UAV attitude estimation system based on joint calculation of IMU and Beidou system

By jointly calculating with the IMU and the BeiDou system, dynamically adjusting the noise covariance and constructing a realistic adjustment factor, the problem of lag in the response of the UAV attitude estimation system in dynamic environments is solved, and the accuracy and reliability of attitude estimation are improved.

CN123065451APending Publication Date: 2026-11-10BEIJING YOUYUN ZHIXIANG AVIATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611569723.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-12
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

In the existing technology, UAV attitude estimation systems based on the extended Kalman filter framework constructed with fixed observation noise are difficult to adapt to the dynamic changes in the flight environment, resulting in a decrease in attitude estimation accuracy and reliability, especially with lag in response to satellite signal interference and drastic changes in flight status.

Method used

By combining IMU and BeiDou system for joint calculation, inertial measurement data and BeiDou satellite observation data are obtained, the initial noise covariance is dynamically adjusted, and signal quality and interference intensity are used for fusion analysis to construct a real adjustment factor, realize attitude correction, and improve the dynamic response capability of attitude estimation.

Benefits of technology

It effectively solves the problem of filter response lag to dynamic environment, improves the dynamic response capability and reliability of attitude estimation, and ensures that attitude estimation can respond to changes in environment and flight status in a timely manner.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123065451A_ABST
    Figure CN123065451A_ABST
Patent Text Reader

Abstract

This invention relates to the field of aircraft attitude measurement technology, specifically to a UAV attitude estimation system based on joint calculation using an IMU and the BeiDou system. The system acquires signal quality data from BeiDou observations and IMU inertial measurement data; determines interference intensity based on the neighborhood distribution of the IMU data, and determines the initial noise covariance of the extended Kalman filter based on the signal quality, obtaining a novelty value through filtering analysis; then determines a fine-tuning factor based on the relative dispersion of the novelty value's neighborhood, filters reference times through similarity analysis between the current and historical times, and weights and fuses the fine-tuning factors of the reference times to obtain a true adjustment factor; finally, it uses the true adjustment factor to correct the initial noise covariance, and performs filtering analysis based on the corrected covariance to estimate the UAV attitude. This effectively solves the problem of filter response lag in dynamic environments, improving the dynamic response capability and reliability of attitude estimation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of aircraft attitude measurement technology, specifically to an unmanned aerial vehicle (UAV) attitude estimation system based on joint calculation by an IMU and the BeiDou system. Background Technology

[0002] With the increasing number of UAVs undertaking autonomous flight missions in complex environments, higher requirements are being placed on the accuracy and robustness of attitude estimation systems under satellite signal interference.

[0003] Existing technologies typically rely on an extended Kalman filter (EKF) framework for navigation, built upon a fixed set of observation noise. Specifically, this involves using high-frequency inertial measurement data from the IMU (Inertial Measurement Unit) as the basis for system state recursion and periodically correcting the data using BeiDou satellite observations to achieve UAV attitude estimation. However, due to the dynamic changes in the flight environment, the fixed observation noise cannot adaptively adjust to fluctuations in satellite signal quality and changes in flight state, leading to filter fusion weight mismatch and decreased attitude estimation accuracy. Furthermore, the volatility and uncertainty of inertial measurement data increase significantly during drastic changes in flight state. The low update frequency of the BeiDou system further exacerbates the filter's lag in responding to dynamic environments, reducing the reliability of UAV attitude estimation. Summary of the Invention

[0004] To address the technical problem that existing technologies typically rely on extended Kalman filter frameworks built with fixed observation noise for navigation, making it difficult to capture changes in flight state and leading to a decrease in the reliability of UAV attitude estimation, the present invention aims to provide a UAV attitude estimation system based on joint calculation by an IMU and the BeiDou system. The specific technical solution adopted is as follows: This invention proposes a UAV attitude estimation system based on joint calculation by IMU and BeiDou system, the system comprising: The data acquisition module is used to acquire the signal quality at each operating moment in the BeiDou satellite observation data, as well as the inertial measurement data of the IMU; The covariance analysis module is used to determine the interference intensity at each operating moment based on the neighborhood distribution of the inertial measurement data; to perform fusion analysis by combining the signal quality and interference intensity to determine the initial noise covariance at each operating moment; and to perform filtering analysis on the BeiDou satellite observation data based on the initial noise covariance to determine the innovation value at each operating moment. The adjustment analysis module is used to determine the fine-tuning factor for each running time based on the relative dispersion of the neighborhood of the innovation value; to perform a similarity analysis between each running time and all its historical running times based on the signal quality and the interference intensity, and to determine all reference running times for each running time; and to perform a weighted fusion of the fine-tuning factors of all reference running times to determine the true adjustment factor for each running time. The attitude correction module is used to correct the initial noise covariance based on the real adjustment factor and determine the target observation noise covariance at each operating moment; and to perform filtering analysis based on the target observation noise covariance to estimate the UAV attitude at each operating moment.

[0005] Furthermore, the method for obtaining the signal quality includes: Obtain the signal-to-noise ratio and position accuracy factor at each operational moment from the BeiDou satellite observation data; The signal quality at each operating moment is obtained based on the signal-to-noise ratio and the position accuracy factor; the signal quality is positively correlated with the signal-to-noise ratio and negatively correlated with the position accuracy factor.

[0006] Furthermore, the method for obtaining the interference intensity includes: Obtain the index value of each inertial index at each running time in the inertial measurement data of the IMU; Within a preset time window at each running moment, the dispersion of the values ​​of each inertial index is analyzed to determine the interference intensity at each running moment.

[0007] Furthermore, the method for obtaining the innovation value includes: Obtain pseudorange observations for each operational moment in the BeiDou satellite observation data; The pseudorange observations at each running time are input into the extended Kalman filter, and the initial noise covariance is used as the control parameter of the extended Kalman filter to perform measurement updates and output the pseudorange predictions at each running time. The information value for each running time is determined based on the deviation between the pseudorange observation value and the pseudorange prediction value.

[0008] Furthermore, the method for obtaining the fine-tuning factor includes: Obtain the coefficient of variation of signal quality within a preset time window at each running moment; Based on the degree of dispersion of the innovation value within a preset time window at each running time, the discrete characteristic value of each running time is determined; based on the coefficient of variation and the discrete characteristic value, the weighted innovation variance of each running time is determined; the weighted innovation variance is positively correlated with the discrete characteristic value and negatively correlated with the coefficient of variation. All running times within a preset time window corresponding to each running time are taken as sample times; the discrete characteristic value of the sample time corresponding to the minimum coefficient of variation is taken as the benchmark innovation variance of each running time; the average dispersion of each running time is determined based on the central tendency of the difference between the weighted innovation variances of any two sample times. Based on the baseline innovation variance and the average dispersion, a fine-tuning factor is determined for each running time step; the fine-tuning factor is positively correlated with the baseline innovation variance and negatively correlated with the average dispersion.

[0009] Furthermore, the method for obtaining the reference runtime includes: Based on the signal quality and the interference intensity, a feature vector is constructed for each operating moment; the similarity of each historical operating moment is determined according to the similarity between the feature vector of each operating moment and the feature vector of each historical operating moment; a threshold is used for filtering based on the similarity to determine all reference operating moments for each operating moment.

[0010] Furthermore, the method for obtaining the true regulating factor includes: Using the similarity of each reference runtime as a weight, the fine-tuning factors of all reference runtimes are weighted and summed to determine the true adjustment factor for each runtime.

[0011] Furthermore, the method for obtaining the target observation noise covariance includes: The product of the actual adjustment factor and the initial noise covariance is used as the target observed noise covariance for each running time.

[0012] Furthermore, the initial noise covariance is the product of the negative correlation mapping value of the signal quality, the interference intensity, and the preset reference noise variance.

[0013] Furthermore, the method for obtaining the average dispersion includes: Take any two sample time points as a sample time point pair, calculate the absolute difference of the weighted innovation variance between the two sample time points in each sample time point pair, and take it as the relative difference value of the sample time point pair; take the mean of the relative difference values ​​of all sample time point pairs as the average dispersion of each running time point.

[0014] The present invention has the following beneficial effects: This invention determines the initial noise covariance for each operational moment by analyzing the interference intensity at each moment and combining it with the signal quality of BeiDou satellite observation data. This allows the initial noise covariance of the extended Kalman filter to be adjusted in real time according to fluctuations in BeiDou signal quality and changes in IMU data, overcoming the deficiency of fixed observation noise covariance in existing technologies that cannot adapt to environmental changes. By weighted fusion of fine-tuning factors from all reference operational moments, the true adjustment factor for each operational moment is determined. Filtering experience from similar historical moments is introduced to predictively correct the current filter parameters, overcoming the lag problem in innovation statistical analysis and enabling attitude estimation to respond promptly to real-time changes in the environment and flight status. Finally, by calculating the corrected target observation noise covariance and performing filtering analysis, the UAV attitude at each operational moment is estimated, effectively solving the problem of filter response lag to dynamic environments and improving the dynamic response capability and reliability of attitude estimation. Attached Figure Description

[0015] Figure 1 This is a structural diagram of a UAV attitude estimation system based on joint calculation by IMU and BeiDou system, provided as an embodiment of the present invention. Figure 2 This is a flowchart of a method for obtaining a fine-tuning factor according to an embodiment of the present invention. Detailed Implementation

[0016] The following description, in conjunction with the accompanying drawings, details a specific scheme for an unmanned aerial vehicle (UAV) attitude estimation system based on joint calculation by an IMU and the BeiDou system, provided by this invention.

[0017] Please see Figure 1 The diagram illustrates a UAV attitude estimation system based on joint calculation by IMU and BeiDou system, according to an embodiment of the present invention. The system includes: The data acquisition module 101 is used to acquire the signal quality at each operating moment in the BeiDou satellite observation data, as well as the inertial measurement data of the IMU.

[0018] During UAV flight, BeiDou satellite signals are susceptible to multipath effects caused by obstructions from buildings, trees, etc., leading to fluctuations in signal quality or even signal loss. Simultaneously, the inertial measurement data output by the IMU (Inertial Measurement Unit) varies significantly depending on the UAV's flight status. Relying solely on data from a single sensor cannot comprehensively reflect the UAV's current environment and flight state, making accurate adjustment of filtering parameters difficult. Therefore, this invention acquires the signal quality at each operational moment from BeiDou satellite observation data, as well as the IMU's inertial measurement data.

[0019] In one specific implementation of this invention, the signal-to-noise ratio (SNR), position accuracy factor (PDOP), and pseudorange observations are used as analytical indicators for BeiDou satellite observation data; velocity and acceleration are used as analytical indicators for IMU inertial measurement data. Specifically, the SNR, PDOP, and pseudorange observations at each BeiDou sampling time are read in real time by an airborne GNSS receiver; the velocity and acceleration of the UAV at each IMU sampling time are obtained by an airborne IMU inertial measurement unit; wherein the velocity and acceleration are both three-dimensional vectors in the UAV's built-in body coordinate system.

[0020] It should be noted that the sampling frequency of the IMU is much higher than the data update frequency of the BeiDou Navigation Satellite System (typically, the IMU sampling frequency is 100Hz to 1000Hz, while the BeiDou data update frequency is 1Hz to 10Hz). In this embodiment, the IMU sampling frequency is set to 200Hz and the BeiDou data update frequency to 5Hz, meaning each BeiDou sampling interval contains 40 IMU sampling times. This can be adjusted within the above range according to the specific implementation scenario. A higher IMU sampling frequency results in more detailed capture of flight state changes and higher attitude tracking accuracy; a higher BeiDou data update frequency results in more frequent BeiDou corrections and more timely drift suppression. However, the greater the frequency difference between the two, the less likely it is to reflect real-time signal quality degradation when the flight environment changes rapidly. Therefore, to capture real-time changes in flight state with higher temporal resolution, this embodiment uses the IMU sampling time as the running time for state recursion and parameter updates, enabling a more refined reflection of the UAV's continuous motion process.

[0021] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining signal quality includes: During UAV flight, BeiDou satellite signals are easily obstructed by buildings, trees, and other obstacles, resulting in reflection and diffraction, creating multipath effects and causing deviations in the pseudorange measured by the receiver. Signal-to-noise ratio (SNR) is a direct indicator of signal strength; a lower SNR indicates more severe interference from multipath effects and other factors, leading to lower reliability of the observations. Simultaneously, the geometric distribution of visible satellites in space directly affects the accuracy of positioning calculations. The position accuracy factor reflects the quality of the satellite's geometric configuration; a smaller value indicates a more favorable spatial distribution for accurate positioning and more reliable positioning results. The former characterizes the signal quality of a single satellite, while the latter characterizes the overall positioning geometry. Both reflect the reliability of BeiDou observation data from different dimensions. Therefore, this embodiment of the invention obtains the SNR and position accuracy factor at each operational moment in the BeiDou satellite observation data.

[0022] Considering that a single indicator cannot fully characterize the reliability of BeiDou observation data, the signal-to-noise ratio (SNR) describes signal strength, but a high SNR combined with a poor satellite geometry distribution can still lead to insufficient positioning accuracy; the position accuracy factor describes satellite geometry distribution, but a good geometry distribution combined with a low SNR cannot guarantee observation quality; further considering that each BeiDou sampling time interval contains multiple operating times, this embodiment of the invention obtains the signal quality for each operating time based on the SNR and the position accuracy factor; the signal quality is positively correlated with the SNR and negatively correlated with the position accuracy factor.

[0023] In one specific implementation of this invention, the system extracts the signal-to-noise ratio (SNR) and position accuracy factor (PDOP) for each BeiDou sampling moment by accessing the data acquisition module. Considering the significant differences in dimensions and magnitudes between SNR and PDOP, this invention uses a minimum-maximum normalization algorithm to map the SNR and PDOP for each BeiDou sampling moment to the range [0,1], obtaining normalized SNR and normalized PDOP. The normalized PDOP is obtained by subtracting the normalized PDOP from the base value (set to 1 in this invention). The product of the normalized SNR and the positive PDOP is used as the signal quality for each BeiDou sampling moment. Considering that the UAV's flight distance is limited between adjacent BeiDou sampling moments, and the satellite signal propagation environment and satellite geometric distribution will not change drastically, this invention uses the zero-order hold method to analyze the signal quality for each BeiDou sampling moment. Specifically, the signal quality obtained at each BeiDou sampling moment is assigned to all operating moments from that moment to the next BeiDou sampling moment, thereby obtaining the signal quality for each operating moment.

[0024] It should be noted that, taking signal-to-noise ratio (SNR) as an example, the minimum and maximum values ​​used in the minimum-maximum normalization algorithm can be obtained by statistically analyzing the historical minimum and maximum SNR values ​​of the same UAV in historical flight operation data. If the SNR at the current BeiDou sampling time is less than the historical minimum, it is set to 0; if the SNR at the current BeiDou sampling time is greater than the historical maximum, it is set to 1. Based on the same method of obtaining the normalized SNR, simply replacing the SNR with the position accuracy factor will yield the normalized position accuracy factor.

[0025] The covariance analysis module 102 is used to determine the interference intensity at each operating moment based on the neighborhood distribution of the inertial measurement data; to perform fusion analysis by combining the signal quality and interference intensity to determine the initial noise covariance at each operating moment; and to perform filtering analysis on the BeiDou satellite observation data based on the initial noise covariance to determine the innovation value at each operating moment.

[0026] Existing technologies typically use IMU data as a source for state prediction. However, the neighborhood distribution of IMU data can reflect the drastic changes and consistency of the UAV within a short period of time. When the data fluctuations within the neighborhood are large, it indicates that the UAV's flight state is complex (such as rapid maneuvers, turns, etc.). In this case, the uncertainty of the IMU data increases significantly, and the prediction reliability decreases. Therefore, this embodiment of the invention determines the interference intensity at each operating moment based on the neighborhood distribution of the inertial measurement data, in order to quantify the impact of the current flight state on the IMU data.

[0027] Considering that the filtering analysis process requires the fusion of BeiDou satellite observation data and inertial measurement data, and that the initial noise covariance determines the filter's level of trust in the BeiDou satellite observation data, the initial noise covariance should be increased when the signal quality of the BeiDou satellite observation data is poor, thus reducing the trust in the BeiDou satellite observation data. Similarly, when the interference intensity of the inertial measurement data is high, the initial noise covariance should also be increased to avoid the filter over-reliance on unreliable inertial measurement data for correction. Based on these characteristics, this embodiment of the invention combines signal quality and interference intensity for fusion analysis to determine the initial noise covariance at each operating moment. To further verify whether the initial noise covariance can accurately reflect the true noise level of the current environment, this embodiment of the invention performs filtering analysis on the BeiDou satellite observation data based on the initial noise covariance to determine the innovation value at each operating moment, which reflects the degree of matching between the current environmental noise and the filtering parameters.

[0028] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the interference intensity includes: Considering that IMU inertial measurement data can reflect the instantaneous motion state of the UAV at each operational moment, and since the UAV's flight state is constantly changing, the inertial index values ​​differ at different times. Therefore, this embodiment of the invention obtains the index value of each inertial index at each operational moment from the IMU inertial measurement data. Furthermore, considering that the inertial index value at a single moment only reflects the motion state at that instant, it cannot determine whether the current motion is stable or rapidly changing, nor can it assess the reliability of the IMU data at that moment. The dispersion of each inertial index within its neighborhood directly reflects the stability or dynamism of the UAV's motion during that period. If the values ​​of each inertial index fluctuate significantly within the neighborhood, it indicates that the UAV is in a maneuvering state, significantly increasing the uncertainty of the IMU measurement data and indicating a high degree of interference from the motion state. Therefore, this embodiment of the invention performs a fusion analysis of the dispersion of the index values ​​of each inertial index within a preset time window at each operational moment to determine the interference intensity at each operational moment.

[0029] In one specific implementation of this invention, taking each running moment as the endpoint, all historical running moments within the nearest preset time period are selected to construct a preset time window; wherein, the preferred range of the preset time period is [3s, 5s], and this embodiment sets it to 4s, which can be adjusted within this preferred range according to the specific implementation scenario. Specifically: if the flight state changes drastically and a rapid response is required, the preset time period can be shortened to ensure that the filtering parameters track the state changes in real time and enhance the system's adaptability; if the flight state is relatively stable, the preset time period can be increased to increase the number of samples, improve statistical reliability, ensure stable output of the filtering parameters, and avoid unnecessary fluctuations. For each inertial index, within a preset time window, a covariance matrix is ​​constructed based on the variance of its index value along the X, Y, and Z axes in the body coordinate system. Since the trace of the covariance matrix represents the total variance of the data and reflects the dispersion of the data, this embodiment of the invention uses a minimum-maximum normalization algorithm to map the trace of the covariance matrix to the range [0,1] to determine the index interference intensity of each inertial index at each operating time. Finally, the product of the index interference intensities of all inertial indices at each operating time is taken as the interference intensity at each operating time. The minimum and maximum values ​​used in the minimum-maximum normalization algorithm can be obtained by statistically analyzing the historical minimum and maximum values ​​of the traces corresponding to the index values ​​of the same UAV in historical flight operation data. If the trace corresponding to the index value at the current operating time is less than the historical minimum value, it is set to 0; if the trace corresponding to the index value at the current operating time is greater than the historical maximum value, it is set to 1.

[0030] It should be noted that, in the initial stage of system startup, due to the insufficient number of previous historical running times, a complete preset time window cannot be constructed. In this embodiment of the invention, all historical running times before the current time are used as the preset time window. In particular, for the first running time, since there are no historical running times, it is not analyzed subsequently.

[0031] Preferably, in some possible implementations of the embodiments of the present invention, considering that when the signal quality of BeiDou satellite observation data is poor, the pseudorange observation value is unreliable, the fusion weight of the observation value should be reduced, i.e., the observation noise covariance should be increased; when the IMU data is affected by changes in flight state leading to a large interference intensity, the predicted state is unreliable, the correction strength of the observation value should also be reduced, i.e., the observation noise covariance should be increased; that is, the observation noise covariance is negatively correlated with signal quality and positively correlated with interference intensity. Therefore, the embodiments of the present invention use the product of the negative correlation mapping value of signal quality, interference intensity, and preset reference noise variance as the initial noise covariance for each operating moment.

[0032] In one specific implementation of this invention, the signal quality is subtracted from the base value (set to 1 in this embodiment) to obtain the negative correlation mapping value; wherein, the preset reference noise variance is the mean of the variances of all innovation values ​​within a preset time window at each running moment.

[0033] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the information value includes: The BeiDou Navigation Satellite System calculates the distance between the receiver and the satellite by measuring the propagation time of the satellite signal from transmission to reception. This distance, which includes various errors such as satellite clock error, receiver clock error, ionospheric delay, tropospheric delay, and multipath effects, is called the pseudorange observation value, which directly reflects the relative positional relationship between the receiver and the satellite. Therefore, this embodiment of the invention obtains the pseudorange observation value at each operating moment in the BeiDou satellite observation data.

[0034] The Extended Kalman Filter (EKF) requires weighted fusion of predicted and observed values ​​to obtain the optimal state estimate. During the measurement update phase, the filter maps the state vector predicted by the IMU to the observation space to calculate the pseudorange prediction. In this process, the initial noise covariance, as a control parameter, influences the calculation of the Kalman gain and determines the correction strength of the pseudorange observation for the state estimate. Therefore, in this embodiment, the pseudorange observation at each running time is input into the EKF, and the initial noise covariance is used as the control parameter of the EKF to perform the measurement update and output the pseudorange prediction for each running time.

[0035] The pseudorange observation value reflects the distance information actually measured by the Beidou satellite, while the pseudorange prediction value reflects the expected observation value calculated based on the IMU prediction state. The deviation between the two reflects the degree of matching between the current filter setting and the actual environment and flight state. Therefore, in this embodiment of the invention, the information value at each operating moment is determined based on the deviation between the pseudorange observation value and the pseudorange prediction value.

[0036] In one specific implementation of this invention, the pseudorange observation value at each running time is input into the extended Kalman filter, and the initial noise covariance is used as the diagonal element value of the observation noise covariance matrix of the extended Kalman filter to perform measurement updates, outputting the pseudorange prediction value at each running time; and the innovation value at each running time is obtained by subtracting the pseudorange prediction value from the pseudorange observation value. It should be noted that the extended Kalman filtering process is a well-known technique, and its specific implementation will not be elaborated here.

[0037] The adjustment analysis module 103 is used to determine the fine-tuning factor for each running time based on the relative dispersion of the neighborhood of the innovation value; to perform a similarity analysis between each running time and all its historical running times based on the signal quality and the interference intensity, and to determine all reference running times for each running time; and to perform weighted fusion of the fine-tuning factors of all reference running times to determine the true adjustment factor for each running time.

[0038] The relative dispersion of the innovation value within its neighborhood directly reflects the degree of matching between the current filter setting and the actual environmental noise. When the relative dispersion of the innovation value within its neighborhood is high, it indicates that the actual environmental noise level exceeds the estimated range of the current observed noise covariance, the weight allocation of the filter between prediction and observation is inappropriate, and the initial covariance needs to be corrected. Therefore, in this embodiment of the invention, the fine-tuning factor for each running time is determined based on the relative dispersion of the innovation value within its neighborhood.

[0039] Because the fine-tuning factor requires the accumulation of data within a certain time window during calculation, the obtained fine-tuning factor naturally has a lag. When the environment and flight status change, the fine-tuning factor calculated based on historical data has not yet reflected this change. To solve this problem, this embodiment of the invention performs a similarity analysis on each operating moment and all its historical operating moments based on the signal quality and interference intensity, determining all reference operating moments for each operating moment. By performing similarity analysis on all historical operating moments, the historical moment closest to the current state can be selected as a reference, thereby overcoming the lag in the calculation of the fine-tuning factor and enabling the adjustment of the filter parameters to reflect changes in the environment and flight status in a timely manner. Furthermore, considering the difference in reference value between different historical operating moments and the current moment, this embodiment of the invention performs weighted fusion of the fine-tuning factors of all reference operating moments to determine the true adjustment factor for each operating moment.

[0040] Preferably, in some possible implementations of the embodiments of the present invention, the specific process of obtaining the fine-tuning factor in the adjustment analysis module 103 includes: Please refer to... Figure 2 The diagram illustrates a flowchart of a method for obtaining a fine-tuning factor according to an embodiment of the present invention, the method comprising: S201: Obtain the coefficient of variation of signal quality within a preset time window at each running moment.

[0041] The dispersion of the innovation value characterizes the overall level of current environmental noise; however, its dispersion is affected not only by environmental noise but also by fluctuations in the quality of the BeiDou signal. When signal quality fluctuations are large, even if the environmental noise remains constant, the innovation value will exhibit a large degree of dispersion due to the instability of the observed values. The coefficient of variation reflects the relative stability of the signal quality; the smaller the coefficient of variation, the more stable the signal quality, and the more accurately the dispersion of the innovation value reflects the level of environmental noise. Therefore, to analyze this interference factor, this embodiment of the invention obtains the coefficient of variation of the signal quality within a preset time window for each operating moment.

[0042] It should be noted that the calculation of the coefficient of variation is a technique well known to those skilled in the art, and its specific implementation process will not be elaborated here; if the mean value of the signal quality is 0 during the calculation of the coefficient of variation, it indicates that the signal is damaged during this period and will not be analyzed.

[0043] S202: Determine the discrete feature value for each running time based on the degree of dispersion of the innovation value within a preset time window for each running time; determine the weighted innovation variance for each running time based on the coefficient of variation and the discrete feature value; the weighted innovation variance is positively correlated with the discrete feature value and negatively correlated with the coefficient of variation.

[0044] After excluding the influence of signal quality on environmental noise assessment, the dispersion of the innovation value directly reflects the overall level of current environmental noise. Therefore, this embodiment of the invention determines the discrete characteristic value for each operating moment based on the dispersion of the innovation value within a preset time window. To further distinguish the influence of signal quality and the dispersion of the innovation value on the current environmental noise assessment, this embodiment of the invention determines the weighted innovation variance for each operating moment based on the coefficient of variation and the discrete characteristic value. The weighted innovation variance is positively correlated with the discrete characteristic value and negatively correlated with the coefficient of variation, such that the smaller the coefficient of variation, the more stable the signal quality, and the more accurately the dispersion of the innovation value reflects the environmental noise level.

[0045] In one specific implementation of this invention, the variance of the innovation value within a preset time window at each running moment is used as the discrete feature value at each running moment; the discrete feature value is used as the numerator, the coefficient of variation is used as the denominator, and the ratio of the two is used as the weighted innovation variance at each running moment.

[0046] It should be noted that since the coefficient of variation is a dimensionless value, the dimension of the weighted innovation variance is the same as that of the discrete eigenvalues, both being m. 2When the coefficient of variation is 0, it indicates that the signal quality within the window is completely stable, and the reliability of the window cannot be effectively evaluated using the coefficient of variation. In this case, the window is marked as an invalid window and will not be included in subsequent analysis.

[0047] S203: Take all running times within the preset time window corresponding to each running time as sample times; take the discrete characteristic value of the sample time corresponding to the minimum value of the coefficient of variation as the benchmark innovation variance of each running time; determine the average dispersion of each running time based on the central tendency of the difference between the weighted innovation variances of any two sample times.

[0048] Considering that the weighted innovation variance at a single moment may deviate from the true level due to instantaneous noise and cannot fully reflect the overall state of the current environmental noise, this embodiment of the invention uses all running moments within a preset time window corresponding to each running moment as sample moments. Since the coefficient of variation reflects the relative stability of signal quality, the sample moment with the smallest coefficient of variation has the most stable signal quality, is least affected by interference such as multipath effects, and its corresponding discrete eigenvalue best represents the true environmental noise. Therefore, this embodiment of the invention uses the discrete eigenvalue of the sample moment corresponding to the minimum coefficient of variation as the benchmark innovation variance for each running moment.

[0049] If the weighted innovation variances at each sample time point are very close, it indicates that the environmental noise level remains consistent within that period, and the filter parameters are set reasonably. Conversely, if the weighted innovation variances at each sample time point differ significantly, it indicates that the environmental noise is unstable or the filter parameters deviate from actual requirements. Therefore, in order to quantify the differences in weighted innovation variances between each sample time point, this embodiment of the invention determines the average dispersion of each running time point based on the central tendency of the differences between the weighted innovation variances of any two sample time points, thereby quantifying the consistency of environmental noise within that period.

[0050] In one specific implementation of this invention, each running moment within a preset time window is taken as a sample moment. The discrete characteristic value of the sample moment corresponding to the minimum coefficient of variation is selected as the baseline innovation variance for each running moment. Any two sample moments are taken as a sample moment pair, and the absolute difference of the weighted innovation variance between the two sample moments in each sample moment pair is calculated as the relative difference value of the sample moment pair. The mean of the relative difference values ​​in all sample moment pairs is taken as the average dispersion of each running moment.

[0051] S204: Based on the baseline innovation variance and the average dispersion, determine the fine-tuning factor for each running time; the fine-tuning factor is positively correlated with the baseline innovation variance and negatively correlated with the average dispersion.

[0052] The baseline innovation variance represents the absolute level of environmental noise, while the average dispersion represents the consistency of environmental noise. Both together determine the direction and magnitude of the fine-tuning factor. A large baseline innovation variance indicates that even during periods of peak signal quality, the innovation value fluctuates significantly, suggesting that the actual noise level of the current environment is inherently high, requiring a corresponding increase in the observation noise covariance to match this noise level. Conversely, a larger average dispersion indicates significant differences in the weighted innovation variance across different sample times, resulting in drastic fluctuations in the innovation sequence. This may be due to a small initial noise covariance, causing the filter to over-respond to observation noise, leading to large fluctuations in the innovation variance at each time point. In this case, the fine-tuning factor needs to be reduced to decrease the observation noise covariance, making the filter smoother and more stable. Therefore, considering both influencing factors, this embodiment of the invention determines the fine-tuning factor for each operating time point based on the baseline innovation variance and the average dispersion. The fine-tuning factor is positively correlated with the baseline innovation variance and negatively correlated with the average dispersion.

[0053] In one specific implementation of this invention, the reference innovation variance is used as the numerator, the average dispersion is used as the denominator, and the ratio of the two is used as the fine-tuning factor, where the fine-tuning factor is a dimensionless non-negative real number. If the fine-tuning factor is less than 1, it indicates that the current initial noise covariance is too large, and the covariance needs to be reduced to increase the confidence in the BeiDou observations. When the fine-tuning factor is greater than 1, it indicates that the current initial noise covariance is too small, and the covariance needs to be increased to reduce the confidence in the BeiDou observations. When the fine-tuning factor is equal to 1, it indicates that the current covariance setting is reasonable and no adjustment is needed, thereby achieving bidirectional adaptive correction of the filter parameters.

[0054] It should be noted that the baseline innovation variance and the average dispersion have the same dimension, both being m. 2 Furthermore, the magnitudes all correspond to the squares of the pseudorange innovation values, falling within the same numerical range under the same flight scenario, allowing for direct division. When the average dispersion is 0, it indicates that the weighted innovation variance at each sample time point is completely consistent. In this case, the fine-tuning factor is set to 1 to ensure that no adjustment is made when consistency is highest, resulting in a smoother filtering result.

[0055] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the reference runtime includes: Signal quality reflects the external reliability of current BeiDou observation data, while interference intensity reflects the internal reliability of IMU prediction data. Both factors jointly determine the adaptive adjustment requirements of the current filtering parameters. Therefore, this embodiment of the invention constructs a feature vector for each operating moment based on the signal quality and interference intensity. This feature vector characterizes the external signal environment and internal motion state of the UAV at the corresponding moment, serving as the basis for subsequent screening and analysis.

[0056] After obtaining the similarity of all historical moments, in order to filter out historical moments that are truly valuable for the current moment, this embodiment of the invention determines the similarity of each historical running moment based on the similarity between the feature vector of each running moment and the feature vector of each historical running moment; based on the similarity, a threshold filtering is performed to determine all reference running moments for each running moment, retaining only historical moments whose similarity to the current moment is higher than the threshold as reference running moments. Threshold filtering can effectively exclude historical data that differs greatly from the current state, ensuring that each historical moment in the reference set has sufficient similarity to the current moment, thereby improving the reliability of the weighted fusion result.

[0057] In one specific implementation of this invention, a two-dimensional row vector is constructed based on the signal quality and interference intensity at each running moment, serving as the feature vector for each running moment. For example: the... The feature vector at each running time can be represented as: ,in Indicates the first Signal quality at each runtime moment Indicates the first The interference intensity at each running time point is calculated. For each running time point and each of its historical running times points, the Euclidean distance between their feature vectors is calculated, and this Euclidean distance is input into a Gaussian kernel function with a bandwidth parameter set to 1, thereby mapping the Euclidean distance to the range [0,1] to obtain the similarity of each historical running time point. Finally, all historical running times points with similarity greater than or equal to a preset similarity threshold are used as reference running times points for each running time point. The preferred range of the preset similarity threshold is [0.7, 0.9], and this embodiment of the invention sets it to 0.8, which can be adjusted within this preferred range according to the specific implementation scenario. If the number of reference running times points is 0, it means that there are no historical times points similar to the current state, so its true adjustment factor is directly set to 1, that is, the initial noise covariance is not corrected.

[0058] It should be noted that if the flight environment is a complex environment such as a city or canyon, due to frequent signal blockage, severe multipath effects, and drastic changes in flight status, the preset similarity threshold can be reduced to increase the diversity of reference samples. If the flight environment is a good environment such as open airspace, due to stable signal quality and relatively stable flight status, the preset similarity threshold can be increased to ensure that only historical moments highly similar to the current state are referenced, thus maintaining the stability of the filtering parameters.

[0059] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the true adjustment factor includes: Considering that higher similarity indicates a greater consistency between the external signal environment and internal motion state of the historical moment and the current moment, the fine-tuning factor has greater reference value for the current moment. Therefore, this embodiment of the invention uses the similarity of each reference running moment as a weight to perform a weighted summation of the fine-tuning factors of all reference running moments to determine the true adjustment factor for each running moment.

[0060] In one specific implementation of this invention, the product of the similarity and the fine-tuning factor is used as the weighted adjustment factor for each reference running time; the sum of the weighted adjustment factors of all reference running times is used as the numerator, the sum of the similarities of all reference running times is used as the denominator, and the ratio of the two is used as the true adjustment factor for each running time, thereby realizing the weighted summation of the fine-tuning factors of all reference running times.

[0061] The attitude correction module 104 is used to correct the initial noise covariance based on the real adjustment factor, determine the target observation noise covariance at each operating moment, and perform filtering analysis based on the target observation noise covariance to estimate the UAV attitude at each operating moment.

[0062] The initial noise covariance is determined based on the signal quality and interference intensity at the current moment, reflecting the real-time characteristics of the current environment and flight status. However, this value may deviate from the true level due to instantaneous fluctuations. The true adjustment factor, on the other hand, is obtained by weighted fusion of fine-tuning factors from similar historical moments, incorporating experience in adjusting filter parameters under similar historical conditions. It reflects the direction and degree of deviation of the current covariance from the true environmental noise level. Therefore, this embodiment of the invention corrects the initial noise covariance based on the true adjustment factor, determining the target observation noise covariance for each operating moment. This ensures that the corrected target observation noise covariance retains the real-time information of the signal quality and interference intensity at the current moment while incorporating successful filtering experience from similar historical moments, enabling a more accurate match to the current actual noise level.

[0063] In one specific implementation of this invention, the product of the actual adjustment factor and the initial noise covariance is used as the target observation noise covariance for each running time.

[0064] Since the target observation noise covariance accurately reflects the reasonable weight distribution between the observed and predicted values ​​at the current operating moment, the measurement update performed by extended Kalman filtering based on this target observation noise covariance allows the filter to fuse IMU predictions and BeiDou observations with optimal fusion weights. This avoids jitter in state estimation due to signal noise interference caused by over-reliance on BeiDou observations, and also avoids error accumulation due to over-reliance on IMU predictions, resulting in timely correction. This ensures that the state estimation results output by the filtering analysis accurately reflect the UAV's true position, velocity, and attitude information at the current moment. Therefore, this embodiment of the invention performs filtering analysis based on the aforementioned target observation noise covariance to estimate the UAV attitude at each operating moment.

[0065] In one specific implementation of this invention, the target observation noise covariance is used as the diagonal element value of the observation noise covariance matrix in the extended Kalman filter measurement update, and substituted into the Kalman gain calculation formula to calculate the Kalman gain. The predicted state is then corrected based on the Kalman gain and the innovation value to obtain the state estimate for the current operating moment. The state estimate includes the UAV's position, velocity, and attitude angle information. It should be noted that the above process is a well-known method for measurement updates using the extended Kalman filter algorithm, and its specific implementation will not be elaborated here.

[0066] In summary, this invention determines the initial noise covariance for each operational moment by analyzing the interference intensity at each time step and combining it with the signal quality of BeiDou satellite observation data. This allows the initial noise covariance of the extended Kalman filter to be adjusted in real time according to fluctuations in BeiDou signal quality and changes in IMU data, overcoming the shortcomings of existing technologies where fixed observation noise covariance cannot adapt to environmental changes. By weighted fusion of the fine-tuning factors of all reference operational moments, the true adjustment factor for each operational moment is determined. Filtering experience from similar historical moments is introduced to predictively correct the current filter parameters, overcoming the lag problem in innovation statistical analysis and enabling attitude estimation to respond promptly to real-time changes in the environment and flight status. Finally, by calculating the corrected target observation noise covariance and performing filtering analysis, the UAV attitude at each operational moment is estimated, effectively solving the problem of filter response lag to dynamic environments and improving the dynamic response capability and reliability of attitude estimation.

Claims

1. A UAV attitude estimation system based on joint calculation by IMU and BeiDou system, characterized in that, The system includes: The data acquisition module is used to acquire the signal quality at each operating moment in the BeiDou satellite observation data, as well as the inertial measurement data of the IMU; The covariance analysis module is used to determine the interference intensity at each operating moment based on the neighborhood distribution of the inertial measurement data; to perform fusion analysis by combining the signal quality and interference intensity to determine the initial noise covariance at each operating moment; and to perform filtering analysis on the BeiDou satellite observation data based on the initial noise covariance to determine the innovation value at each operating moment. The adjustment analysis module is used to determine the fine-tuning factor for each running time based on the relative dispersion of the neighborhood of the innovation value; to perform a similarity analysis between each running time and all its historical running times based on the signal quality and the interference intensity, and to determine all reference running times for each running time; and to perform a weighted fusion of the fine-tuning factors of all reference running times to determine the true adjustment factor for each running time. The attitude correction module is used to correct the initial noise covariance based on the real adjustment factor and determine the target observation noise covariance at each operating moment; and to perform filtering analysis based on the target observation noise covariance to estimate the UAV attitude at each operating moment.

2. The UAV attitude estimation system based on joint calculation of IMU and BeiDou system according to claim 1, characterized in that, The method for obtaining the signal quality includes: Obtain the signal-to-noise ratio and position accuracy factor at each operational moment from the BeiDou satellite observation data; The signal quality at each operating moment is obtained based on the signal-to-noise ratio and the position accuracy factor; the signal quality is positively correlated with the signal-to-noise ratio and negatively correlated with the position accuracy factor.

3. The UAV attitude estimation system based on joint calculation by IMU and BeiDou system according to claim 1, characterized in that, The method for obtaining the interference intensity includes: Obtain the index value of each inertial index at each running time in the inertial measurement data of the IMU; Within a preset time window at each running moment, the dispersion of the values ​​of each inertial index is analyzed to determine the interference intensity at each running moment.

4. The UAV attitude estimation system based on joint calculation of IMU and BeiDou system according to claim 1, characterized in that, The method for obtaining the new information value includes: Obtain pseudorange observations for each operational moment in the BeiDou satellite observation data; The pseudorange observations at each running time are input into the extended Kalman filter, and the initial noise covariance is used as the control parameter of the extended Kalman filter to perform measurement updates and output the pseudorange predictions at each running time. The information value for each running time is determined based on the deviation between the pseudorange observation value and the pseudorange prediction value.

5. The UAV attitude estimation system based on joint calculation by IMU and BeiDou system according to claim 3, characterized in that, The method for obtaining the fine-tuning factor includes: Obtain the coefficient of variation of signal quality within a preset time window at each running moment; Based on the degree of dispersion of the innovation value within a preset time window at each running time, the discrete characteristic value of each running time is determined; based on the coefficient of variation and the discrete characteristic value, the weighted innovation variance of each running time is determined; the weighted innovation variance is positively correlated with the discrete characteristic value and negatively correlated with the coefficient of variation. All running times within a preset time window corresponding to each running time are taken as sample times; the discrete characteristic value of the sample time corresponding to the minimum coefficient of variation is taken as the benchmark innovation variance of each running time; the average dispersion of each running time is determined based on the central tendency of the difference between the weighted innovation variances of any two sample times. Based on the baseline innovation variance and the average dispersion, a fine-tuning factor is determined for each running time step; the fine-tuning factor is positively correlated with the baseline innovation variance and negatively correlated with the average dispersion.

6. The UAV attitude estimation system based on joint calculation by IMU and BeiDou system according to claim 1, characterized in that, The method for obtaining the reference runtime includes: Based on the signal quality and the interference intensity, a feature vector is constructed for each operating moment; the similarity of each historical operating moment is determined according to the similarity between the feature vector of each operating moment and the feature vector of each historical operating moment; a threshold is used for filtering based on the similarity to determine all reference operating moments for each operating moment.

7. The UAV attitude estimation system based on joint calculation by IMU and BeiDou system according to claim 6, characterized in that, The method for obtaining the true regulating factor includes: Using the similarity of each reference runtime as a weight, the fine-tuning factors of all reference runtimes are weighted and summed to determine the true adjustment factor for each runtime.

8. The UAV attitude estimation system based on joint calculation by IMU and BeiDou system according to claim 1, characterized in that, The method for obtaining the target observation noise covariance includes: The product of the actual adjustment factor and the initial noise covariance is used as the target observed noise covariance for each running time.

9. The UAV attitude estimation system based on joint calculation by IMU and BeiDou system according to claim 1, characterized in that, The initial noise covariance is the product of the negative correlation mapping value of the signal quality, the interference intensity, and the preset reference noise variance.

10. The UAV attitude estimation system based on joint calculation of IMU and BeiDou system according to claim 5, characterized in that, The method for obtaining the average dispersion includes: Take any two sample time points as a sample time point pair, calculate the absolute difference of the weighted innovation variance between the two sample time points in each sample time point pair, and take it as the relative difference value of the sample time point pair; take the mean of the relative difference values ​​of all sample time point pairs as the average dispersion of each running time point.