, error sources can be divided into temperature-dependent errors and time-dependent errors. Temperature-dependent errors are mainly caused by the Shupe effect and are typically compensated for by establishing a mapping model between temperature and drift. Time-dependent errors are time-varying residual errors caused by dynamic temperature changes (such as thermal hysteresis and temperature variation rate) and device aging. Existing temperature compensation methods for three-axis integrated fiber optic gyroscope inertial navigation systems typically employ single-point temperature measurement to establish a static polynomial or neural network model based on a single temperature variable to correct the gyroscope's output. This method can eliminate most linear thermal drift in uniform temperature-changing environments without relying on complex sensing networks.

[0003] However, in miniaturized, highly integrated triaxial fiber optic gyroscope inertial navigation systems, high-power heat sources such as field-programmable gate array (FPGA) chips and light source drivers are integrated internally. Due to the structural thermal conductivity, a non-uniform "weak spatial temperature gradient field" often forms inside the system. Limited by traditional single-point temperature measurement methods, existing approaches cannot perceive this complex spatial thermal distribution information, and therefore cannot correct the non-reciprocal phase shift error caused by the gradient field using a simple linear model. To avoid navigation accuracy degradation due to these unmodeled gradient errors, traditional methods often sacrifice the system's dynamic performance or add complex temperature control devices. Furthermore, when devices experience aging deviations or external shocks due to long-term operation, existing offline static models lack online parameter adjustment capabilities. The residual zero bias not only increases the error amplitude but also causes the compensated output to slowly diverge over time, potentially leading to accuracy failure. Summary of the Invention

[0004] The purpose of this application is to provide a two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field, which can achieve high-precision error compensation for fiber optic gyroscope errors throughout the entire life cycle of a three-axis integrated fiber optic gyroscope inertial navigation system.

[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field, including: Distributed temperature data is acquired within a three-axis integrated fiber optic gyroscope inertial navigation system. This distributed temperature data is transmitted by a temperature sensor array within the three-axis integrated fiber optic gyroscope inertial navigation system. The temperature sensor array is distributed on the surface of the heat-generating device and the fiber optic gyroscope enclosure according to the rules of covering the heat source and the heat conduction path. Based on the distributed temperature data, the predicted value of static thermal drift error is determined using an offline-trained first-level model; Using an online dynamic compensation secondary model and combined with external reference information, the predicted value of the static thermal drift error is calibrated to correct the error offset caused by time-varying factors, thus obtaining the final accurate error prediction value; the final accurate error prediction value is used to compensate the output of the fiber optic gyroscope.

[0006] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a two-stage progressive temperature compensation method for fiber optic gyroscopes based on a weak spatial temperature gradient field. First, distributed temperature data within a three-axis integrated fiber optic gyroscope inertial navigation system is acquired. This data is obtained from a temperature sensor array distributed on the surface of the heating device and the fiber optic gyroscope's casing according to rules covering the heat source and heat conduction path, thus containing information about the weak spatial temperature gradient field within the three-axis integrated fiber optic gyroscope inertial navigation system. Based on this distributed temperature data, a first-level model trained offline is used to determine the predicted value of the static thermally induced drift error. Next, a second-level model with online dynamic compensation, combined with external reference information, is used to calibrate the predicted value of the static thermally induced drift error to correct the error offset caused by time-varying factors in the three-axis integrated fiber optic gyroscope inertial navigation system, obtaining the final accurate error prediction value, which is then used to compensate the output of the fiber optic gyroscope. Through the above-mentioned two-stage progressive fiber optic gyroscope temperature compensation scheme, this application can not only correct the nonlinear error (the main error source) caused by the complex temperature field inside the three-axis integrated fiber optic gyroscope inertial navigation system, but also effectively overcome the first-level model failure problem caused by time-varying factors during the long-term operation of the three-axis integrated fiber optic gyroscope inertial navigation system, thus ensuring the high-precision performance of the three-axis integrated fiber optic gyroscope inertial navigation system throughout its entire life cycle. Attached Figure Description

[0007] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1This is an application environment diagram of a two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field in one embodiment of this application. Figure 2 A flowchart illustrating a two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field, provided as an embodiment of this application; Figure 3 Simulation cloud map of the internal temperature field of a three-axis integrated fiber optic gyroscope inertial navigation system provided in an embodiment of this application; Figure 4 A schematic diagram of the spatial layout of internal components and sensors of a three-axis integrated fiber optic gyroscope inertial navigation system provided in an embodiment of this application; Figure 5 This is a multi-channel temperature response curve obtained by a sensor array under varying temperature conditions, provided in one embodiment of this application. Figure 6 A block diagram illustrating the principle of a two-stage progressive fiber optic gyroscope temperature compensation system provided in an embodiment of this application; Figure 7 This is a comparison diagram of the gyroscope drift compensation effect provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0009] This application addresses the issue of fiber optic gyroscope accuracy being affected by the complex internal thermal environment and weak spatial temperature gradient field in a three-axis integrated fiber optic gyroscope inertial navigation system. To meet the high-precision application requirements of the three-axis integrated fiber optic gyroscope inertial navigation system under complex internal thermal environments and throughout its entire life cycle, a novel temperature compensation strategy is researched. To address the problem that existing single-point models cannot sense weak spatial gradients and cannot adapt to long-term changes, a two-stage progressive error compensation method based on a weak spatial temperature gradient field is proposed. This strategy divides the compensation process into two levels: first, an internal temperature field sensing network is constructed to extract gradient information. N A temperature feature vector is input into an offline-trained first-level model (the first-level static model) to correct nonlinear errors caused by complex temperature fields. Then, using the output of this first-level model as a benchmark, an online dynamically compensated second-level model is used to estimate and correct fine-tuning coefficients in real time. Considering that weak spatial temperature gradients are the main error source affecting the accuracy of the static model during the execution of this strategy, spatial feature extraction and dynamic adaptive calibration functions are added to the traditional model, thereby suppressing the accuracy loss of the high-precision navigation system in dynamic environments and long-term operation.

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0012] The two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send distributed temperature data from the three-axis integrated fiber optic gyroscope inertial navigation system to server 104. After receiving the distributed temperature data, server 104 uses an offline-trained first-level model to determine the predicted value of static thermal drift error; it then uses an online dynamic compensation second-level model, combined with external reference information, to calibrate the predicted value of static thermal drift error to correct the error shift caused by time-varying factors, obtaining the final accurate error prediction value. Server 104 can then feed back the obtained final accurate error prediction value to terminal 102. In addition, in some embodiments, the two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field can also be implemented by the server 104 or the terminal 102 separately. For example, the terminal 102 can directly process the distributed temperature data, or the server 104 can obtain the distributed temperature data from the data storage system and process it.

[0013] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers, or it can be a cloud server.

[0014] In one exemplary embodiment, such as Figure 2 As shown, a two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1Taking server 104 as an example, the explanation includes the following steps 201 to 203. Wherein: Step 201: Obtain distributed temperature data within the three-axis integrated fiber optic gyroscope inertial navigation system; the distributed temperature data is transmitted by a temperature sensor array within the three-axis integrated fiber optic gyroscope inertial navigation system; the temperature sensors in the array are distributed on the surface of the heat-generating device and the fiber optic gyroscope packaging shell according to the rules of covering the heat source and the heat conduction path.

[0015] Preferably, the temperature sensors in the array are evenly distributed on the surface of the heating device and the fiber optic gyroscope enclosure according to the rules of covering the heat source and the heat conduction path.

[0016] Step 202: Based on the distributed temperature data, determine the predicted value of the static thermal drift error using an offline-trained first-level model. This offline-trained first-level model characterizes the mapping relationship between distributed temperature and static thermal drift error within the three-axis integrated fiber optic gyroscope inertial navigation system.

[0017] Step 203: Using the online dynamic compensation secondary model and combined with external reference information, the predicted value of the static thermal drift error is calibrated to correct the error offset caused by time-varying factors, and the final accurate error prediction value is obtained; the final accurate error prediction value is used to compensate the output of the fiber optic gyroscope.

[0018] As an optional implementation, the offline-trained first-level model is a machine learning model trained on a dataset; the dataset includes a distributed temperature data sample set and a corresponding label dataset; the process of acquiring the dataset includes: 1-1) The three-axis integrated fiber optic gyroscope inertial navigation system was placed in a temperature-controlled experimental chamber.

[0019] 1-2) Using a temperature-controlled experimental chamber, the three-axis integrated fiber optic gyroscope inertial navigation system was subjected to cyclic heating and cooling at different temperature change rates to obtain a distributed temperature data sample set and the original drift error dataset of the fiber optic gyroscope in a zero-input state.

[0020] That is, different temperature change rates are used in each heating and cooling cycle.

[0021] 1-3) The original drift error dataset is determined as the corresponding label dataset.

[0022] As an optional implementation, the online dynamic compensation secondary model is a Kalman filter; the calibration process for the predicted static thermal drift error includes: 2-1) Construct a state vector containing the scaling correction coefficient and the bias correction coefficient.

[0023] 2-2) Establish the state transition equation based on the state vector.

[0024] 2-3) Construct an observation matrix based on the predicted static thermal drift error and establish an observation equation.

[0025] 2-4) When external reference information exists, the state vector is updated using a Kalman filter with the actual drift error determined based on the external reference information as the observation, to obtain the scaling correction coefficient and bias correction coefficient at the current time.

[0026] 2-5) When there is no external reference information, the scaling correction factor and offset correction factor of the previous time step are used as the scaling correction factor and offset correction factor of the current time step.

[0027] 2-6) The predicted value of static thermal drift error is calibrated using the proportional correction coefficient and the bias correction coefficient at the current time.

[0028] As an optional implementation, the observation matrix is ​​determined in real time based on the predicted static thermal drift error output by the offline-trained first-level model; the form of the observation matrix is... ,in, Indicates time The predicted value of static thermally induced drift error, Indicates the current iteration step The corresponding moment.

[0029] As an optional implementation, the calibration process for the predicted static thermal drift error further includes: 3-1) Determine whether the observation residual exceeds a preset threshold; the observation residual is determined based on the observation matrix and the observations.

[0030] 3-2) When the judgment result is yes, pause the state update of the Kalman filter and determine the static thermal drift error prediction value as the final accurate error prediction value.

[0031] As an optional implementation, step 2-6) "calibrating the predicted static thermal drift error value using the proportional correction coefficient and bias correction coefficient at the current moment" specifically includes: 2-6-1) Multiply the predicted value of the static thermal drift error by the proportional correction coefficient to obtain the first product.

[0032] 2-6-2) Add the first product to the bias correction coefficient to obtain the final accurate error prediction value.

[0033] As an optional implementation, before determining the predicted static thermally induced drift error using an offline-trained first-level model, the following steps are also included: The distributed temperature data is preprocessed, including moving average filtering and time alignment. The preprocessed distributed temperature data is then used as input to the offline-trained first-level model.

[0034] As an optional implementation, the primary model is a support vector regression model or a backpropagation neural network model.

[0035] As an optional implementation, external reference information includes Global Navigation Satellite System (GNSS) velocity observations and Zero Velocity Update (ZUPT).

[0036] As an optional implementation, the offline-trained first-level model is implemented using a lookup table method or a polynomial kernel computation.

[0037] To aid understanding by those skilled in the art, further explanation is provided below.

[0038] The overall structure is as follows: Part One analyzes the error mechanism of weak spatial temperature gradient field affecting the accuracy of a three-axis integrated fiber optic gyroscope inertial navigation system. Part Two describes the construction and feature extraction method of the internal temperature field sensing network. Part Three elucidates the mechanism of two-stage progressive temperature error modeling. Part Four presents the system architecture and implementation process. Part Five describes the design method of the online calibration filter. Part Six presents the effect verification and conclusions.

[0039] (1) Analysis of the error of weak spatial temperature gradient field affecting the accuracy of three-axis integrated fiber optic gyroscope inertial navigation system.

[0040] In a three-axis integrated fiber optic gyroscope inertial navigation system, the core thermal factor affecting the accuracy of the fiber optic gyroscope is no longer simply the external ambient temperature, but rather a dynamic temperature gradient field formed by the combined effects of the high-power components and the thermal conductivity of the compact structure within the system. This section focuses on analyzing the physical causes of this "weak spatial temperature gradient" and its nonlinear influence mechanism on the Shupe error of the fiber optic gyroscope.

[0041] (1.1) Physical causes of internal thermal environment and weak spatial gradient.

[0042] To achieve miniaturization and high integration, three-axis integrated fiber optic gyroscope inertial navigation systems typically encapsulate high-heat-generating components such as the main control FPGA chip and the light source driving circuit with the fiber optic gyroscope optical path module in the same sealed housing.

[0043] During system operation, the internal heat source continuously generates heat, which is conducted to the fiber optic loop through the metal substrate, fiber optic frame, and encapsulating colloid. Due to significant differences in the heat capacity and thermal conductivity of the various components within the system, the heat transfer path exhibits anisotropy. This non-uniform heat conduction process creates a dynamically changing, non-uniform temperature field within the system.

[0044] The temperature field exhibits typical characteristics of "weak spatial temperature gradient," meaning that although the absolute temperature difference between points in the fiber loop may be on the order of extremely small magnitude, the direction vector and intensity of the gradient will undergo slow and complex nonlinear evolution as the system preheating state, working mode switching and external environment changes.

[0045] (1.2) Mechanism of the influence of gradient field on Shupe error.

[0046] To intuitively illustrate this physical phenomenon, this embodiment presents a steady-state thermodynamic simulation of the internal structure of a three-axis integrated fiber optic gyroscope inertial navigation system.

[0047] See Figure 3 Simulation results show that, in core heat sources such as FPGAs (see...) Figure 3 The temperature in the region near the red cube (the middle part) is significantly higher than that at the edge of the system, forming a complex spatial temperature gradient distribution. The fiber loop situated in this non-uniform temperature field (see...) Figure 3 The temperature change rate sensed by the fiber optic micro-elements at different locations in the central ring section. Completely different. According to the Shupe effect principle, the non-reciprocal phase shift error generated in the optical fiber... It is the integral of the rate of temperature change with the fiber position weight: ; in, This represents the rate of change of the average temperature of the fiber optic loop over time. Represents the local position coordinates of an optical fiber element along the optical fiber path; Indicates the total length of the fiber optic loop; This represents a weak spatial temperature gradient term that is location-dependent. Due to the existence of this gradient term, even if the average rate of change of the ambient temperature is constant, changes in the internal gradient can introduce unpredictable drift errors.

[0048] Figure 3 In the text, time = 3600s indicates that the simulation time is 3600s.

[0049] (1.3) Limitations of the traditional single-point model.

[0050] Traditional temperature compensation methods are typically based on the "homogeneous body assumption" or the "linear gradient assumption," and only collect the temperature at a single location (usually a point on the surface or inside the gyroscope). Modeling and compensation are performed. However, in a three-axis integrated fiber optic gyroscope inertial navigation system, the output drift of the fiber optic gyroscope... B ( t In fact, it is the vector of the entire three-dimensional temperature field distribution. T ( t Multivariate nonlinear functions: ; in, These represent the temperatures of key thermal nodes within the system; This represents the spatial gradient characteristic. If we ignore this high-dimensional gradient effect and only use single-point temperature... Building a model is equivalent to forcibly simplifying a complex multivariate function into a univariate function, which will lead to... The resulting nonlinear error terms cannot be decoupled and eliminated, leaving a large amount of uncompensated error in the final output. Therefore, constructing an error model capable of sensing and analyzing this complex temperature field is key to improving the accuracy of a three-axis integrated fiber optic gyroscope inertial navigation system.

[0051] (2) Construction and feature extraction of internal temperature field sensing network.

[0052] To capture the "weak spatial temperature gradient" identified in Part 1 and effectively decouple the Shupe error, relying solely on single-point temperature measurement is insufficient. This section details how to construct a temperature field sensing network within an integrated system and extract high-dimensional temperature feature vectors from it.

[0053] (2.1) Sensor layout strategy based on thermal simulation.

[0054] Inside the compact, sealed housing of the navigation system, the following are arranged: N indivual( N ≥ 3) High-precision digital temperature sensor. Figure 4 This paper showcases the spatial layout of the core components and temperature sensor nodes within a three-axis integrated fiber optic gyroscope inertial navigation system. The sensor layout is not uniformly distributed but optimized based on the system's internal heat conduction paths. To maximize the observability of the sensed gradients, the following two core layout principles are followed: 1) Cover the main disturbance heat source: such as Figure 4 As shown, sensors are deployed at the sources of heat dissipation from high-power devices within the system. For example, sensor T1 is deployed close to the surface of the main control FPGA chip; sensor T2 is deployed on the optical module base containing the Y-waveguide and SLD light source. These locations are most sensitive to changes in operating conditions and can detect thermal shock inputs immediately.

[0055] 2) Cover critical heat conduction paths: such as Figure 4As shown, sensors are deployed on the surface of the fiber optic gyroscope package, which is most sensitive to temperature gradients. Key details include deploying sensor T3 on the light-receiving side (near-heat end) of the fiber optic gyroscope package near the heat source, and sensor T4 on the backlight side (far-heat end) away from the heat source. The temperature difference at these locations (such as the difference between T3 and T4) visually reflects the hysteresis and spatial attenuation characteristics of heat transfer through the metal substrate and dielectric.

[0056] Through this non-uniform "source + path" layout, the sensor network can capture the complete heat conduction chain from the heat source to the sensitive device at the physical level.

[0057] (2.2) Construction and analysis of multidimensional temperature feature vectors.

[0058] From this N A sensing array consisting of spatially distributed sensors is responsible for real-time acquisition of temperature field data inside the three-axis integrated fiber optic gyroscope inertial navigation system.

[0059] like Figure 5 As shown, in the full-temperature-range variable-temperature experiment, due to the different physical paths of each sensor from the heat source, their output temperature response curves exhibit significant differences: a. Amplitude difference: The temperature amplitude of the (sensor) probe (such as probe 3) closer to the heat source fluctuates more, while the amplitude of the probe (such as probe 1) farther away from the heat source is relatively flat.

[0060] b. Phase difference: There is a significant time delay (phase difference) between different curves.

[0061] These dynamic differences are not merely measurement noise; they inherently contain information about the direction vector and intensity changes of the temperature gradient within the system. If only single-point temperature measurement is used, this rich three-dimensional thermal field information will be lost.

[0062] The acquired multi-channel data is simultaneously sampled and preprocessed to construct a real-time... N 3D temperature eigenvector T ( t ): T ( t )=[ T 1( t ), T 2( t ),…, T N ( t )] T ; Among them, superscript T This indicates transpose.

[0063] This vector not only contains the common-mode component that reflects the average ambient temperature, but also implicitly contains the differential-mode (gradient) component that characterizes the spatial heat distribution pattern. Using it as input to the subsequent error model is sufficient to support the decoupling and reconstruction of complex nonlinear Shupe errors in the two-level model.

[0064] (3) Two-stage progressive temperature error modeling mechanism.

[0065] To address the mixed characteristics of "deterministic nonlinear thermal drift" and "random time-varying deviation" in the error sources of a three-axis integrated fiber optic gyroscope inertial navigation system, this application establishes a two-level progressive error model combining "offline high-dimensional static mapping" and "online low-dimensional dynamic adaptation". This architecture aims to use the static model to solve complex gradient field effects and the dynamic model to overcome the long-term variability of the device.

[0066] (3.1) First-level model: offline static mapping based on machine learning.

[0067] The primary model, serving as the baseline model of the system, has the core task of establishing a multidimensional temperature feature vector. T ( t ) and fiber optic gyroscope Shupe error The global mapping relationship between them.

[0068] Because the internal heat conduction path of a three-axis integrated fiber optic gyroscope inertial navigation system involves thermal resistance coupling at multiple material interfaces, its physical equations are extremely complex and difficult to solve analytically. Therefore, this embodiment adopts a data-driven "black box" modeling strategy.

[0069] a. Offline calibration and data acquisition: Before leaving the factory, the system is placed in a precision temperature-controlled experimental chamber and calibrated to cover the entire operating temperature range (e.g., -40℃ to +60℃). During the experiment, the ambient temperature is cyclically increased and decreased at different rates (e.g., 0.5℃ / min, 1℃ / min) to fully stimulate different forms of weak spatial temperature gradients within the system. The sensor array is recorded synchronously throughout the process. N Dimensional output T ( t The original drift error of fiber optic gyroscopes in zero-input state Construct a massive training dataset { T ( t ), }

[0070] b. Nonlinear regression modeling: Using Support Vector Regression (SVR) or Backpropagation Neural Network (BPNN), the above dataset is trained to obtain a nonlinear function. : ; The model It can keenly capture the nonlinear combination relationship between the components of the input vector (i.e., gradient features), thereby accurately fitting the deterministic drift caused by complex temperature fields.

[0071] c. Model output: During the real-time operation phase, the temperature characteristics collected in real time are input into the model to calculate the predicted value of static thermal drift error. This step can eliminate more than 90% of the nonlinear errors caused by ambient temperature and internal gradients, providing the system with a high-precision zero-bias reference.

[0072] (3.2) Second-level model: online dynamic adaptive calibration.

[0073] Although the Level 1 model effectively addresses the thermal drift problem, it is a "static" model, meaning its parameters remain fixed after training. During the long-term, full-lifecycle operation of a three-axis integrated fiber optic gyroscope inertial navigation system, the following factors can lead to prediction bias in the Level 1 model: Device aging: The decrease in light source power and the change in detector responsivity lead to scale factor drift.

[0074] Mechanical stress relief: Long-term vibration or impact causes changes in the stress state of the fiber optic loop encapsulation colloid, resulting in a shift of the zero-bias reference.

[0075] Startup characteristic differences: Small differences in circuit state may introduce random constant biases each time power is applied.

[0076] To address the aforementioned "model mismatch" problem, this application cascades a second-level model after the first-level model. This model assumes true drift. Compared with the predicted value There exists a local linear relationship: ; in, This represents the final compensation prediction value. The model contains two time-varying parameters to be estimated: (Proportional Correction Factor): Used to correct proportionality errors caused by changes in optical path gain or temperature gradient coupling strength; initial value is 1.

[0077] (Bias Correction Factor): Used to correct additive constant errors (i.e., changes in the zero-bias intercept) caused by device aging or startup randomness; the initial value is 0.

[0078] These two parameters are not preset, but are estimated and updated online in real time during the actual operation of the system by using a Kalman filter (see Part 5 for details) to back-calculate the residuals obtained from external reference information (such as GNSS velocity observations or zero-velocity correction ZUPT).

[0079] Through this two-level architecture, the first-level model is responsible for "explaining" complex physical laws (temperature gradient effect), while the second-level model is responsible for "adapting" to changes in the system's state (time aging effect). The two complement each other, achieving a dual improvement in accuracy and robustness.

[0080] (4) System architecture and implementation process.

[0081] In this embodiment, the proposed temperature compensation scheme for a three-axis integrated fiber optic gyroscope inertial navigation system adopts a "hardware-software hybrid" architecture in engineering implementation. The system includes not only a hardware network for sensing the physical field but also an embedded computing core running a two-level progressive algorithm. The overall architecture and data processing flow of this embodiment are as follows: Figure 6 As shown.

[0082] (4.1) Overall hardware architecture.

[0083] like Figure 6 As shown, the system hardware mainly consists of the following three parts: a. Physical perception layer: Fiber optic gyroscope component: As the object being compensated, it outputs the raw angular velocity signal, including thermal errors, in real time. .

[0084] Distributed temperature sensor array: consisting of key thermal nodes deployed within the system N A high-precision digital temperature sensor ( These sensors are composed of [missing information - likely a series of components]. They are connected to the main control unit via SPI or I2C bus to synchronously acquire temperature field data inside the system at high frequencies (such as 100Hz).

[0085] b. Core processing layer: Navigation computers are typically implemented using high-performance field-programmable gate arrays (FPGAs) or digital signal processors (DSPs). This layer carries the algorithm and integrates a feature extraction module, a first-level static inference engine, and a second-level dynamic filter.

[0086] c. Application output layer: Output high-precision angular velocity after double compensation It is used for inertial navigation system (INS) calculations or integrated navigation (GNSS / INS).

[0087] (4.2) Data processing and compensation implementation process.

[0088] The core operation of the system is a closed-loop real-time processing procedure, and the specific steps are detailed below: Step 1: Synchronous acquisition and preprocessing of multidimensional data.

[0089] The navigation computer reads temperature values ​​from the sensor array in parallel via a bus. To eliminate the quantization noise inherent in the sensor, the raw temperature data is first filtered using a moving average. Then, the temperature data from each channel are time-aligned to construct a time-series... t of N 3D temperature eigenvector: ; superscript T This indicates the transpose. This vector fully preserves the differential mode information that reflects the weak spatial temperature gradient.

[0090] Step 2: Offline inference of the first-level model (coarse compensation).

[0091] In this embodiment, the constructed feature vector Input into the SVR static regression model pre-stored in Flash memory.

[0092] In FPGAs / DSPs, this model is typically implemented using lookup tables or polynomial kernel computation. The arithmetic unit calculates based on the input... Quickly calculate the predicted static thermal drift caused by the temperature gradient at the current moment. : ; This step takes an extremely short time (microseconds) and can respond in real time to rapid changes in ambient temperature, eliminating the nonlinear principal components in the drift.

[0093] Step 3: Online calibration (fine compensation) of the second-level dynamic filter.

[0094] To correct residual errors caused by device aging, startup shocks, or model mismatch, the system runs a low-dimensional Kalman filter (KF) in parallel.

[0095] State observation: When the system is in a "zero-speed state" (e.g., the vehicle is stationary) or has a high-precision external reference (e.g., GNSS positioning is effective), the system calculates the true drift observations of the fiber optic gyroscope. .

[0096] Parameter update: The filter is updated based on the actual drift observation. The observed residuals are calculated by combining the output of the first-level model, and the proportional correction coefficients are updated in real time using a recursive algorithm. and bias correction coefficient These two parameters reflect the slight deviation of the current device state from its factory-calibrated state.

[0097] Step 4: Final drift synthesis and subtraction.

[0098] By combining the static thermally induced drift predictions from the first-level model with the dynamic correction parameters from the second-level model, the final total drift estimate is calculated. : ; Finally, from the raw output of the fiber optic gyroscope Subtracting the total drift from the final high-precision angular velocity output yields the final output. : ; Through the above process, the system completes a full cycle from physical field perception to two-level error stripping within each millisecond of the operation cycle, ensuring that the output data is always in the best accuracy state.

[0099] (5) Design of online calibration filter.

[0100] To implement the scaling correction coefficient in the second-level model k ( t and bias correction coefficient b ( t To achieve real-time accurate estimation of GNSS velocity, this embodiment designs a low-dimensional, high-efficiency linear Kalman filter. This filter operates within the navigation computer, utilizing external observation information (such as GNSS velocity observations or zero-velocity correction conditions) as feedback to continuously correct the prediction bias of the first-level model.

[0101] (5.1) Construction of state vector and state equation.

[0102] 1) Definition of state vector: Define the state vector of the filter A column vector containing two parameters to be estimated: ; in: This represents the correction gain for the output amplitude of the first-order model. Its physical meaning is to compensate for scale factor drift caused by optical path attenuation or changes in photodetector gain. Ideally... .

[0103] This indicates a correction bias for the zero-point output of the first-level model. Its physical meaning is to compensate for additive constant errors caused by startup randomness or stress release. Ideally... .

[0104] 2) State transition equation: Considering that device aging and environmental characteristics change slowly (with a time constant much larger than the filtering period) in the actual operation of a three-axis integrated fiber optic gyroscope inertial navigation system, the changes in these two parameters can be modeled as a random walk process. The discretized state transition equation is as follows: ; in: This is the state transition matrix. Since the parameters are considered as slowly varying constants, it is taken as the second-order identity matrix. Subscript " " is the most classic and rigorous general expression of discrete Kalman filtering (emphasizing the discrete transition process from time k-1 to time k). For the sake of standardization, we maintain a high degree of consistency with the notation in classic control theory textbooks.

[0105] The process noise vector follows a Gaussian distribution. Covariance matrix The settings reflect the system's tolerance to time-varying parameters: The larger the value, the faster the filter tracks parameter changes, but the noise immunity decreases. The smaller the value, the smoother the parameter estimation.

[0106] (5.2) Establishment of dynamic observation equations.

[0107] The observation equations serve as a bridge connecting the "mathematical model" and "physical observations." The ingenious design of the observation equations in this method lies in their incorporation of the output of the first-level model. It is dynamically embedded into the observation matrix.

[0108] 1) Obtaining observations: When the system has external reference information available (e.g., when the carrier is stationary and performing zero-velocity correction (ZUPT), or when the high-precision heading information provided by GNSS is effective), it can obtain the current true drift observation value of the fiber optic gyroscope. .

[0109] For example, at zero speed, the theoretical output of a gyroscope should be the Earth's rotation component. At this time, the actual output The deviation in the middle is called drift: ; 2) Form of the observation equation: Based on the linear correction model established in Part 3 The following observation equation can be established: ; 3) Dynamic characteristics of the observation matrix: The observation matrix is ​​not a constant, but changes in real time with the output of the first-level model: ; : Indicates state k ( t The coefficient of ) is the predicted value of the first-level model at the current moment. This reflects the idea of ​​"using the first-level model as a benchmark".

[0110] : Indicates state b ( t ) is an additive constant.

[0111] The observation noise mainly consists of the angle random walk (ARW) of the fiber optic gyroscope and quantization noise, and its covariance matrix is... It can be determined in advance through Allan analysis of variance.

[0112] (5.3) Filtering recursion and parameter update.

[0113] In each filtering cycle (e.g., 10ms or 100ms), the system executes the following standard Kalman filter recursive steps: Step 1: One-step prediction.

[0114] Predict the current state based on the estimate from the previous time step (for a random walk model, the predicted value equals the estimate from the previous time step): ; in, Indicates the current time (current iteration step). The predicted state value, and It is based on the previous moment (the previous iteration step). State estimate It's confirmed.

[0115] Simultaneously update the prediction error covariance matrix: ; in, Indicates the current time The prediction error covariance matrix, and It is based on the previous moment Prediction error covariance matrix Certain; Indicates the current time The process noise vector covariance matrix.

[0116] Step 2: Calculate the Kalman gain.

[0117] Calculate the degree of confidence the filter has in the new observation: ; Where the superscript -1 denotes the inverse of the matrix, Indicates the current time The observation noise covariance matrix, Indicates the current time The observation matrix.

[0118] Note that, due to Includes real-time Therefore, gain It is time-varying and can automatically adjust the correction intensity according to the current temperature gradient intensity.

[0119] Step 3: Status update.

[0120] State estimation corrected using observation residuals: ; At this point, the optimal fine-tuning coefficient for the current moment is obtained. and .

[0121] Step 4: Covariance update.

[0122] .

[0123] in, Indicates the current time The prediction error covariance matrix.

[0124] (5.4) Filter initialization and adaptive mechanism.

[0125] To ensure the stability of the filter during the startup phase, the following initialization strategy is adopted: Initial state values: This means that, under the default initial state, the first-level model is accurate, with no bias or gain errors.

[0126] Initial value of covariance: Set it to a small diagonal matrix to avoid drastic parameter oscillations in the initial stage.

[0127] Furthermore, to prevent filter divergence during drastic temperature changes (which increase the residuals of the first-order model), a novelty sequence monitoring mechanism can be introduced. When observing the residuals... When the threshold is exceeded, it is determined to be an abnormal disturbance, the status update is temporarily frozen, and only the first-level static compensation is retained, thereby further improving the robustness of the system.

[0128] (6) Effect verification and conclusion.

[0129] To verify the effectiveness of the two-stage progressive error compensation method based on weak spatial temperature gradient field sensing proposed in this application, this embodiment built a dynamic test environment covering the entire temperature range and conducted a long-term variable temperature drift test on the fiber optic gyroscope in the three-axis integrated fiber optic gyroscope inertial navigation system. Qualitative and quantitative analyses were performed on the data from different compensation stages.

[0130] (6.1) Experimental environment and testing plan.

[0131] The experimental subject was a prototype of a three-axis integrated fiber optic gyroscope inertial navigation system. To simulate the harsh thermal environment that might be encountered in real-world applications, the experiment was conducted on a precision temperature-controlled turntable. The test setup was as follows: Temperature profile: The set temperature range covers -40℃ to +60℃, adopts a "heating-holding-cooling" cycle mode, and sets a rapid temperature change rate of 1℃ / min during the heating and cooling stages to fully stimulate the weak spatial temperature gradient inside the system.

[0132] Data Acquisition: The system is positioned on a stationary base and synchronously records the temperature feature vector of the internal sensor array at a frequency of 100Hz. And the raw angular velocity output of the fiber optic gyroscope.

[0133] Comparison group setup: To verify the necessity of the stratification strategy, the following data were calculated: "uncompensated original data", "data compensated using only the first-level static model", and "data compensated using the two-level progressive model of this application".

[0134] (6.2) Comparative analysis of compensation effects.

[0135] Experimental results are as follows Figure 7 As shown in the figure, the zero-bias drift curves under three different treatment states within the same time period are displayed.

[0136] 1) Original drift characteristics (grey curve).

[0137] like Figure 7As shown by the gray curve (original drift), the output of the fiber optic gyroscope in the three-axis integrated fiber optic gyroscope inertial navigation system exhibits drastic fluctuations without compensation. Due to the Shupe effect and internal non-uniform thermal stress, the peak-to-peak drift exceeds 15 degrees per hour, and the fluctuation trend is highly correlated with changes in ambient temperature. If this large nonlinear error is not eliminated, it will cause the navigation solution to diverge rapidly.

[0138] 2) Compensation effect of the first-level static model (blue dashed line).

[0139] The blue dashed line (first-stage residual) shows the residual after compensation using only the first-stage SVR static model.

[0140] Gradient effect elimination: It can be seen that the large sinusoidal fluctuations in the original curve have been effectively "flattened," and the residual curve has become relatively smooth. This proves that the static model based on the multidimensional sensing network successfully captured the weak spatial temperature gradient inside the system and decoupled the main nonlinear thermally induced errors.

[0141] Residual bias analysis: However, the residual curve did not return to the zero axis as a whole, but instead showed a significant bias of approximately -15 deg / h, and exhibited a slow drifting trend over time. This verifies the aforementioned analysis: due to the time interval between offline calibration and actual operation, differences in device aging and startup states introduce constant biases that cannot be predicted by the static model.

[0142] 3) Two-level progressive model compensation effect (red solid line).

[0143] The solid red line (the proposed method) shows the final output after introducing a second-stage online calibration filter.

[0144] Adaptive Correction: Within a short time after the filter starts, the red curve rapidly converges from the bias position of the blue dashed line to near the 0 deg / h baseline. This indicates that the online estimation algorithm successfully identified the fine-tuning coefficients. k ( t )and b ( t ).

[0145] Accuracy Preservation: During subsequent temperature variations, despite continuous and drastic changes in ambient temperature, the residual remained stable near zero with minimal fluctuations. Statistically, the final output exhibits zero-bias stability (1... This represents an improvement of two orders of magnitude compared to the original data.

[0146] (6.3) Conclusion: In summary, the two-stage progressive fiber optic gyroscope temperature compensation method based on weak spatial temperature gradient field (sensing) proposed in this application has achieved significant technical results, addressing the technical challenges of complex internal thermal environment and long-term time-varying characteristics of devices in three-axis integrated fiber optic gyroscope inertial navigation systems. The gradient sensing problem has been solved: by constructing a multi-point temperature sensing network, the limitation of traditional single-point temperature measurement in being unable to characterize the internal non-uniform thermal field has been overcome, and the core error sources affecting accuracy have been captured from a physical level.

[0147] It achieves refined error stripping: "Offline global learning" eliminates complex nonlinear gradient effects, and "online local fine-tuning" adapts to device aging and environmental changes. The two-level model complements each other, ensuring both high responsiveness to thermal dynamics and zero-position stability during long-term operation.

[0148] The system's environmental adaptability has been improved: Experiments have shown that this method enables the three-axis integrated fiber optic gyroscope inertial navigation system to maintain high precision performance in harsh environments with a wide temperature range and large temperature gradients without the need for expensive temperature control measures, which has extremely high engineering application value.

[0149] The purpose of the method proposed in this application is to address the problem in three-axis integrated fiber optic gyroscope inertial navigation systems where the accuracy of the fiber optic gyroscope is limited by the internal dynamic temperature gradient field and cannot be adapted to device aging using a single static model. The proposed method captures weak spatial temperature gradient features by constructing a multi-point sensing network. After eliminating the main thermally induced errors through offline global mapping, it adapts to the long-term time-varying characteristics of the system through online local fine-tuning. The proposed method reveals the impact of weak spatial gradients on accuracy through mechanistic analysis, plans a two-stage processing flow from sensing to output, and designs an online calibration filter suitable for this strategy. Furthermore, the proposed method adopts a progressive combination of "offline static + online dynamic" approaches, significantly improving the environmental adaptability and long-term stability of the integrated system.

[0150] The contributions of the proposed method are as follows: 1) The formation of the weak spatial temperature gradient field inside the three-axis integrated fiber optic gyroscope inertial navigation system and its influence mechanism on the Shupe error of the fiber optic gyroscope were innovatively analyzed, and an internal temperature field sensing network based on multi-point layout was constructed.

[0151] 2) A two-level progressive error modeling mechanism is proposed, which integrates offline static mapping based on machine learning and online dynamic calibration based on Kalman filtering, and solves the problem of nonlinear compensation under complex thermal environments.

[0152] 3) The proposed method introduces an online estimation mechanism for fine-tuning coefficients, which effectively overcomes the model failure problem caused by device aging or sudden environmental changes, and ensures the high-precision performance of the system throughout its entire life cycle.

[0153] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 8 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores distributed temperature data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field.

[0154] Those skilled in the art will understand that Figure 8 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0155] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0156] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0157] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0158] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0159] 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.

[0160] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field, characterized in that, include: Acquire distributed temperature data within a three-axis integrated fiber optic gyroscope inertial navigation system; The distributed temperature data is transmitted by a temperature sensor array within the three-axis integrated fiber optic gyroscope inertial navigation system; the temperature sensors in the array are distributed on the surface of the heat-generating device and the fiber optic gyroscope enclosure according to the rule of covering the heat source and the heat conduction path. Based on the distributed temperature data, the predicted value of static thermal drift error is determined using an offline-trained first-level model; Using an online dynamic compensation secondary model and combined with external reference information, the predicted value of the static thermal drift error is calibrated to correct the error offset caused by time-varying factors, thus obtaining the final accurate error prediction value; the final accurate error prediction value is used to compensate the output of the fiber optic gyroscope.

2. The two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field according to claim 1, characterized in that, The offline-trained first-level model is a machine learning model trained on a dataset; the dataset includes a distributed temperature data sample set and a corresponding label dataset; the process of acquiring the dataset includes: The three-axis integrated fiber optic gyroscope inertial navigation system was placed in a temperature-controlled experimental chamber. Using a temperature-controlled experimental chamber, the three-axis integrated fiber optic gyroscope inertial navigation system was subjected to cyclic heating and cooling at different temperature change rates to obtain a distributed temperature data sample set and the original drift error dataset of the fiber optic gyroscope in a zero-input state. The original drift error dataset is identified as the corresponding label dataset.

3. The two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field according to claim 1, characterized in that, The online dynamic compensation two-level model is a Kalman filter; The calibration process for the predicted static thermal drift error includes: Construct a state vector that includes a scaling correction coefficient and a bias correction coefficient; Establish a state transition equation based on the state vector; An observation matrix is ​​constructed based on the predicted static thermal drift error, and an observation equation is established. When external reference information exists, the state vector is updated using a Kalman filter with the actual drift error determined based on the external reference information as the observation, to obtain the proportional correction coefficient and bias correction coefficient at the current moment. When no external reference information is available, the scaling correction factor and offset correction factor of the previous time step are used as the scaling correction factor and offset correction factor of the current time step. The predicted value of static thermal drift error is calibrated using the proportional correction factor and bias correction factor at the current moment.

4. The two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field according to claim 3, characterized in that, The observation matrix is ​​determined in real time based on the predicted static thermal drift error output by the offline-trained first-level model; the form of the observation matrix is ​​as follows: ,in, Indicates time The predicted value of static thermally induced drift error, Indicates the current iteration step The corresponding moment.

5. The two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field according to claim 4, characterized in that, The calibration process for the predicted static thermal drift error also includes: Determine whether the observation residual exceeds a preset threshold; the observation residual is determined based on the observation matrix and the observations. When the judgment result is yes, the state update of the Kalman filter is paused, and the static thermal drift error prediction value is determined as the final accurate error prediction value.

6. The two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field according to claim 3, characterized in that, The static thermally induced drift error prediction value is calibrated using the proportional correction factor and bias correction factor at the current moment, specifically including: The predicted static thermal drift error is multiplied by the proportional correction coefficient to obtain the first product; Add the first product to the bias correction coefficient to obtain the final accurate error prediction value.

7. The two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field according to claim 1, characterized in that, Before determining the predicted static thermal drift error using the offline-trained first-level model, the following steps are also included: The distributed temperature data is preprocessed; the preprocessing includes moving average filtering and time alignment. The preprocessed distributed temperature data is used as the input to the offline-trained first-level model.

8. The two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field according to claim 1, characterized in that, The primary model is either a support vector regression model or a backpropagation neural network model.

9. The two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field according to claim 1, characterized in that, The external reference information includes global navigation satellite system velocity observations and zero-velocity corrections.

10. The two-stage progressive fiber optic gyroscope temperature compensation method based on a weak spatial temperature gradient field according to claim 8, characterized in that, The offline-trained first-level model is implemented using a lookup table method or a polynomial kernel computation.

Images