A polarization smearing error estimation method based on meta-learning in a high dynamic environment

Abstract

The application provides a polarization tailing error estimation method based on meta-learning in a high dynamic environment, and belongs to the field of bionic polarization navigation, and comprises the following steps: a polarization angle error database caused by tailing interference under different rotating speeds is established; in an offline stage, a rotating speed related network and a rotating speed independent network are trained based on a meta-learning strategy, polarization angle errors are decoupled into rotating speed related features and rotating speed independent features; in an online stage, when performing linear motion, a polarization / inertial navigation correction dynamic model three-dimensional attitude error is used; in turning motion, the rotating speed related features are taken as states, the polarization angle inverted by a carrier dynamic model is taken as a measurement, the rotating speed independent features are taken as a measurement matrix, the rotating speed related features are estimated, the rotating speed related features and the rotating speed independent features are fused, and the estimation of the polarization tailing error is completed. The application combines offline and online learning to complete adaptive estimation and compensation of the polarization tailing error under different rotating speeds in a high dynamic environment, and improves the navigation performance of a bionic polarization integrated navigation system.

Description

Technical Field

[0001] This invention belongs to the field of biomimetic polarization navigation, specifically involving a polarization tail error estimation method based on meta-learning in high dynamic environments. It realizes adaptive estimation and compensation of polarization angle error under different speed tail interference environments, thereby improving the navigation performance of the biomimetic polarization integrated navigation system in turning maneuvering environments. Background Technology

[0002] Many organisms in nature, such as bees, dragonflies, and monarch butterflies, can sense polarized light in the sky for autonomous navigation and positioning. Inspired by this biological navigation mechanism, biomimetic polarization navigation has received widespread attention in recent years. It boasts advantages such as immunity to electromagnetic interference and high dynamic range, enabling passive, autonomous navigation. The polarization angle is calculated by integrating the polarization sensor over one sampling period. However, during high-speed maneuvers of the carrier, the sampling frequency of the polarization sensor is lower than the change in the carrier's angular velocity, resulting in a delayed response and polarization angle trailing error, which reduces the performance of biomimetic polarization navigation. Therefore, compensating for the polarization angle trailing error during turning maneuvers is the core key to improving the performance and environmental adaptability of biomimetic polarization navigation.

[0003] The invention "A Multi-Source Error Calibration Method for a Biomimetic Polarization Sensor Based on Adaptive UKF" (Application No.: CN201811414147.X) establishes a system state model based on installation error, scaling factor, polarization degree coefficient, polarization angle, and polarization degree as state variables of the polarization navigation system. It establishes a measurement model of the polarization navigation system using light intensity measurements containing multi-source errors as output. An adaptive unscented Kalman filter is designed to estimate installation error, scaling factor, polarization angle, and polarization degree, thereby estimating and compensating for polarization angle errors under multi-source error conditions. The invention "An Inertial Navigation / Polarization Combined Navigation Method Based on Non-Rayleigh Scattering Model Error" (Application No.: CN202111417764.7) introduces non-Rayleigh scattering model errors and jointly constructs an extended-dimensional system state vector with the strapdown inertial navigation error state. Simultaneously, it incorporates modeling of polarization angle errors caused by the non-Rayleigh scattering model into the measurement equations, achieving online estimation and compensation of polarization angle errors under the non-Rayleigh scattering model, thus improving the system's adaptability to adverse weather conditions. Furthermore, existing technologies address anomalies caused by large maneuvers by designing an adaptive random filtering method that considers measurement correlation entropy based on the time-varying non-Gaussian nature of polarization navigation measurement noise, thereby improving the system's anti-interference capability under large maneuvers.

[0004] The existing polarization sensor models described above consider the influence of various environmental factors on the polarization sensor. However, they do not account for the impact of the tailing error generated by the polarization sensor during high-dynamic turning maneuvers. This leads to a significant increase in polarization angle error during turning maneuvers, resulting in a substantial decrease in the accuracy of heading and attitude estimation. Therefore, how to estimate and compensate for polarization tailing errors at different speeds to improve the attitude accuracy of biomimetic polarization navigation urgently needs further research. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention proposes a meta-learning-based method for estimating polarization tail error in high-dynamic environments. A database of polarization angle errors caused by tail interference at different rotational speeds is established, including angular velocity, acceleration, polarized light intensity, and polarization angle error. In the offline phase, a rotational speed-dependent network and a rotational speed-independent network are trained separately based on a meta-learning strategy to decouple the polarization angle error into rotational speed-dependent and rotational speed-independent features. In the online phase, during the carrier's linear motion, a combined navigation system using a polarization sensor and inertial navigation is first employed. Subsequently, the estimated heading and horizontal attitude are used to correct the dynamically derived three-dimensional attitude. During the carrier's turning motion, rotational speed-dependent features are used as the state, the polarization angle derived from the carrier's dynamics is used as the measurement, and rotational speed-independent features are used as the measurement to complete the online correction of rotational speed-dependent features. Finally, the rotational speed-dependent and rotational speed-independent features are fused to achieve adaptive estimation and compensation of polarization angle error under polarization tail interference conditions.

[0006] To achieve the above objectives, the technical solution adopted by this invention is as follows: a polarization tailing error estimation method based on meta-learning in a high dynamic environment, comprising the following steps:

[0007] Step 1: Establish a database of polarization angle errors at different rotational speeds, including angular velocity. acceleration Polarized light intensity and polarization angle error ,in These are the components of the angular velocity along the three axes. These are the components of acceleration along the three axes. The light intensity of a certain channel, Number of channels;

[0008] Step 2: Build a speed-related network based on an offline meta-learning strategy. Rotation speed independent network , with angular velocity acceleration and polarized light intensity As input to the network, contrastive learning loss is used. Generate counter-loss and reconstruction losses Constrain and train and The network will reduce the polarization angle error. Decoupling into speed-dependent features Rotational speed-independent characteristics , where m is the dimension of rotational speed-related and unrelated features;

[0009] Step 3, Online Phase: When the carrier is moving in a straight line, the polarization / inertial navigation system is used to estimate the carrier's three-dimensional attitude and correct the three-dimensional attitude of the carrier's dynamic model, using the attitude error of the carrier's dynamic model. and angular velocity error As a state Establish the system state equations The heading angle obtained by polarization / inertial navigation fusion Roll angle and pitch angle As a measurement, a measurement equation is established. Correcting the three-dimensional attitude of the carrier dynamics model , and Indicates and Related functions, Indicates and Relevant measurement values; during vehicle maneuvering and turning, based on the dynamic model and speed-independent characteristics, the speed-related characteristics are corrected online, and the speed-related characteristics are adjusted accordingly. as a state vector Establish the system state equations Based on rotational speed independent characteristics Polarization angle error derived from carrier dynamics model inversion Establish system measurement equations To achieve speed-related features Online adaptive updates yield corrected speed-related features. , and Indicates and Related functions, Indicates and Relevant measurement values;

[0010] Step 4, and The coupling completes the estimation and compensation of polarization angle tailing error.

[0011] The advantages of this invention compared to the prior art are as follows:

[0012] This invention proposes a meta-learning-based method for estimating polarization tail error in high-dynamic environments. It achieves adaptive estimation and compensation of polarization angle error under tail interference conditions with different rotational speeds, improving the navigation performance of the biomimetic polarization integrated navigation system in turning maneuvers. Based on a meta-learning strategy, this invention decouples the rotational speed-dependent and rotational speed-independent features of the polarization angle error in the offline stage. In the online stage, it constructs a rotational speed-dependent feature model, completing the adaptive estimation and compensation of polarization angle error under tail interference conditions with different rotational speeds, thus improving the navigation performance of the biomimetic polarization integrated navigation system in turning maneuvers. Attached Figure Description

[0013] Figure 1 This invention provides a method for estimating polarization tailing error based on meta-learning in a high dynamic environment.

[0014] Figure 2 This is a simulation trajectory setting diagram of an embodiment of the present invention;

[0015] Figure 3 This is a simulated angular velocity diagram of an embodiment of the present invention;

[0016] Figure 4 This is a simulation heading angle diagram of an embodiment of the present invention;

[0017] Figure 5 This is a simulation roll angle diagram of an embodiment of the present invention;

[0018] Figure 6 The results show the comparison of heading angle errors. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0020] The polarization angle is calculated by integrating the polarization sensor over one sampling period. However, during turning maneuvers, the calculation error of the polarization angle is amplified due to rapid environmental changes, thus affecting the navigation accuracy of the polarization sensor. Therefore, as... Figure 1 As shown, the present invention proposes a polarization tailing error estimation method based on meta-learning in a high dynamic environment. The specific implementation steps are as follows:

[0021] Step 1: Establish a database of polarization angle errors at different rotational speeds, including the angular velocity output by the gyroscope. Accelerometer output acceleration The intensity of polarized output light And the polarization angle error obtained based on the reference system. ,in These are the components of the angular velocity along the three axes. These are the components of acceleration along the three axes. The light intensity of a certain channel of the polarization sensor. This represents the number of channels in the polarization sensor.

[0022] Step 2: Build a speed-related network based on an offline meta-learning strategy. Rotation speed independent network , with angular velocity acceleration and polarized light intensity As input to the network, contrastive learning loss is used. Generate counter-loss and reconstruction losses Constrain and train and The network will reduce the polarization angle error. Decoupling into speed-dependent features Rotational speed-independent characteristics , where m is the dimension of rotational speed-related and unrelated features;

[0023] Step 3, Online Phase: When the carrier is moving in a straight line, the polarization / inertial navigation system is used to estimate the carrier's three-dimensional attitude and correct the three-dimensional attitude of the carrier's dynamic model, using the attitude error of the carrier's dynamic model. and angular velocity error As a state Establish the system state equations The heading angle obtained by polarization / inertial navigation fusion Roll angle and pitch angle As a measurement, a measurement equation is established. Correcting the three-dimensional attitude of the carrier dynamics model , and Indicates and Related functions, Indicates and Relevant measurement values; during vehicle maneuvering and turning, based on the dynamic model and speed-independent characteristics, the speed-related characteristics are corrected online, and the speed-related characteristics are adjusted accordingly. as a state vector Establish the system state equations Based on rotational speed independence characteristics Polarization angle error derived from carrier dynamics model inversion Establish system measurement equations To achieve speed-related features Online adaptive updates yield corrected speed-related features. , and Indicates and Related functions, Indicates and Relevant measurement values;

[0024] Step 4, and The coupling process completes the estimation and compensation of polarization angle tailing error. By compensating for the polarization angle error under tailing interference, the fusion accuracy of polarization / inertial navigation in a maneuvering environment can be effectively improved.

[0025] Specifically, in step 1:

[0026] Solar Vector polarization vector under system b The relationship can be represented as:

[0027] (1)

[0028] Among them, the solar vector The polarization vector under system b can be calculated using an astronomical almanac. It can be represented as , The attitude matrix can be calculated from the heading angle and horizontal attitude angle provided by the reference, from which the polarization angle can be solved. Combined with the polarization angle measured by the polarization sensor This allows us to obtain the label for the polarization angle error. , can be represented as:

[0029] (2)

[0030] This allows us to establish a database of polarization angle errors at different rotational speeds.

[0031] Specifically, in step 2,

[0032] and Both are based on the angular velocity output by the gyroscope. Accelerometer output acceleration The intensity of polarized output light As input, and The outputs are speed-related features. Rotational speed-independent characteristics Therefore, the polarization angle error can be decoupled as and The above process can be represented as:

[0033] (3)

[0034] Speed-related network Rotation speed independent network The loss function can be expressed as:

[0035] (4)

[0036] in, Represent the coefficients of reconstruction loss, contrastive learning loss, and generative adversarial loss, respectively. Reconstruction loss This can be expressed as:

[0037] (5)

[0038] in, Given the 1-norm, the contrast learning loss This can be expressed as:

[0039] (6)

[0040] in, For positive samples, the loss can be expressed as:

[0041] (7)

[0042] in, It is a 2-norm. and It comes from the same rotational speed The following data and , The negative sample loss can be expressed as:

[0043] (8)

[0044] in, and It comes from different speeds and The following data. Contrastive learning loss can be improved by bringing together speed-related features at the same engine speed and distancing them from those at different engine speeds. The ability to extract rotational speed-related features. Generating adversarial loss. This can be expressed as:

[0045] (9)

[0046] in, For discriminator, The data comes from zero speed. It comes from non-zero speed The following data is used to generate adversarial losses. By making the discriminator unable to distinguish speed-independent features from zero-speed and non-zero-speed data, it is possible to promote... The ability to extract speed-independent features. Therefore, it is possible to train a system capable of accurately extracting both speed-independent and relevant features. and .

[0047] Specifically, in step 3:

[0048] When the carrier moves in a straight line, the three-dimensional attitude of the carrier is estimated using polarization / inertial navigation combined navigation. This is used to correct the three-dimensional attitude of the carrier dynamics model. The system state equation of the dynamics model can be expressed as:

[0049] (10)

[0050] in, , The reference angle for the three-axis attitude error. This refers to the triaxial angular velocity error. Noise representing the state equation, The state transition matrix can be represented as:

[0051] (11)

[0052] in, It is the inertia matrix. Represents a diagonal matrix, indicating the three-dimensional attitude using polarization / inertial navigation integrated navigation. The measurement model of the modified dynamic model can be expressed as:

[0053] (12)

[0054] Therefore, polarization / inertial navigation combined with navigation is used to correct the recursive error of the dynamic model on straight segments, so as to ensure that the dynamic model can derive accurate three-dimensional attitude. .

[0055] When the vehicle is turning, the system's state equation can be expressed as:

[0056] (13)

[0057] in, , Noise representing the state equation, The state transition matrix can be represented as:

[0058] (14)

[0059] in, As the attenuation coefficient, the measurement equation can be expressed as:

[0060] (15)

[0061] This allows us to obtain the corrected speed-related characteristics. .

[0062] Specifically, step 4 includes: and The coupling process completes the estimation and compensation of polarization angle tailing error. By compensating for the polarization angle error under tailing interference, the fusion accuracy of polarization / inertial navigation in a maneuvering environment can be effectively improved.

[0063] The polarization angle that compensates for the tailing error can be expressed as:

[0064] (16)

[0065] in, The polarization angle is measured by the polarization sensor. To compensate for the polarization angle after trailing interference. By compensating for the polarization angle error under trailing interference, the fusion accuracy of polarization / inertial navigation in a maneuvering environment can be effectively improved.

[0066] Example:

[0067] The simulated trajectory consists of 8 maneuvers, each including: a leftward roll at 0.05° / s for 2.5 seconds; a leftward turn at a specific speed for 10 seconds; a leftward roll at 0.05° / s for 2.5 seconds; uniform linear motion for 5 seconds; a leftward roll at 0.05° / s for 2.5 seconds; a rightward turn at a specific speed for 10 seconds; a leftward roll at 0.05° / s for 2.5 seconds; and uniform linear motion for 5 seconds. The speeds for each maneuver are 1 / 7 / 2.5 / 11.5 / 16 / 20.5 / 23.5 / 25° / s, respectively. The accelerometer bias is 10ug, and the velocity exhibits random walk. Gyroscope zero bias 10° / h, angle random walk The polarized light intensity is increased by uniformly distributed noise with an amplitude of 10 Lux. For example... Figure 2 The diagram shown is a simulation trajectory setting diagram for this embodiment. Figure 3 The figure shown is a simulated angular velocity diagram of this embodiment. Figure 4 This is the simulated heading angle diagram for this embodiment. Figure 5 This is a simulated roll angle diagram for this embodiment.

[0068] Using the method of this invention, in the offline stage, the polarization angle error is decoupled into rotational speed-dependent features and rotational speed-independent features based on a meta-learning strategy. In the online stage, when the carrier is moving in a straight line, the three-dimensional attitude error of the carrier dynamics model is corrected using polarization / inertial navigation fusion. During turning motion, the rotational speed-dependent features are estimated using the polarization angle and rotational speed-independent features derived from carrier dynamics inversion, thus completing the adaptive estimation and compensation of polarization angle error under polarization tailing interference. A comparison is made between not compensating for tailing error, using only offline stage compensation, and the compensation method of this invention.

[0069] The comparison results of heading angle error are as follows Figure 6 As shown, the straight line represents the uncompensated tail error, the circled area represents compensation using offline learning, and the crossed area represents the results of compensation using both offline and online learning. It is clear that offline learning effectively compensates for the tail error, while adding online learning stabilizes the heading error around 0. Quantitative analysis shows that the mean absolute error (RMSE) of the heading angle without compensation is 0.60°, the RMSE after offline learning compensation is 0.24°, an improvement of 60.0%, and the RMSE after both offline and online learning compensation is 0.10°, an improvement of 58.3% compared to offline learning compensation alone. Thus, the polarization angle error estimation based on meta-learning under tail interference conditions is completed.

[0070] Although the illustrative specific embodiments of the present invention have been described above to enable those skilled in the art to understand the invention, it should be understood that the invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes will be obvious as long as they are within the spirit and scope of the invention as defined and determined by the appended claims, and all inventions utilizing the concept of the present invention are protected.

Claims

1. A polarization tailing error estimation method based on meta-learning in high dynamic environments, characterized in that, Includes the following steps: Step 1: Establish a database of polarization angle errors at different rotational speeds, including angular velocity. acceleration Polarized light intensity and polarization angle error ,in These are the components of the angular velocity along the three axes. These are the components of acceleration along the three axes. The light intensity of a certain channel, Number of channels; Step 2: Build a speed-related network based on an offline meta-learning strategy. Rotation speed independent network , with angular velocity acceleration and polarized light intensity As input to the network, contrastive learning loss is used. Generate counter-loss and reconstruction losses Constrain and train and The network will reduce the polarization angle error. Decoupling into speed-dependent features Rotational speed-independent characteristics , where m is the dimension of rotational speed-related and unrelated features; Step 3, Online Phase: When the carrier is moving in a straight line, the polarization / inertial navigation system is used to estimate the carrier's three-dimensional attitude and correct the three-dimensional attitude of the carrier's dynamic model, using the attitude error of the carrier's dynamic model. and angular velocity error As a state Establish the system state equations The heading angle obtained by polarization / inertial navigation fusion Roll angle and pitch angle As a measurement, a measurement equation is established. Correcting the three-dimensional attitude of the carrier dynamics model , and Indicates and Related functions, Indicates and Relevant measurement values; during vehicle maneuvering and turning, based on the dynamic model and speed-independent characteristics, the speed-related characteristics are corrected online, and the speed-related characteristics are adjusted accordingly. as a state vector Establish the system state equations Based on rotational speed independent characteristics Polarization angle error derived from carrier dynamics model inversion Establish system measurement equations To achieve speed-related features Online adaptive updates yield corrected speed-related features. , and Indicates and Related functions, Indicates and Relevant measurement values; Step 4, and The coupling completes the estimation and compensation of polarization angle tailing error.

2. The polarization tailing error estimation method based on meta-learning in a high dynamic environment according to claim 1, characterized in that: In step (1), a polarization angle error database is established at different rotational speeds, including the angular velocity output by the gyroscope. Accelerometer output acceleration The intensity of polarized output light And the polarization angle error obtained based on the reference system. ,in, The light intensity of a certain channel of the polarization sensor. This represents the number of channels in the polarization sensor. Solar Vector polarization vector under system b The relationship is represented as: (1) Among them, the solar vector Using astronomical almanac calculations, the polarization vector under system b is... Represented as , The attitude matrix is ​​calculated from the heading angle and horizontal attitude angle provided by the reference system, from which the polarization angle is solved. Combined with the polarization angle measured by the polarization sensor Thus, the polarization angle error is obtained. , is represented as: (2) Therefore, a database of polarization angle errors under different rotational speeds was established.

3. The polarization tailing error estimation method based on meta-learning in a high dynamic environment according to claim 2, characterized in that: In step (2), and Both are based on the angular velocity output by the gyroscope. Accelerometer output acceleration The intensity of polarized light output As input, and The outputs are speed-related features. Rotational speed-independent characteristics Therefore, the polarization angle error is decoupled as and The above process can be represented as follows: (3)。 4. The polarization tailing error estimation method based on meta-learning in a high dynamic environment according to claim 1, characterized in that: Speed-related network Rotation speed independent network The loss function is expressed as: (4) in, Represent the coefficients of reconstruction loss, contrastive learning loss, and generative adversarial loss, respectively. Reconstruction loss Expressed as: (5) in, Given the 1-norm, the contrast learning loss Expressed as: (6) in, For positive samples, the loss is expressed as: (7) in, and It comes from the same rotational speed The following data and , The negative sample loss is expressed as: (8) in, It is a 2-norm. and It comes from different speeds and The following data; Generate adversarial loss Expressed as: (9) in, For discriminator, The data comes from zero speed. It comes from non-zero speed Based on the data below, a model for extracting rotational speed-independent and correlated features is trained. and .

5. The polarization tailing error estimation method based on meta-learning in a high dynamic environment according to claim 3, characterized in that: In step (3), when the carrier moves in a straight line, the three-dimensional attitude of the carrier is estimated using polarization / inertial navigation combined navigation. This is used to correct the three-dimensional attitude of the carrier dynamics model, and the system state equation of the dynamics model is expressed as: (10) in, , The three-axis attitude error angles, This refers to the triaxial angular velocity error. Noise representing the state equation, Let be the state transition matrix.

6. The polarization tailing error estimation method based on meta-learning in a high dynamic environment according to claim 5, characterized in that: State transition matrix Represented as: (11) in, It is the inertia matrix. This represents a diagonal matrix.

7. The polarization tailing error estimation method based on meta-learning in a high dynamic environment according to claim 6, characterized in that: Three-dimensional attitude using polarization / inertial navigation The measurement model of the modified dynamic model is expressed as follows: (12) Therefore, by correcting the recursive error of the dynamic model through polarization / inertial navigation combined with navigation on straight segments, the dynamic model can be used to derive the three-dimensional attitude. .

8. The polarization tailing error estimation method based on meta-learning in a high dynamic environment according to claim 7, characterized in that: When the vehicle is turning, the system's state equation can be expressed as: (13) in, , Noise representing the state equation, Let be the state transition matrix.

9. The polarization tailing error estimation method based on meta-learning in a high dynamic environment according to claim 8, characterized in that: State transition matrix Represented as: (14) in, The attenuation coefficient; The measurement equation is expressed as: (15) Thus, the corrected speed-related characteristics are obtained. .

10. The polarization tailing error estimation method based on meta-learning in a high dynamic environment according to claim 9, characterized in that: In step (4), the polarization angle for compensating for tailing error is expressed as: (16) in, The polarization angle is measured by the polarization sensor. To compensate for the polarization angle after trailing interference.

Images