Hemispherical resonator gyroscope inertial navigation system temperature compensation method and compensation system

Abstract

The invention discloses a hemispherical resonator gyroscope inertial navigation system temperature compensation method and compensation system, and the method comprises the steps: constructing a dynamic temperature space model, and achieving temperature drift characteristic modeling; designing an incremental adaptive neural network structure to enable the compensation model to adapt to the temperature change, and optimizing the compensation precision through a successive approximation incremental learning mode; temperature compensation and increment optimization are carried out in real time through combination of a dynamic temperature space model and an increment adaptive neural network, and compensation precision is maintained through dynamic updating of a temperature space and an increment learning mechanism of the increment adaptive neural network; output signal drift and compensation errors are monitored through a self-adaptive feedback mechanism, a dynamic temperature space model and incremental self-adaptive neural network parameters are dynamically adjusted, and the stability and the compensation effect are improved through an online correction and model self-optimization mechanism. According to the method, the problems of insufficient temperature drift modeling precision, poor adaptability and compensation precision attenuation after long-time operation in a traditional compensation method are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of inertial navigation systems, and in particular to a temperature compensation method and a compensation system for a hemispherical resonant gyro inertial navigation system. Background Art

[0002] With the continuous development of inertial navigation technology, hemispherical resonant gyroscope (HRG) is widely used in aerospace, ocean exploration, precision weapons and other fields due to its high precision and high stability. However, HRG often faces the impact of temperature changes on its performance in practical applications, especially when operating in complex environments. Temperature fluctuations often cause the drift of HRG output signals, which seriously affects the accuracy and stability of the inertial navigation system.

[0003] Temperature drift is one of the main factors affecting HRG performance, which is usually manifested as a strong correlation between the HRG output signal and the ambient temperature. Specifically, as the temperature rises or falls, the HRG's oscillation frequency, sensitivity, and system noise characteristics will change, and this change will cause its output signal to shift significantly. In order to ensure that the HRG can maintain high-precision working performance within the temperature range, temperature compensation technology has become the key to solving this problem.

[0004] Traditional temperature compensation methods are mostly based on linear models, which obtain the output data of HRG at different temperatures in advance and use methods such as linear regression to establish the relationship between temperature and HRG drift. Although this linear compensation method can provide certain corrections within a certain temperature range, its compensation accuracy often cannot meet the requirements of high-precision navigation systems. Especially in environments with drastic temperature changes, the temperature drift of HRG often shows highly nonlinear characteristics. Traditional linear models cannot effectively capture and correct this nonlinear change, resulting in a serious decline in the compensation effect.

[0005] In addition, with the continuous advancement of high-precision inertial navigation systems, the accuracy requirements for temperature compensation are becoming increasingly higher. The compensation strategy of traditional linear models fails to fully consider the complex, dynamic, nonlinear relationship between HRG output drift and temperature, resulting in large errors in the temperature compensation effect of the system under different working conditions. In order to overcome this bottleneck, researchers urgently need to explore new temperature compensation methods, especially advanced models and algorithms that can describe the nonlinear characteristics of HRG output drift and thus improve compensation accuracy.

[0006] Therefore, the temperature drift of the hemispherical resonant gyro (HRG) has obvious nonlinear characteristics. The traditional linear model is obviously difficult to accurately describe its complex temperature characteristics. It is urgent to establish a more complex nonlinear model to improve the accuracy and adaptability of temperature compensation. Summary of the invention

[0007] In order to solve the above problems, the present invention provides a temperature compensation method and compensation system for a hemispherical resonant gyro inertial navigation system.

[0008] To achieve the above purpose, the technical solution adopted by the present invention is as follows: On the one hand, the present invention discloses a temperature compensation method for a hemispherical resonant gyro inertial navigation system, comprising: Step 1: By collecting HRG output signal data under different temperature conditions, a dynamic temperature space model is constructed, and a high-order polynomial mapping function is used to dynamically adjust the model coefficients in combination with the temperature change rate to achieve temperature drift characteristic modeling; Step 2: Design an incremental adaptive neural network structure, and combine it with a local error adjustment mechanism to achieve dynamic adjustment of the network structure and learning rate, so that the compensation model can adapt to temperature changes, and optimize the compensation accuracy through a step-by-step incremental learning method; Step 3: Through the combination of the dynamic temperature space model and the incremental adaptive neural network, temperature compensation and incremental optimization are performed in real time, and the compensation accuracy is maintained through the dynamic update of the temperature space and the incremental learning mechanism of the incremental adaptive neural network; Step 4: Monitor the output signal drift and compensation error through the adaptive feedback mechanism, dynamically adjust the dynamic temperature space model and incremental adaptive neural network parameters, and improve the stability and compensation effect through online correction and model self-optimization mechanism.

[0009] Further: Step 1 comprises: Obtain the output signal data of HRG under different temperature conditions, and record the temperature value and its change rate at each experimental moment; By analyzing the experimental data, a multi-dimensional dynamic temperature space model that is dynamically updated with temperature changes is constructed to adjust the drift characteristics of the HRG according to different temperatures and change rates; According to the characteristics of experimental data, considering the influence of temperature and its changing rate on HRG output, the coefficients of the polynomial are determined through regression analysis; By measuring the temperature change rate and temperature gradient in real time, the coefficients of the high-order polynomial model are dynamically adjusted to adapt to rapid or slow changes in temperature, ensuring the accuracy of temperature compensation.

[0010] Further: Step 2 includes: Incremental learning neural network that dynamically adjusts the network structure based on real-time input signals to ensure that the temperature compensation of the HRG can adapt to changes in ambient temperature in real time; By designing a local error adjustment mechanism, the network can dynamically adjust the learning rate according to the temperature change rate and the change rate of HRG output drift, so that the compensation process can adapt to temperature changes; Through the incremental learning method of step-by-step approximation, the incremental adaptive neural network can fine-tune the network weights each time data arrives.

[0011] Further: Step 3 comprises: By combining the dynamic temperature space model with the incremental adaptive neural network, each time the temperature changes, local modeling is first performed through the dynamic temperature space model, and then incremental optimization is performed through the incremental adaptive neural network, and finally an accurate compensation signal is output; The dynamic temperature space model updates the geometric characteristics of the temperature space in real time according to the temperature changes of the environment; The incremental learning mechanism of the incremental adaptive neural network dynamically adjusts the learning rate and network weights according to real-time error feedback to ensure that the compensation model adapts to temperature changes.

[0012] Further, the step 4 comprises: By real-time monitoring of the drift and compensation error of the HRG output signal and combining the characteristics of temperature changes, the parameters of the dynamic temperature space model and the incremental adaptive neural network are dynamically adjusted to ensure compensation accuracy; By accumulating historical data and optimizing the model regularly, the stability and compensation effect of the system can be gradually improved; The HRG output drift is compensated through incremental learning and feedback correction.

[0013] On the other hand, the present invention discloses a temperature compensation system for a hemispherical resonant gyro inertial navigation system, comprising: Temperature space modeling and initial model setting module: By collecting HRG output signal data under different temperature conditions, a dynamic temperature space model is constructed, and the model coefficients are dynamically adjusted using a high-order polynomial mapping function combined with the temperature change rate to achieve temperature drift characteristic modeling; Incremental adaptive neural network structure module: Design an incremental adaptive neural network structure, combine the local error adjustment mechanism to achieve dynamic adjustment of the network structure and learning rate, so that the compensation model can adapt to temperature changes, and optimize the compensation accuracy through incremental learning method of step-by-step approximation; Temperature compensation dynamic optimization module: Through the combination of dynamic temperature space model and incremental adaptive neural network, temperature compensation and incremental optimization are carried out in real time, and compensation accuracy is maintained through dynamic update of temperature space and incremental learning mechanism of incremental adaptive neural network; Adaptive feedback and correction mechanism module: monitors output signal drift and compensation error through adaptive feedback mechanism, dynamically adjusts dynamic temperature space model and incremental adaptive neural network parameters, and improves stability and compensation effect through online correction and model self-optimization mechanism.

[0014] Further: the temperature space modeling and initial model setting module includes: Obtain the output signal data of HRG under different temperature conditions, and record the temperature value and its change rate at each experimental moment; By analyzing the experimental data, a multi-dimensional dynamic temperature space model that is dynamically updated with temperature changes is constructed to adjust the drift characteristics of the HRG according to different temperatures and change rates; According to the characteristics of experimental data, considering the influence of temperature and its changing rate on HRG output, the coefficients of the polynomial are determined through regression analysis; By measuring the temperature change rate and temperature gradient in real time, the coefficients of the high-order polynomial model are dynamically adjusted to adapt to rapid or slow changes in temperature, ensuring the accuracy of temperature compensation.

[0015] Further: the incremental adaptive neural network structure module includes: Incremental learning neural network that dynamically adjusts the network structure based on real-time input signals to ensure that the temperature compensation of the HRG can adapt to changes in ambient temperature in real time; By designing a local error adjustment mechanism, the network can dynamically adjust the learning rate according to the temperature change rate and the change rate of HRG output drift, so that the compensation process can adapt to temperature changes; Through the incremental learning method of step-by-step approximation, the incremental adaptive neural network can fine-tune the network weights each time data arrives.

[0016] Further: the temperature compensation dynamic optimization module includes: By combining the dynamic temperature space model with the incremental adaptive neural network, each time the temperature changes, local modeling is first performed through the dynamic temperature space model, and then incremental optimization is performed through the incremental adaptive neural network, and finally an accurate compensation signal is output; The dynamic temperature space model updates the geometric characteristics of the temperature space in real time according to the temperature changes of the environment; The incremental learning mechanism of the incremental adaptive neural network dynamically adjusts the learning rate and network weights according to real-time error feedback to ensure that the compensation model adapts to temperature changes.

[0017] Further: the adaptive feedback and correction mechanism module includes: By real-time monitoring of the drift and compensation error of the HRG output signal and combining the characteristics of temperature changes, the parameters of the dynamic temperature space model and the incremental adaptive neural network are dynamically adjusted to ensure compensation accuracy; By accumulating historical data and optimizing the model regularly, the stability and compensation effect of the system can be gradually improved; The HRG output drift is compensated through incremental learning and feedback correction.

[0018] Compared with the prior art, the present invention has the following technical advances: Traditional linear compensation methods cannot accurately describe the complex nonlinear characteristics of HRG temperature drift, and often cannot provide sufficient compensation accuracy when the temperature changes greatly. The present invention introduces a dynamic temperature space model and an incremental adaptive neural network, combined with the complex nonlinear relationship between temperature and drift output signals, to accurately capture the multi-dimensional impact of temperature changes on HRG performance. By customizing the nonlinear mapping function, this method can effectively model in different temperature ranges, significantly improving the accuracy of compensation, especially in environments with large temperature fluctuations, the compensation effect can still remain efficient and stable.

[0019] The present invention adopts an incremental adaptive neural network, which enables the system to automatically adjust the compensation model according to the real-time temperature and HRG output data. Unlike the traditional static training method, the dynamic temperature space model can optimize the model in real time through the incremental learning strategy of "progressive approximation" after each temperature fluctuation. Whenever new data arrives, the dynamic temperature space model can adjust the weight according to the current error and dynamically correct the network structure, thereby continuously improving the compensation accuracy of the system and avoiding the lag effect caused by over-reliance on historical data.

[0020] Another outstanding advantage of the present invention is the design of an adaptive feedback mechanism, which automatically adjusts the compensation strategy by real-time monitoring of the output signal drift of the HRG and combining the error between the model prediction value and the actual drift. When the system finds that the error caused by temperature change exceeds the preset threshold, the adaptive feedback mechanism will be triggered quickly and dynamically adjust the parameters of the compensation model according to the error size. This feedback mechanism not only corrects the instantaneous error when it occurs, but also gradually optimizes the dynamic temperature space model and the structure of the dynamic temperature space model according to the feedback data accumulated over a long period of time, thereby improving the long-term stability of the system.

[0021] The present invention also combines online correction and model self-optimization mechanisms, and can automatically optimize according to historical data and temperature change characteristics in long-term operation. After multiple runs and temperature fluctuations, the system can make regular corrections to the dynamic temperature space model and the dynamic temperature space model, so that the model continues to adapt to the temperature change trend and continuously improves the compensation accuracy. Even in a complex and dynamic working environment, the system can maintain high accuracy and stability, effectively avoiding the problem of accuracy degradation caused by long-term operation in traditional temperature compensation methods.

[0022] By combining the dynamic temperature space model with the dynamic temperature space model, the present invention can perform real-time dynamic compensation while the temperature changes. When the system detects a temperature change, the dynamic temperature space model first locally models the current temperature-drift relationship, and then performs incremental optimization through the dynamic temperature space model, and finally outputs a real-time compensation signal. This process ensures that the HRG output signal can be accurately corrected as the temperature changes, which not only reduces the hysteresis phenomenon in the traditional compensation method, but also improves the real-time nature of the compensation accuracy. In an environment with extreme temperature changes or long-term temperature fluctuations, traditional linear models are often unable to cope with the severity of temperature changes, resulting in a significant decrease in compensation accuracy. The present invention can not only provide high-precision compensation within the normal temperature range by adjusting the compensation model in real time, but also can perform effective self-correction under extreme environmental conditions, greatly improving the environmental adaptability of the system and ensuring that it maintains stable performance under variable temperature conditions.

[0023] In summary, the present invention effectively solves the problems of insufficient temperature drift modeling accuracy, poor adaptability and attenuation of compensation accuracy after long-term operation in traditional compensation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0025] In the attached picture: Figure 1 is a flow chart of the present invention; Figure 2 This is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0026] The following specific embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0027] Example 1 like Figure 1 As shown, the present invention discloses a temperature compensation method for a hemispherical resonant gyro inertial navigation system, comprising: Step 1: By collecting HRG output signal data under different temperature conditions, a dynamic temperature space model is constructed, and a high-order polynomial mapping function is used to dynamically adjust the model coefficients in combination with the temperature change rate to achieve temperature drift characteristic modeling; Step 2: Design an incremental adaptive neural network structure, and combine it with a local error adjustment mechanism to achieve dynamic adjustment of the network structure and learning rate, so that the compensation model can adapt to temperature changes, and optimize the compensation accuracy through a step-by-step incremental learning method; Step 3: Through the combination of the dynamic temperature space model and the incremental adaptive neural network, temperature compensation and incremental optimization are performed in real time, and the compensation accuracy is maintained through the dynamic update of the temperature space and the incremental learning mechanism of the incremental adaptive neural network; Step 4: Monitor the output signal drift and compensation error through the adaptive feedback mechanism, dynamically adjust the dynamic temperature space model and incremental adaptive neural network parameters, and improve the stability and compensation effect through online correction and model self-optimization mechanism.

[0028] Specifically, step 1 includes: Data acquisition: First, obtain the output signal data of HRG under different temperature conditions, and record the temperature value and its change rate at each experimental moment.

[0029] Dynamic temperature space: By analyzing the experimental data, a multi-dimensional dynamic temperature space model is constructed. This space is not static, but a surface space that is dynamically updated with temperature changes, and can adjust the drift characteristics of the HRG according to different temperatures and change rates.

[0030] Local high-order polynomial mapping function: According to the characteristics of the experimental data, a suitable high-order polynomial mapping function is selected, the influence of temperature and its rate of change on the HRG output is considered, and the coefficients of the polynomial are determined by regression analysis.

[0031] Dynamic adjustment: By measuring the temperature change rate and temperature gradient in real time, the coefficients of the high-order polynomial model are dynamically adjusted to adapt to rapid or slow changes in temperature to ensure the accuracy of temperature compensation.

[0032] In this step, the present embodiment first needs to construct a dynamic temperature space model and design a local high-order polynomial mapping function to achieve accurate modeling of the HRG output signal drift. This model will be based on the complex nonlinear relationship between temperature change and HRG drift characteristics, and can dynamically adjust the compensation parameters in different temperature ranges.

[0033] 1.1 Construction of dynamic temperature space model: In order to deal with the nonlinear effect of temperature on HRG output, this embodiment first needs to construct a dynamic temperature space. This space is not just a traditional static mapping, but a dynamic, multi-dimensional temperature drift surface. As the temperature changes, the shape of this space will change dynamically (such as curvature, slope, etc.).

[0034] 1.1.1 Nonlinear relationship between temperature and HRG output: The output drift of HRG usually shows a complex nonlinear law as the temperature changes. This law is affected by multiple factors, including but not limited to ambient temperature, material properties, and the design characteristics of the gyro itself. Therefore, the model needs to take these factors into consideration to establish a temperature space that accurately reflects the output drift of HRG.

[0035] 1.1.2 Space construction method: First, it is necessary to collect a large amount of experimental data, that is, to record the output signal of HRG at different temperatures. The data needs to cover a sufficiently wide temperature range, such as changes from extremely low temperature to high temperature environment.

[0036] Temperature space representation: In this embodiment, the output drift of the HRG is represented by a multidimensional function, and the relationship between the output signal and the temperature can be described by the following function: in: is the output drift signal of HRG; It's time Temperature value at the moment; is the gradient or rate of temperature change (i.e., rate of temperature change); is the noise term that represents the random error in the relationship between temperature and HRG drift.

[0037] This function represents a nonlinear mapping between temperature and output drift, and this mapping changes over time (i.e., changes in ambient temperature).

[0038] 1.1.3 Local geometric characteristics of dynamic temperature space: Temperature gradient and rate of change: This example introduces the rate of change of temperature ( ) is an important factor affecting HRG output drift. In different temperature ranges, the rate of temperature change ( ) may cause significant changes in the output drift characteristics. For example, when the temperature changes sharply, the drift of HRG is often no longer a simple linear relationship, but a more complex nonlinear behavior.

[0039] Therefore, the geometry of the dynamic temperature space needs to take into account the temperature gradient ( ) and temperature difference ( ) on the spatial morphology. These effects determine the way in which the parameters are adjusted in the compensation model.

[0040] 1.2 Design of local high-order polynomial mapping function: Based on the construction of the dynamic temperature space, this embodiment needs to design a local high-order polynomial mapping function to describe the complex nonlinear relationship between temperature and HRG output drift. This polynomial function must not only be able to reflect the immediate impact of temperature, but also consider the impact of temperature gradient and change rate on HRG output.

[0041] 1.2.1 Local high-order polynomial mapping function definition: Assume that the output drift of the HRG It can be expressed by a high-order polynomial function and the current temperature and temperature change rate In association, this embodiment defines the following local high-order polynomial mapping function: in: are the polynomial coefficients, representing the interactive effect of temperature and temperature change rate; It's time Temperature value at the moment; It's time The rate of temperature change at a given moment; and are the temperature power of the polynomial and the power of the rate of temperature change, respectively; is the error term, which is used to represent the noise or the part that cannot be fitted by the polynomial.

[0042] 1.2.2 How to choose the polynomial order: In order to ensure the accuracy and computational efficiency of the model, this embodiment can use a stepwise regression method to determine the appropriate order and ,In the preliminary modeling, we can start from a lower order, and gradually increase the order by fitting the experimental data,until the model can accurately fit the output signal within the range of the experimental data.

[0043] In actual implementation, a technique called cross-validation is often used to select the order of the polynomial. By dividing the data set into a training set and a validation set, the performance of models of different orders is evaluated and an optimal polynomial order is selected.

[0044] 1.2.3 Dynamically adjusted local temperature perception function: During the working process of HRG, the temperature not only changes in time, but may also present different dynamic characteristics in different physical environments. Therefore, the local high-order polynomial mapping function needs to dynamically adjust its order and coefficient according to the actual temperature gradient and change rate in different temperature ranges.

[0045] For example, in an environment where the temperature changes rapidly, the rate of temperature change It may become a major influencing factor. In this case, this embodiment can capture these high-frequency changes by adding high-order terms. When the temperature changes slowly, the low-order terms may be able to accurately fit the data.

[0046] Through these steps, this embodiment can establish a dynamic temperature space model that can cope with nonlinear temperature drift and perform efficient compensation.

[0047] In step 1, this embodiment successfully establishes a nonlinear mapping relationship between temperature and HRG output drift through a dynamic temperature space model. However, since the temperature change is dynamic and time-varying in actual work, in order to effectively compensate for the drift of the HRG output signal, this embodiment must design an incremental adaptive neural network so that the system can adjust the network structure and weights in real time with new data input, thereby achieving dynamic adaptation. More importantly, when the temperature changes drastically, the incremental adaptive neural network should be able to optimize the compensation strategy in real time through an incremental learning mechanism, thereby effectively compensating for the drift error caused by temperature changes.

[0048] Specifically, step 2 includes: Incremental adaptive neural network structure design: By designing an incremental learning neural network that can dynamically adjust the network structure according to real-time input signals, it is ensured that the temperature compensation of HRG can adapt to changes in ambient temperature in real time.

[0049] Local error adjustment mechanism: By designing a local error adjustment mechanism, the network can dynamically adjust the learning rate according to the rate of change of temperature and the rate of change of HRG output drift, so that the compensation process can quickly adapt to drastic temperature changes while avoiding over-adjustment when the temperature change is small.

[0050] Progressive approximation and incremental learning: Through the incremental learning method of progressive approximation, the incremental adaptive neural network can fine-tune the network weights every time data arrives, ensuring the stability and efficiency of compensation accuracy.

[0051] This incremental adaptive neural network design can dynamically adjust its structure and learning process in the face of nonlinear temperature drift and rapidly changing environments, thereby achieving high-precision temperature compensation.

[0052] 2.1 The proposal of incremental adaptive neural network: Traditional neural networks usually rely on a large amount of training data for batch training, and cannot be adjusted dynamically after training is completed. However, in the present invention, the incremental adaptive neural network is different from the traditional training method. It can gradually approach the target value according to the temperature change and the real-time data of HRG output, and continuously adjust its internal structure and parameters during operation. Whenever the system receives new temperature and HRG output data, the incremental adaptive neural network will dynamically adjust the model parameters according to the current error and input signal to refine the compensation process.

[0053] 2.1.1 Design of network structure: The basic structure of the incremental adaptive neural network consists of an input layer, a hidden layer, and an output layer. The input layer receives temperature data, temperature change rate, and HRG output signals from the temperature space model. The hidden layer dynamically adjusts the weight parameters based on the temperature change rate and feedback information of the historical compensation error. The output layer provides a real-time compensation signal to correct the drift of the HRG.

[0054] The network structure has the ability to be incrementally adjusted: The network structure of the incremental adaptive neural network is relatively simple in the initial setting, but as new data arrives, the network can gradually expand its complexity by adjusting the number of nodes and connection weights in the hidden layer, thereby better fitting the temperature drift relationship of the HRG.

[0055] 2.1.2 Incremental learning mechanism: Stepwise approximation training strategy: Unlike the one-time training of traditional neural networks, the incremental adaptive neural network adopts a stepwise approximation strategy. When new temperature data and HRG output signals arrive, the incremental adaptive neural network will adjust the network weights and biases based on the error between the current network output and the actual value. This adjustment is incremental rather than global, so it can adapt to new input data efficiently and quickly without the need for retraining.

[0056] Fine-tuning mechanism: The network makes small adjustments every time it receives new data to ensure that the model always maintains high compensation accuracy in a changing temperature environment.

[0057] 2.2 Design of local error adjustment mechanism: The core of incremental learning lies in how to dynamically adjust the learning rate and increment size of the network according to the temperature change rate and the change rate of HRG output drift. To this end, this embodiment designs an innovative local error adjustment mechanism. Its core goal is to adjust the learning process through real-time feedback so that the network can quickly and effectively adapt to and maintain a high compensation accuracy when facing a complex environment with temperature changes.

[0058] 2.2.1 Error feedback and increment adjustment: After receiving new temperature and HRG output data each time, the incremental adaptive neural network calculates the error between the current output signal and the target signal (i.e., the compensation error): in: is the actual measured HRG output signal; is the HRG output signal predicted by the incremental adaptive neural network model.

[0059] 2.2.2 Error correction and dynamic adjustment of learning rate: In traditional neural networks, the learning rate is fixed, while incremental adaptive neural networks use a dynamic learning rate adjustment mechanism. When the temperature is large, or when the drift change rate of the HRG output is large, the local error adjustment mechanism will adjust the network's learning rate according to the current error size and temperature change rate ( ) to speed up the network's adaptation to these rapid changes.

[0060] Specifically, the learning rate The adjustment formula can be expressed as: in: is the initial learning rate; is the factor affecting the learning rate by the rate of temperature change; is the factor that affects the learning rate; It is the absolute value of the temperature change rate, reflecting the severity of the temperature change; It is the absolute value of the current compensation error, reflecting the gap between the model prediction and the actual value.

[0061] Through this dynamic adjustment mechanism, the learning rate can be increased in the face of rapidly changing temperatures so that the network responds quickly, and decreased when the temperature changes are smaller and the error is smaller, avoiding over-adjustment and overtraining.

[0062] 2.3 Learning process of incremental adjustment: In the incremental learning process, the incremental adaptive neural network will gradually update the weights and biases of the network based on the current error and learning rate. The update formula is: in: It is weights; is the weight increment adjusted by error feedback.

[0063] The key to this process is that the network gradually updates its parameters at each time step, rather than making large-scale adjustments at one time, ensuring that the compensation model can be fine-tuned within a stable range and achieve high accuracy.

[0064] In step one and step two, by closely combining the dynamic temperature space model and the incremental adaptive neural network during system operation, this embodiment can, through a dynamic real-time optimization compensation process, enable the HRG output signal to be accurately corrected following the temperature change each time a temperature change occurs, thereby ensuring the stability and high precision of the system.

[0065] Specifically, step 3 includes: Real-time temperature compensation: By combining the dynamic temperature space model with the incremental adaptive neural network, the system can first perform local modeling through the dynamic temperature space model every time the temperature changes, and then perform incremental optimization through the incremental adaptive neural network, and finally output an accurate compensation signal.

[0066] Dynamic temperature space update: The dynamic temperature space model will update the geometric characteristics of the temperature space in real time according to the temperature changes of the environment to ensure the stability and accuracy of the compensation strategy.

[0067] Incremental learning mechanism: The incremental learning mechanism of the incremental adaptive neural network dynamically adjusts the learning rate and network weights according to real-time error feedback, ensuring that the compensation model can quickly adapt to changes in temperature.

[0068] 3.1 Real-time dynamic compensation based on dynamic temperature space model and incremental adaptive neural network: The temperature compensation process of the system is an iterative process. Whenever the system detects a temperature change, it first locally models the relationship between the current temperature and the HRG output drift through a dynamic temperature space model, then incrementally optimizes the current model using an incremental adaptive neural network, and finally outputs a real-time compensation signal. The specific process can be divided into the following steps: 3.1.1 Temperature detection and real-time modeling of dynamic temperature space model: When the system detects new temperature data When arriving, the dynamic temperature space model uses the current temperature value and temperature gradient (such as ) Local modeling of the temperature space is performed to calculate the temperature drift characteristics at the current moment. This local modeling process can generate an immediate temperature drift prediction to provide a preliminary reference for subsequent compensation.

[0069] For example, the dynamic temperature space model can generate a preliminary compensation mapping function based on the local temperature gradient and previous historical data: in: is the current predicted HRG output drift; and are the current temperature and temperature gradient respectively; It is the system prediction error, which is used to reflect the prediction deviation of the dynamic temperature space model.

[0070] 3.1.2 Incremental Learning Optimization of Incremental Adaptive Neural Networks: After the dynamic temperature space model provides a preliminary prediction of the temperature drift characteristics, the incremental adaptive neural network will dynamically adjust the structure of the neural network and optimize the compensation effect based on the current prediction error and real-time temperature data using an incremental learning mechanism. Specifically, the incremental adaptive neural network will feedback and adjust the error between the current prediction and the actual measurement value, gradually optimizing the weights and structure of the network to adjust the compensation model in real time.

[0071] In this process, the incremental learning mechanism of the incremental adaptive neural network can dynamically update the weights of the network through the following feedback mechanism: in: It is weights; is the weight increment caused by error feedback; is the dynamically adjusted learning rate, determined by the local error adjustment mechanism (as described in step 2); is the error of the current model, reflecting the gap between the prediction and the actual output.

[0072] 3.1.3 Real-time compensation signal output: Finally, through the incremental optimization of the incremental adaptive neural network, the compensation signal will be updated and output to the HRG system. The output signal will correct the HRG output according to the real-time compensation strategy to ensure that the drift of the HRG can be optimally compensated at the current temperature.

[0073] 3.2 Real-time update of dynamic temperature space: The dynamic temperature space model is not just a static temperature drift map, but a dynamic space that can be updated in real time according to environmental changes. During system operation, as the temperature changes, the dynamic temperature space model will continuously update the geometric characteristics of the temperature space (such as temperature curvature, slope, temperature gradient, etc.) to ensure the accuracy of the compensation strategy.

[0074] 3.2.1 Update of temperature space geometric characteristics: The dynamic update of the temperature space model relies on real-time input data. When the temperature changes greatly, the dynamic temperature space model can detect these changes and dynamically adjust the modeling method of the temperature drift relationship. This update is not just a simple mapping of the current temperature, but also includes the optimization of local geometric characteristics such as temperature gradient and temperature change rate. For example, when the temperature changes sharply, the dynamic temperature space model will identify this nonlinear feature and optimize the model through incremental learning.

[0075] Mathematical expression for temperature space adjustment: in: is the updated temperature drift mapping function; is the initial temperature drift map; It is the compensation adjustment caused by new temperature data and temperature gradient.

[0076] 3.2.2 Self-correction mechanism in extreme environments: In some extreme temperature change environments, the dynamic temperature space model will self-correct its compensation model to cope with nonlinear temperature drift relationships. When the system detects that the drift of HRG presents unprecedented nonlinear characteristics, the dynamic temperature space model will adjust the compensation model through an incremental learning mechanism to further optimize the compensation accuracy.

[0077] 3.3 The core role of incremental learning mechanism: The incremental learning mechanism is the core of this step. It can dynamically adjust the network weights, optimize the compensation strategy, and adapt to the rapidly changing temperature environment during system operation. The incremental learning mechanism works through the following key links: 3.3.1 Real-time error feedback and dynamic adjustment of learning rate: After each new temperature data is input, the incremental adaptive neural network calculates the real-time error and dynamically adjusts the learning rate to ensure that the network can quickly adapt and compensate for HRG drift in a rapidly changing temperature environment.

[0078] 3.3.2 Fine-tuning of compensation strategy: The incremental learning mechanism can make fine compensation adjustments for each temperature change instead of relying solely on historical data, which ensures that the drift of the HRG can be accurately corrected even in extreme temperature environments.

[0079] 3.3.3 Dynamic weight update: The incremental learning mechanism enables the incremental adaptive neural network to optimize the compensation effect by fine-tuning the network weights each time new data is received, thereby ensuring that the compensation signal can perfectly match the real-time temperature changes.

[0080] Through this dynamic optimization mechanism that combines the dynamic temperature space model with the incremental adaptive neural network, temperature compensation can not only be performed in real time, but also adaptively respond to various complex temperature environments, ensuring that the HRG output signal always maintains high accuracy and stability.

[0081] In the previous steps, this embodiment designed the basic framework of temperature compensation through a dynamic temperature space model and an incremental adaptive neural network, combined with incremental learning and dynamic adjustment mechanisms to ensure that the system can respond to and compensate for temperature changes in real-time operation. However, in order to further improve the compensation accuracy of the system, especially to cope with the nonlinear characteristics of temperature and HRG drift during long-term operation, it becomes crucial to design an adaptive feedback mechanism and an online correction and model self-optimization mechanism. Through the adaptive feedback mechanism and online correction, this embodiment can realize a dynamic closed-loop temperature compensation system to ensure that the model can adaptively improve accuracy and adapt to changing trends.

[0082] Specifically, step 4 includes: The adaptive feedback mechanism monitors the drift and compensation error of the HRG output signal in real time, combines the characteristics of temperature change, and dynamically adjusts the parameters of the dynamic temperature space model and the incremental adaptive neural network to ensure the compensation accuracy.

[0083] The online correction and model self-optimization mechanism gradually improves the system stability and compensation effect by accumulating historical data and optimizing the model regularly.

[0084] The incremental adaptive network algorithm achieves accurate compensation for HRG output drift through incremental learning and feedback correction, enhancing the system's adaptability and accuracy to long-term temperature changes.

[0085] 4.1 Design of adaptive feedback mechanism: The core goal of the adaptive feedback mechanism is to monitor the drift of the HRG output signal in real time, and adjust the parameters of the dynamic temperature space model and the incremental adaptive neural network model according to the error between the output signal and the model prediction output to improve the compensation accuracy. Specifically, the adaptive feedback mechanism optimizes the system through the following steps: 4.1.1 Drift monitoring and error calculation: After each temperature change, the system generates the current HRG output drift prediction value through the combined model of dynamic temperature space model and incremental adaptive neural network. and the actual measured value For comparison, the error This is the difference between these two: The error reflects the temperature compensation effect of the current model. If the error exceeds the preset threshold, it is considered that the compensation accuracy is insufficient, and the adaptive feedback mechanism will be activated to start adjusting the parameters of the dynamic temperature space model and the incremental adaptive neural network.

[0086] 4.1.2 Dynamic error feedback and threshold triggering: When the error is detected When the error is greater than the set threshold, the adaptive feedback mechanism will start the automatic adjustment process. Specifically, the adaptive feedback mechanism will compare the current error with the trend of the historical error and combine it with the temperature change rate. To dynamically adjust the learning rate and the correction amplitude of the model.

[0087] For example, when the system detects that the temperature change rate is large and the error is large, the system will increase the correction of the dynamic temperature space model and the incremental adaptive neural network model parameters to improve the compensation accuracy.

[0088] 4.1.3 Real-time update of dynamic temperature space model and incremental adaptive neural network parameters: Dynamic temperature space model correction: By adaptively adjusting the temperature drift mapping function of the dynamic temperature space model to more accurately describe the HRG output behavior under the current temperature change, the updated mathematical expression is: in, It represents the dynamic correction triggered by the adaptive feedback mechanism, which is determined by the current error and the temperature change rate.

[0089] Incremental adaptive neural network parameter adjustment: The adaptive feedback mechanism dynamically adjusts the learning rate of the incremental adaptive neural network based on the characteristics of error and temperature changes .

[0090] Incremental update mechanism: incrementally adjust the weights of the neural network by incrementally , to adapt to new compensation needs.

[0091] 4.2 Online correction and model self-optimization: In order to ensure that the system can continuously improve the compensation accuracy in long-term operation, in addition to real-time feedback adjustment, it is also necessary to introduce an online correction and model self-optimization mechanism, that is, to regularly correct the dynamic temperature space model and incremental adaptive neural network through accumulated historical data and feedback information. This mechanism not only responds to instantaneous temperature changes, but also gradually optimizes the model after multiple operation cycles to enhance the overall stability and compensation effect of the system.

[0092] 4.2.1 Historical data accumulation and trend analysis: The system will analyze the error trend based on the temperature change and drift signal data during long-term operation. By statistically analyzing the historical error data, the system can identify the potential rules in the temperature drift relationship, thereby optimizing the temperature space model of the dynamic temperature space model and the learning algorithm of the incremental adaptive neural network.

[0093] 4.2.2 Regular self-optimization process: During long-term operation, the system will periodically (for example, after a certain period of time or amount of data) self-optimize the model. This self-optimization is not only local incremental learning, but also includes the reconstruction and fine-tuning of the entire temperature compensation framework.

[0094] Through adaptive correction of the dynamic temperature space model and the incremental adaptive neural network, the system can continuously adjust the model structure to cope with long-term changing environmental conditions and HRG output characteristics. For example, when the system accumulates a certain amount of temperature change and compensation data, the dynamic temperature space model will globally optimize the temperature drift mapping of the model, while the incremental adaptive neural network will retrain some network nodes to adapt to the new temperature compensation requirements.

[0095] 4.2.3 Corrected compensation output: After completing the model correction, the system will update the output of the compensation signal so that the drift error of the HRG can be corrected more accurately at the next temperature change. The final compensation output signal It can be expressed by the following formula: in, It is the modified temperature drift mapping function of the dynamic temperature space model, which represents the compensation for the HRG output drift after adaptive correction.

[0096] 4.3 Combination of adaptive feedback and incremental adaptive network algorithm: By combining the adaptive feedback mechanism and the incremental adaptive neural network, this embodiment ensures that the system's compensation capability can not only adapt to each temperature change in real time, but also self-optimize and correct according to long-term operating data. The incremental adaptive network algorithm enables the system to quickly adjust the compensation model after each temperature fluctuation through its incremental learning characteristics, while the adaptive feedback mechanism enables the system to continuously improve the compensation accuracy in complex environments through real-time error feedback and dynamic adjustment mechanisms.

[0097] The design of this closed-loop adaptive feedback and online correction mechanism enables the system to maintain efficient and stable temperature compensation functions in complex and changing temperature environments, and continuously optimize the compensation accuracy during long-term operation.

[0098] Example 2 like Figure 2 As shown, the present invention discloses a temperature compensation system for a hemispherical resonant gyro inertial navigation system, comprising: Temperature space modeling and initial model setting module: By collecting HRG output signal data under different temperature conditions, a dynamic temperature space model is constructed, and the model coefficients are dynamically adjusted using a high-order polynomial mapping function combined with the temperature change rate to achieve temperature drift characteristic modeling; Incremental adaptive neural network structure module: Design an incremental adaptive neural network structure, combine the local error adjustment mechanism to achieve dynamic adjustment of the network structure and learning rate, so that the compensation model can adapt to temperature changes, and optimize the compensation accuracy through incremental learning method of step-by-step approximation; Temperature compensation dynamic optimization module: Through the combination of dynamic temperature space model and incremental adaptive neural network, temperature compensation and incremental optimization are carried out in real time, and compensation accuracy is maintained through dynamic update of temperature space and incremental learning mechanism of incremental adaptive neural network; Adaptive feedback and correction mechanism module: monitors output signal drift and compensation error through adaptive feedback mechanism, dynamically adjusts dynamic temperature space model and incremental adaptive neural network parameters, and improves stability and compensation effect through online correction and model self-optimization mechanism.

[0099] Specifically, the temperature space modeling and initial model setting module includes: Obtain the output signal data of HRG under different temperature conditions, and record the temperature value and its change rate at each experimental moment; By analyzing the experimental data, a multi-dimensional dynamic temperature space model that is dynamically updated with temperature changes is constructed to adjust the drift characteristics of the HRG according to different temperatures and change rates; According to the characteristics of experimental data, considering the influence of temperature and its changing rate on HRG output, the coefficients of the polynomial are determined through regression analysis; By measuring the temperature change rate and temperature gradient in real time, the coefficients of the high-order polynomial model are dynamically adjusted to adapt to rapid or slow changes in temperature, ensuring the accuracy of temperature compensation.

[0100] Specifically, the incremental adaptive neural network structure module includes: Incremental learning neural network that dynamically adjusts the network structure based on real-time input signals to ensure that the temperature compensation of the HRG can adapt to changes in ambient temperature in real time; By designing a local error adjustment mechanism, the network can dynamically adjust the learning rate according to the temperature change rate and the change rate of HRG output drift, so that the compensation process can adapt to temperature changes; Through the incremental learning method of step-by-step approximation, the incremental adaptive neural network can fine-tune the network weights each time data arrives.

[0101] Specifically, the temperature compensation dynamic optimization module includes: By combining the dynamic temperature space model with the incremental adaptive neural network, each time the temperature changes, local modeling is first performed through the dynamic temperature space model, and then incremental optimization is performed through the incremental adaptive neural network, and finally an accurate compensation signal is output; The dynamic temperature space model updates the geometric characteristics of the temperature space in real time according to the temperature changes of the environment; The incremental learning mechanism of the incremental adaptive neural network dynamically adjusts the learning rate and network weights according to real-time error feedback to ensure that the compensation model adapts to temperature changes.

[0102] Specifically, the adaptive feedback and correction mechanism module includes: By real-time monitoring of the drift and compensation error of the HRG output signal and combining the characteristics of temperature changes, the parameters of the dynamic temperature space model and the incremental adaptive neural network are dynamically adjusted to ensure compensation accuracy; By accumulating historical data and optimizing the model regularly, the stability and compensation effect of the system can be gradually improved; The HRG output drift is compensated through incremental learning and feedback correction.

[0103] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of protection of the claims of the present invention.

Claims

1. A temperature compensation method for a hemispherical resonant gyro inertial navigation system, characterized in that: include: Step 1: By collecting HRG output signal data under different temperature conditions, a dynamic temperature space model is constructed, and a high-order polynomial mapping function is used to dynamically adjust the model coefficients in combination with the temperature change rate to achieve temperature drift characteristic modeling; Step 2: Design an incremental adaptive neural network structure, and combine it with a local error adjustment mechanism to achieve dynamic adjustment of the network structure and learning rate, so that the compensation model can adapt to temperature changes, and optimize the compensation accuracy through a step-by-step incremental learning method; Step 3: Through the combination of the dynamic temperature space model and the incremental adaptive neural network, temperature compensation and incremental optimization are performed in real time, and the compensation accuracy is maintained through the dynamic update of the temperature space and the incremental learning mechanism of the incremental adaptive neural network; Step 4: Monitor the output signal drift and compensation error through the adaptive feedback mechanism, dynamically adjust the dynamic temperature space model and incremental adaptive neural network parameters, and improve the stability and compensation effect through online correction and model self-optimization mechanism.

2. A temperature compensation method for a hemispherical resonant gyro inertial navigation system according to claim 1, characterized in that: The step 1 comprises: Obtain the output signal data of HRG under different temperature conditions, and record the temperature value and its change rate at each experimental moment; By analyzing the experimental data, a multi-dimensional dynamic temperature space model that is dynamically updated with temperature changes is constructed to adjust the drift characteristics of the HRG according to different temperatures and change rates; According to the characteristics of experimental data, considering the influence of temperature and its changing rate on HRG output, the coefficients of the polynomial are determined through regression analysis; By measuring the temperature change rate and temperature gradient in real time, the coefficients of the high-order polynomial model are dynamically adjusted to adapt to rapid or slow changes in temperature, ensuring the accuracy of temperature compensation.

3. A temperature compensation method for a hemispherical resonant gyro inertial navigation system according to claim 2, characterized in that: The step 2 comprises: Incremental learning neural network that dynamically adjusts the network structure based on real-time input signals to ensure that the temperature compensation of the HRG can adapt to changes in ambient temperature in real time; By designing a local error adjustment mechanism, the network can dynamically adjust the learning rate according to the temperature change rate and the change rate of HRG output drift, so that the compensation process can adapt to temperature changes; Through the incremental learning method of step-by-step approximation, the incremental adaptive neural network can fine-tune the network weights each time data arrives.

4. A temperature compensation method for a hemispherical resonant gyro inertial navigation system according to claim 3, characterized in that: The step 3 comprises: By combining the dynamic temperature space model with the incremental adaptive neural network, each time the temperature changes, local modeling is first performed through the dynamic temperature space model, and then incremental optimization is performed through the incremental adaptive neural network, and finally an accurate compensation signal is output; The dynamic temperature space model updates the geometric characteristics of the temperature space in real time according to the temperature changes of the environment; The incremental learning mechanism of the incremental adaptive neural network dynamically adjusts the learning rate and network weights according to real-time error feedback to ensure that the compensation model adapts to temperature changes.

5. A temperature compensation method for a hemispherical resonant gyro inertial navigation system according to claim 4, characterized in that: The step 4 comprises: By real-time monitoring of the drift and compensation error of the HRG output signal and combining the characteristics of temperature changes, the parameters of the dynamic temperature space model and the incremental adaptive neural network are dynamically adjusted to ensure compensation accuracy; By accumulating historical data and optimizing the model regularly, the stability and compensation effect of the system can be gradually improved; The HRG output drift is compensated through incremental learning and feedback correction.

6. A temperature compensation system for a hemispherical resonant gyro inertial navigation system, characterized in that: include: Temperature space modeling and initial model setting module: By collecting HRG output signal data under different temperature conditions, a dynamic temperature space model is constructed, and the model coefficients are dynamically adjusted using a high-order polynomial mapping function combined with the temperature change rate to achieve temperature drift characteristic modeling; Incremental adaptive neural network structure module: Design an incremental adaptive neural network structure, and combine it with a local error adjustment mechanism to achieve dynamic adjustment of the network structure and learning rate, so that the compensation model can adapt to temperature changes, and optimize the compensation accuracy through a step-by-step incremental learning method; Temperature compensation dynamic optimization module: Through the combination of dynamic temperature space model and incremental adaptive neural network, temperature compensation and incremental optimization are carried out in real time, and compensation accuracy is maintained through dynamic update of temperature space and incremental learning mechanism of incremental adaptive neural network; Adaptive feedback and correction mechanism module: monitors output signal drift and compensation error through adaptive feedback mechanism, dynamically adjusts dynamic temperature space model and incremental adaptive neural network parameters, and improves stability and compensation effect through online correction and model self-optimization mechanism.

7. A temperature compensation system for a hemispherical resonant gyro inertial navigation system according to claim 6, characterized in that: The temperature space modeling and initial model setting module includes: Obtain the output signal data of HRG under different temperature conditions, and record the temperature value and its change rate at each experimental moment; By analyzing the experimental data, a multi-dimensional dynamic temperature space model that is dynamically updated with temperature changes is constructed to adjust the drift characteristics of the HRG according to different temperatures and change rates; According to the characteristics of experimental data, considering the influence of temperature and its changing rate on HRG output, the coefficients of the polynomial are determined through regression analysis; By measuring the temperature change rate and temperature gradient in real time, the coefficients of the high-order polynomial model are dynamically adjusted to adapt to rapid or slow changes in temperature, ensuring the accuracy of temperature compensation.

8. The temperature compensation system of the hemispherical resonant gyro inertial navigation system according to claim 7, characterized in that: The incremental adaptive neural network structure module includes: Incremental learning neural network that dynamically adjusts the network structure based on real-time input signals to ensure that the temperature compensation of the HRG can adapt to changes in ambient temperature in real time; By designing a local error adjustment mechanism, the network can dynamically adjust the learning rate according to the temperature change rate and the change rate of HRG output drift, so that the compensation process can adapt to temperature changes; Through the incremental learning method of step-by-step approximation, the incremental adaptive neural network can fine-tune the network weights each time data arrives.

9. The temperature compensation system of the hemispherical resonant gyro inertial navigation system according to claim 8, characterized in that: The temperature compensation dynamic optimization module includes: By combining the dynamic temperature space model with the incremental adaptive neural network, each time the temperature changes, local modeling is first performed through the dynamic temperature space model, and then incremental optimization is performed through the incremental adaptive neural network, and finally an accurate compensation signal is output; The dynamic temperature space model updates the geometric characteristics of the temperature space in real time according to the temperature changes of the environment; The incremental learning mechanism of the incremental adaptive neural network dynamically adjusts the learning rate and network weights according to real-time error feedback to ensure that the compensation model adapts to temperature changes.

10. The temperature compensation system of the hemispherical resonant gyro inertial navigation system according to claim 9, characterized in that: The adaptive feedback and correction mechanism module includes: By real-time monitoring of the drift and compensation error of the HRG output signal and combining the characteristics of temperature changes, the parameters of the dynamic temperature space model and the incremental adaptive neural network are dynamically adjusted to ensure compensation accuracy; By accumulating historical data and optimizing the model regularly, the stability and compensation effect of the system can be gradually improved; The HRG output drift is compensated through incremental learning and feedback correction.

Images