A deep space autonomous navigation method and system based on sensor combination optimization

Abstract

The application relates to the technical field of deep space exploration autonomous navigation, and discloses a deep space autonomous navigation method and system based on sensor combination optimization. The method comprises the following steps: acquiring and preprocessing measurement data of a multi-source heterogeneous system, determining a sensor combination according to an optimization objective function value; performing distributed bottom-layer preprocessing on measurement data of a geometry and kinematics sensitive sensor and a sun sensitive sensor, inputting local state estimation values and covariance matrices obtained into a main filter together with measurement data of a space physical field sensitive sensor, and performing centralized global joint estimation combined with evolutionary orbit dynamics constraints; taking an initial reference short-arc orbit as priori, calibrating physical field parameters according to an observability condition, and updating related models; performing consistency checking on updated theoretical measurement data and actual measurement data; outputting a navigation state and physical field parameters when a preset allowable threshold is met; otherwise, triggering a fault-tolerant isolation mechanism. The application realizes autonomous navigation and physical field parameter inversion in a complex deep space environment.

Description

Technical Field

[0001] This application relates to the field of autonomous navigation technology for deep space exploration, and in particular to a method and system for autonomous navigation in deep space based on sensor combination optimization. Background Technology

[0002] Deep space exploration missions are characterized by long orbital periods, large communication delays, and limited ground-based telemetry and control resources. Therefore, highly autonomous and precise onboard navigation capabilities are crucial for ensuring the safe operation and mission execution of the probe. To adapt to the extremely complex space environment, modern deep space probes typically no longer rely on single-sensor systems. Instead, they widely employ multi-source heterogeneous sensor fusion navigation systems composed of inertial measurement units (IMUs), star sensors, solar sensors, visible / infrared cameras, and physical field measurement instruments (such as magnetometers and gravitational gradiometers).

[0003] However, existing deep-space multi-source information fusion navigation technologies still face the following problems in practical applications in the extreme and unknown environments of deep space: First, there is a lack of an autonomous optimization mechanism for sensor measurement combinations in dynamic and complex space environments. During deep space flight, the space geometry of the probe, external illumination, and the gradient of the space physical field change drastically. For example, when the probe enters the shadow region of a celestial body, faces weak reflections, or is subjected to strong direct sunlight, the optical imaging quality of the target celestial body may be severely degraded or even temporarily malfunction; the star sensor's ability to extract star points is greatly limited when there is strong glare or obstructed field of view; and the signal-to-noise ratio of field sensors such as gravitational gradiometers and magnetometers is also highly correlated with the distance of the probe from the celestial body surface and the local physical field gradient. Existing fusion navigation methods usually adopt static, fixed sensor configurations or passive channel fault detection strategies, with all available sensors participating in data fusion in a normalized manner. It is impossible to dynamically quantify and weigh the system observability contribution of each sensor and the cost of onboard computing power and power consumption based on the current space state. This not only consumes valuable onboard computing and energy resources, but also easily introduces low signal-to-noise ratio observation information that is on the verge of degradation into the system, causing unnecessary noise pollution and even reducing the accuracy of the navigation reference.

[0004] Second, it is difficult to achieve a balance between high accuracy and low computational cost in multi-source heterogeneous information fusion architectures. Existing multi-source fusion algorithms are mainly divided into centralized and distributed types. Although centralized filtering can theoretically maintain the complete cross-covariance between states and approximate the global optimal estimate, due to the heterogeneous sampling frequency of deep space sensors and the extremely high dimensionality of states and measurements, the computational cost of matrix inversion and covariance update increases exponentially with the number of sensors. The limited computing power of spaceborne chips in deep space cannot meet its real-time requirements, and its single filter structure has extremely poor fault tolerance and isolation capabilities. Although traditional distributed filtering reduces single-point computing power and provides fault isolation through sub-filter decoupling, in the highly dynamic and nonlinearly coupled environment of deep space, local sub-filters often suffer from insufficient information utilization or inaccurate dynamic models, resulting in a global synthesis accuracy that is significantly worse than that of centralized filtering.

[0005] Third, the weak prior model of the target astrophysical field and the strong coupling between the probe's state and the target object's physical field restrict deep-space environment exploration and precise orbit determination. In missions to explore unknown or poorly explored deep-space objects such as Neptune, the physical field parameters of the target object, such as the gravitational field harmonic coefficients, internal mass distribution, and magnetic field topology, often have extremely low prior accuracy or are even unknown. In this case, the evolution of the probe's orbital state is highly dependent on the astrophysical field model, while the inversion of physical field parameters must be based on the precise spatial pose of the probe. The two are mutually input and constraint, forming a strongly nonlinear and strongly coupled system. Traditional navigation methods treat the physical field model as a known nominal input, which can easily lead to distortion of dynamics and divergence in state estimation when the prior model is distorted. If all unknown physical field parameters are directly augmented into the state vector for simultaneous estimation, numerical singularities or filter collapse may occur in the initial stage of the mission due to insufficient initial observability and extremely high nonlinearity.

[0006] In summary, how to dynamically and autonomously select the optimal subset of sensor measurements in real time based on system observability and on-board resource constraints, and effectively achieve decoupled and collaborative estimation of deep space state and unknown physical field parameters under a lightweight and high-precision two-level fusion framework, is a key problem that urgently needs to be solved in the field of deep space autonomous navigation. Summary of the Invention

[0007] The purpose of this application is to overcome the problems of large computational load in centralized deep space multi-source information fusion navigation, insufficient accuracy of traditional distributed systems, and the strong coupling between spacecraft motion state and physical field parameters leading to initial unobservability or filter divergence when the prior accuracy of unknown astrophysical fields is limited. Therefore, this application provides a deep space autonomous navigation method and system based on sensor combination optimization.

[0008] To address the aforementioned technical problems, this application provides a deep-space autonomous navigation method based on sensor combination optimization, comprising: Acquire measurement data of a multi-source heterogeneous system and preprocess it, calculate the optimization objective function value of each candidate sensor combination, and determine the candidate sensor combination with the largest optimization objective function value as the sensor combination; The measurement data of the geometry and kinematics sensors and the solar sensor in the sensor array are preprocessed in a distributed low-level manner to obtain the local state estimates and their covariance matrices of each branch. The local state estimates and their covariance matrices of each branch and the measurement data of the space physics field sensor are input into the main filter. Based on the augmented state vector, evolutionary orbital dynamics constraints and strongly coupled measurement model, global elastic fusion is performed in the form of information filtering to perform centralized global joint estimation. Using the initial reference short-arc orbit in the global joint estimation as a priori, the physical field parameters are released and calibrated step by step according to the current observability conditions to obtain the error increments of the gravitational field parameters and the magnetic field parameters. The orbital dynamics model is updated based on the error increments of the gravitational field parameters, and the magnetic field measurement prediction model is updated based on the error increments of the magnetic field parameters to obtain the updated theoretical measurement data. The updated theoretical measurement data and actual measurement data are subjected to consistency verification. When the verification value of consistency verification is less than or equal to the preset tolerance threshold, the detector navigation status and the final calibrated physical field parameters are output. When the verification value of consistency verification is greater than the preset tolerance threshold, the fault tolerance isolation mechanism is triggered, the measurement noise covariance and fusion weight of the current physical field measurement data are adjusted, and global joint estimation and consistency verification are performed based on subsequent measurement data.

[0009] According to the present invention, a deep space autonomous navigation method based on sensor combination optimization is provided, wherein the acquisition and preprocessing of measurement data from a multi-source heterogeneous system are described, and the preprocessing specifically includes: The measurement data of the multi-source heterogeneous system are analyzed using a state-space model, the expression of which is: ; in, The first derivative of the system state vector with respect to time. It is a vector-valued function. Let be the system state vector. For the system input vector, For the first The measurement output of each sensor, It is a nonlinear measurement function. and These are process noise and measurement noise, respectively. It is a time variable.

[0010] According to the deep space autonomous navigation method based on sensor combination optimization provided by the present invention, the calculation of the optimization objective function value of each candidate sensor combination specifically includes: Fisher information matrices for each candidate sensor combination are constructed based on the motion state of the multi-source heterogeneous system, the Fisher information matrices, and the onboard computational and power consumption costs of each candidate sensor. Then, the calculation is performed using an optimization objective function. The expression is: ; in, For the candidate sensor set, This represents the Fisher information matrix for each candidate sensor. For the entire set of available sensors, To call the first The computational cost and power consumption cost of each sensor and These are adaptive weighting coefficients; The expression for the Fisher information matrix of each candidate sensor is as follows: ; in, Let Jacobian matrix be the equation of measurement. To measure the noise covariance matrix, For transpose; The Jacobian matrix is ​​obtained through the nonlinear measurement function. Taking the partial derivative with respect to the state, its expression is: ; in, For the current moment The system state estimate; The nonlinear measurement function is derived from the nonlinear measurement equation. The first-order expression is obtained by performing a Taylor expansion near the current state estimate: .

[0011] According to the deep space autonomous navigation method based on sensor combination optimization provided by the present invention, the distributed low-level preprocessing is completed by several locally distributed filters operating in parallel, wherein the locally distributed filters include: The first local distributed filter performs time integration and dimensionality reduction by pre-integrating the force and angular velocity measurements to obtain the estimated local relative orbital motion state. and the corresponding local covariance matrix ; The second local distributed filter, by establishing an attitude fusion filter loop, estimates and corrects the gyroscope constant bias and integral angle drift online and in real time, thus obtaining the inertial attitude state estimate. and the corresponding local attitude covariance matrix ; The third local distributed filter constructs a deterministic pointing constraint on a specific axis of the spacecraft using the solar direction vector. Combined with the pre-integrated angular velocity, it establishes a dimension-reduced filtering loop to correct the gyroscope's drift error in the direction perpendicular to the solar vector in real time, thus obtaining an estimate of the solar line-of-sight reference attitude state. and the corresponding local covariance matrix .

[0012] According to the deep space autonomous navigation method based on sensor combination optimization provided by the present invention, the centralized global joint estimation specifically includes: Within the main filter, global elastic fusion is performed using information filtering. The update rule for global elastic fusion is as follows: ; in, and These are the estimated local optimal state and covariance matrix after dimensionality reduction, respectively. The prediction information matrix of the master filter before measurement update. The main filter's prediction information vector before measurement updates. The branch numbers of the local filters participating in global fusion. The total number of local filter branches participating in global fusion. Local adaptive fusion weights are assigned based on system observability optimization. For tensor measurement by gravitational gradiometer Vector measurement with magnetometer The global physical field measurement vector constituted and These are the corresponding global measurement Jacobian matrix and measurement noise covariance matrix, respectively; An elastic adjustment factor introduced for global measurement of physical fields. This is a transpose.

[0013] According to the deep space autonomous navigation method based on sensor combination optimization provided by the present invention, the expression of the augmented state vector is as follows: ; in, and These represent the position and velocity of the probe in the inertial frame of the central celestial body, respectively. For attitude quaternions, and These are the zero bias values ​​for the accelerometer and gyroscope, respectively. The increment of the gravitational field parameters of the target celestial body; The increment of the target celestial body's magnetic field parameter error. For transpose; The strongly coupled measurement model is constructed based on a gravitational gradiometer and a magnetometer: The tensor measurement equation of the gravitational gradiometer is: ; in, For the gravitational gradient tensor model of the target celestial body, These are the initial values ​​of the gravitational field parameters. For attitude quaternions, This is the attitude transformation matrix. To measure noise for the gravitational gradiometer; The vector measurement equation of the magnetometer is expressed as: ; in, For the target celestial body's magnetic field evolution vector model, These are the initial values ​​for the magnetic field parameters. This is for the noise in the vector measurement of the magnetometer.

[0014] According to the present invention, a deep space autonomous navigation method based on sensor combination optimization is provided, wherein the step-by-step release calibration of physical field parameters according to the current observability conditions specifically includes: In the initial calibration stage of the physical field parameters, the nominal value of the initial reference short arc segment orbit is fixed or the update range of the initial reference short arc segment orbit is limited, the orbit state covariance is preserved and its uncertainty is propagated to the physical field parameter estimation process. Based on the Fisher information matrix, physical field parameters whose observability meets the preset threshold are released, while physical field parameters whose observability does not meet the preset threshold are kept in the prior value until the observability meets the preset threshold.

[0015] According to the deep space autonomous navigation method based on sensor combination optimization provided by the present invention, the method of updating the orbital dynamics model based on the error increment of gravitational field parameters and updating the magnetic field measurement prediction model based on the error increment of magnetic field parameters specifically includes: The incremental error of the gravitational field parameters is superimposed onto the prior nominal gravitational field model to update the orbital dynamics model; The magnetic field parameter error increment is updated into the target celestial body's magnetic field evolution model, thereby updating the magnetic field measurement prediction model.

[0016] According to the deep space autonomous navigation method based on sensor combination optimization provided by the present invention, the consistency verification includes: The updated theoretical measurement data is compared with the actual sensor observation data to construct the first... Integrated measurement residual vector at time step Based on the comprehensive measurement residual vector Calculate the normalized Mahalanobis distance test value And determine the allowable threshold based on the critical value of the chi-square distribution at the preset confidence level. ; The normalized Mahalanobis distance test value The expression is: ; in, The residual covariance matrix is... For the measurement matrix Jacobian, To predict the covariance matrix in one step, To measure the noise covariance matrix, This is the composite measurement residual vector; The comprehensive measurement residual vector The expression is: ; in, and These are the actual measurements from the gravitational gradiometer and magnetometer at the current moment, respectively. and These are the theoretically predicted measurements calculated based on the updated physical field parameters. For the first The augmented state vector estimate at time step 1; when At that time, output the detector navigation status. and the final calibrated higher-order physical field parameters ,in, This is the vector of spherical harmonic coefficients of the gravitational field of the target celestial body, as determined in the final calibration. This is the final calibrated parameter vector of the target celestial magnetic field dipole; when At this time, the fault-tolerant isolation mechanism is triggered, temporarily suspending the update of abnormal physical field parameters and maintaining the converged value of the previous cycle. The detector navigation state is output based on the local state estimates of the geometric and kinematic sensors. The covariance of the abnormal physical field measurement noise is increased and its fusion weight is applied. Global joint estimation and consistency verification are performed again using subsequent measurement data to obtain a new normalized Mahalanobis distance verification quantity. ; like To restore the update of abnormal physical field parameters, if The measurement channel for abnormal physical fields is cut off and the operation is downgraded.

[0017] The present invention also provides a deep space autonomous navigation system based on sensor combination optimization, for executing any of the deep space autonomous navigation methods described above, comprising: The multi-source heterogeneous sensor acquisition module is used to acquire measurement data from the multi-source heterogeneous system and perform preprocessing. The observation availability determination and target / measurement information optimization module is used to construct the Fisher information matrix of each candidate sensor combination. Based on the motion state of the multi-source heterogeneous system, the Fisher information matrix and the on-board computing cost and power consumption cost of each candidate sensor, the module calculates the optimization objective function value of each candidate sensor combination and determines the candidate sensor combination with the largest optimization objective function value as the sensor combination. The first-level local fusion module is used to perform distributed low-level preprocessing on the measurement data of the geometric and kinematic sensors and the measurement data of the solar sensor in the sensor combination to obtain the local state estimates and their covariance matrices of each branch. The second-level global joint estimation module is used to input the local state estimates of each branch and their covariance matrix and the raw data of the space physics field sensor into the main filter. Based on the augmented state vector, evolutionary orbital dynamics constraints and strongly coupled measurement model, it performs global elastic fusion in the form of information filtering to perform centralized global joint estimation. The physics parameter estimation and model update module is used to perform step-by-step release calibration of the physics parameters based on the initial reference short-arc orbit in the global joint estimation and the current observability conditions, obtaining the error increments of the gravitational field parameters and magnetic field parameters. Based on the gravitational field parameter error increments, the orbital dynamics model is updated; based on the magnetic field parameter error increments, the magnetic field measurement prediction model is updated, resulting in updated theoretical measurement data. The closed-loop consistency verification and navigation output module is used to perform consistency verification on the updated theoretical measurement data and actual measurement data. When the verification value of the consistency verification is less than or equal to the allowable threshold, the module outputs the detector navigation status and the final calibrated physical field parameters. When the verification value of the consistency verification is greater than the allowable threshold, the module triggers the fault-tolerant isolation mechanism, adjusts the measurement noise covariance and fusion weight of the current physical field measurement data, and performs global joint estimation and consistency verification based on subsequent measurement data.

[0018] Compared with the prior art, the advantages of this application are: 1. This invention proposes an autonomous optimization method for multi-source heterogeneous information based on optimal observability: by constructing a combined Fisher information matrix and setting a multi-objective optimization criterion that comprehensively balances observability and computational cost, the real-time autonomous optimization of heterogeneous measurement information sources and observation targets on the satellite is realized, effectively avoiding invalid redundant calculations and noise pollution in the complex environment of deep space.

[0019] 2. This invention constructs a two-tiered fusion architecture of evolutionary orbital dynamics constraints and elasticity, overcoming the limitation of traditional deep-space navigation heavily relying on known high-order prior models. At the bottom layer, distributed local filtering is used to achieve visual / astronomical inertial preprocessing and dimensionality reduction; at the top layer, evolutionary dynamic constraints are introduced, adaptively upgrading from low to high order, in the main filter. Combined with observability analysis and physical field elastic weight adjustment factors, stable extrapolation is achieved based on orbital dynamics even in astronomical observation shadow blind zones, significantly improving the balance between system computing power and accuracy in extreme space environments.

[0020] 3. This invention proposes a joint estimation method based on prior-guided decoupling and closed-loop tolerance verification. Addressing the unobservability and filter divergence issues caused by the strong coupling and mutual initial constraints between the physical field parameters of unknown deep-space astrophysical objects and the probe's navigation state, a sequential decoupling strategy is constructed, starting with low-order prior geometry-guided initiation, inverse physical field construction and calibration, dynamic model upgrade and update, and finally closed-loop iterative convergence. By introducing a quantitative discrimination mechanism based on the normalized Mahalanobis distance and chi-square distribution threshold for the allowable deviation range, adaptive closed-loop feedback iterative solution is achieved when the measurement deviation exceeds the tolerance, and stable output is achieved when the tolerance requirements are met. This method ensures high-precision autonomous orbit determination and attitude assessment of the probe in complex deep-space environments, endowing the probe with the ability to adaptively invert high-precision physical field parameters in unknown deep-space environments. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the logical flow for autonomous selection of observation targets and measurement information based on the Fisher information matrix; Figure 2 A schematic diagram of the structure of a distributed-centralized two-level information fusion filtering algorithm; Figure 3 This is a schematic diagram of a three-stage decoupled joint estimation and closed-loop verification process based on prior guidance. Detailed Implementation

[0022] The technical solutions provided in this application are further illustrated below with reference to the embodiments.

[0023] Example 1 This invention employs a multi-source heterogeneous detection system consisting of a star sensor, a solar sensor, a visible / infrared camera, a magnetometer, a gravitational gradiometer, and an inertial measurement unit (IMU). The specific implementation steps are as follows: Dynamic autonomous optimization of sensor combinatorial observability: Based on the current motion state of the system, Lie derivative observability and Fisher information matrix quantification analysis are performed on multi-source heterogeneous measurement information. The geometric observation contribution of each sensor subset to the state is comprehensively weighed against the on-board computing and power consumption costs. Degraded or redundant measurement sources are dynamically eliminated, and the optimal sensor combinatorial combination at the current moment is selected. The solar sensor is directly input into the underlying processing as a resident hardware redundancy benchmark. Fusion of geometric kinematics preprocessing and intensive estimation of physical fields: Constructing an elastic two-level filtering architecture. At the first level, local distributed filters are set up in parallel for geometric and kinematic sensors such as visible light / infrared cameras, star sensors, solar sensors, and IMUs. These filters respectively complete relative orbit measurement denoising, high-precision full-attitude benchmark calculation, and coarse benchmark constraints for solar line-of-sight, achieving decoupling of local motion states and dimensionality reduction of high-dimensional measurements. At the second level, the main filter receives the kinematic state outputs of each local filter at the bottom level and directly connects to the tensor / vector data of space physical field sensors such as gravitational gradiometers and magnetometers. Combined with evolutionary orbital dynamic constraints, a global centralized joint optimization is achieved using an elastic information filtering framework with adaptive adjustment factors. Prior-guided sequential decoupling and closed-loop tolerance verification: In the global joint estimation, the high-precision benchmark short-arc orbit in the initial stage is used as a strong prior, and the physical field parameters are calibrated step by step according to the current observability conditions; the inverted gravitational field error increment is used to upgrade the orbital dynamics evolution model, and the magnetic field error increment is used to correct the theoretical magnetic field measurement prediction model; closed-loop consistency verification is implemented by constructing a Mahalanobis distance and chi-square distribution deviation judgment criterion: when the tolerance is met, the navigation state and calibration parameters are output in convergence; when the deviation exceeds the limit, the measurement noise covariance is adaptively expanded and fault-tolerant isolation and degraded operation are triggered to ensure the smooth and continuous navigation solution of the detector and adaptive fault-tolerant convergence.

[0024] 1. Algorithm for Autonomous Target Selection of Multi-Source Measurement Information and Observation Targets To avoid invalid data consuming computing power and introducing noise, this invention designs an information autonomous optimization mechanism based on observability. The state-space model of the nonlinear multi-source heterogeneous observation system is defined as follows:

[0025] in, Let be the system state vector. The nonlinear measurement equation represents the first... The measurement output of each sensor, and These are process noise and measurement noise, respectively. For the system input vector, The first derivative of the system state vector with respect to time. For time variables, It is a nonlinear measurement function. It is a vector-valued function that, after substituting the system state vector, system input vector, and time variable, outputs a column vector with the same dimension as the system state vector.

[0026] In order to quantify the first This invention assesses the contribution of a combination of sensors to the current system state, using the Fisher Information Matrix (FIM) as the evaluation criterion for observability. The Fisher Information Matrix of the combined sensors is defined as follows:

[0027] in, For the candidate sensor set, Let Jacobian matrix be the equation of measurement. To measure the noise covariance matrix, This is a transpose.

[0028] Through the nonlinear measurement equations of each sensor Establish the relationship between system state and measurement data, and analyze the nonlinear measurement functions within it. Taking the partial derivative with respect to the state, we obtain the measurement Jacobian matrix.

[0029] The expression for the Jacobian matrix is:

[0030] in, For the current moment System state estimate, nonlinear measurement equation A Taylor expansion is performed near the current state estimate, and its first-order expression is: .

[0031] This invention sets the optimal observability criterion as maximizing the determinant of the Fisher information matrix to ensure the minimization of the error ellipsoid volume. The optimization objective function is designed as follows:

[0032] in, For the entire set of available sensors, To call the first The computational cost and power consumption cost of each sensor and These are adaptive weighting coefficients. By solving this optimization problem, online real-time optimization of measurement information can be achieved.

[0033] 2. Strongly Coupled System Models and State Augmentation Unlike traditional navigation methods that only estimate the six degrees of freedom of the probe, this invention introduces the unknown physical field parameters of target celestial bodies such as Neptune into the state space. This invention incorporates prior knowledge, modeling the physical field parameters as a combination of a prior low-order model and the increment of the error to be estimated.

[0034] Define augmented state vector , used to represent the system state vector The unknown quantity in the expression is:

[0035] in, and These represent the position and velocity of the probe in the inertial frame of the central celestial body, respectively. For attitude quaternions, and These are the zero biases of the accelerometer and gyroscope, respectively. The key augmentation term is the estimated error increment: The spherical harmonic coefficients of the gravitational field of the target celestial body (e.g.) , The perturbation term parameters are compared to the initial values ​​of the spherical harmonic coefficients of the gravitational field in the prior low-order model. The correction vector to be estimated; The initial values ​​of the target celestial body's magnetic field dipole parameters (magnetic moment vector and tilt angle) compared to the initial values ​​of the dipole parameters in the prior low-order model. The correction vector to be estimated.

[0036] Given the shadow blind zone in astronomical observations, this invention introduces evolutionary orbital dynamic constraints into the system dynamics model. In the initial stage of system operation, nominal orbital dynamic equations based on prior low-order parameters are used. As the physical field error increment is continuously calibrated and updated in subsequent closed-loop verifications, these equations are gradually upgraded and evolved into high-precision orbital dynamic constraints. In the event of astronomical sensor failure, the detector state is deduced through the evolved high-precision orbital dynamics model, thereby compensating for the information loss due to inaccuracies in the physical field model and the unavailability of astronomical observations.

[0037] To achieve the co-inversion of state and unknown physical field parameters, this invention combines a gravitational gradiometer and a magnetometer to construct a strongly coupled measurement model.

[0038] The measurement equations for the gravitational gradiometer are expressed as follows:

[0039] in, This is a tensor model of the gravitational gradient of a target celestial body, with gravitational field parameters (the sum of the initial values ​​of the gravitational field spherical harmonic coefficients and the correction vector to be estimated) as variables. For attitude quaternions The calculated attitude transformation matrix is ​​used to project the gravitational gradient tensor in the inertial frame onto the gravitational gradiometer measurement coordinate system; To measure noise for the gravitational gradiometer.

[0040] The vector measurement equation of the magnetometer is expressed as:

[0041] in, This is a model for the evolution of the magnetic field of a target celestial body, with magnetic field parameters (the sum of the initial values ​​of the dipole parameters and the correction vector to be estimated) as variables. This is for the noise in the vector measurement of the magnetometer.

[0042] 3. Two-level information fusion architecture To resolve the contradiction between the excessive computational power consumption of centralized fusion algorithms and the limited global accuracy of distributed filtering, and to eliminate the risk of filter divergence caused by the introduction of low-order physical models, this invention proposes the following... Figure 2 The two-stage fusion architecture under the elastic filtering framework shown: Level 1 (Distributed underlying preprocessing): For local sensors with high sampling frequencies and high measurement dimensions, such as star sensors, optical cameras (visible / infrared), and IMUs, parallel-operating local distributed filters (e.g., Error State Kalman Filter, ESKF) are established to perform noise reduction and local state dimensionality reduction calculations. This system, through a parallel architecture, includes at least the following two local preprocessing subsystems: 1) Time integration dimensionality reduction of high-dynamic IMU specific force and angular velocity measurements is performed using pre-integration techniques. For the long-distance approach and orbit phases of deep space (defined internationally as a space region ≥ 2 million kilometers from Earth), macroscopic geometric features such as the line-of-sight (LOS) vector, centroid deviation, or limb contour of the target celestial body are extracted from the image plane to avoid the direct participation of pixel-level high-dimensional image matrices in global calculations. Combined with IMU pre-integration dynamics, outliers and measurement noise in local optical observations are smoothed to calculate the estimated local relative orbital motion state corresponding to that branch. (Including the position and velocity relative to the target celestial body) and the corresponding local covariance matrix .

[0043] 2) "Star Sensor + IMU" Astronomical Inertial Subsystem: An attitude fusion filtering loop is established based on the high-precision absolute attitude observation data of the inertial system obtained through star sensor line-of-sight vector matching. Utilizing the absolute orientation capability of the star sensor, the constant zero bias and integral angle drift of the IMU gyroscope are estimated online and corrected in real time, resulting in a high-precision estimate of the inertial attitude state. (Including attitude quaternions) Zero bias of the gyroscope and the corresponding local attitude covariance matrix .

[0044] 3) "Solar Sensor + IMU" Astronomical Inertial Subsystem: For extreme conditions such as intense light interference in deep space and high angular rate maneuvers, the solar sensor, as a resident, low-power, high-reliability sensor, independently outputs the space solar direction vector (two-degree-of-freedom line-of-sight vector). Although the solar sensor can be classified as a geometric and kinematic sensor, it serves as a resident redundant reference. This subsystem bypasses front-end information optimization and directly connects to the corresponding underlying local filter. It utilizes the solar direction vector to construct a deterministic pointing constraint for a specific axis of the spacecraft. Combined with the IMU's pre-integrated angular velocity, it establishes a dimension-reduced filtering loop to correct the gyroscope's drift error perpendicular to the solar vector direction in real time. It independently calculates the estimated solar line-of-sight reference attitude state for this branch. and the corresponding local covariance matrix This subsystem forms a hardware and software complementarity redundancy with the star sensor branch, seamlessly providing the main filter with uninterrupted coarse attitude reference and safety margin constraints when the star sensor is blinded or fails due to strong stray light.

[0045] Each local distributed filter operates in parallel, outputting its own dimensionality-reduced local optimal state estimate to the top-level master filter. and covariance matrix , This refers to the branch number of the local filter that participates in global fusion.

[0046] Level 2 (Elastic Centralized Global Joint Estimation): The top-level master filter receives geometric pose references and motion state estimates from the outputs of each bottom-level local filter, and directly inputs raw tensor / vector measurement data from space physics field sensors such as gravitational gradiometers and magnetometers. Within the master filter, global elastic fusion is performed using information filtering. Combining the aforementioned sensor ensemble autonomous optimization mechanism and evolutionary orbital dynamics constraints, a global information matrix is ​​defined. and information vector The elastic update rules are as follows: ; in, The prediction information matrix of the master filter before measurement update. The main filter's prediction information vector before measurement updates. The branch numbers of the local filters participating in global fusion. The total number of local filter branches participating in global fusion. Local adaptive fusion weights are assigned based on system observability optimization. For tensor measurement by gravitational gradiometer Vector measurement with magnetometer The global physical field measurement vector constituted and These are the corresponding global measurement Jacobian matrix and measurement noise covariance matrix, respectively; An elastic adjustment factor introduced for global measurement of physical fields. This is a transpose.

[0047] When the physical field model is coarse and the inversion accuracy is lower than the geometric benchmark based on pure optics / astronomy and nominal orbital dynamics in the early stages of the mission, the system adaptively lowers the accuracy. The weights are adjusted to avoid contaminating the global state estimate with initially unreliable physical field data; and the system's resilience is enhanced after high-precision gravitational gradient and magnetic field parameters are derived through closed-loop verification, or when the detector enters the shadow region of a celestial body, causing optical observations to degrade / temporarily fail. The weights are determined by precise physical field constraints to ensure global continuous high-precision orbit determination and smooth parameter estimation.

[0048] 4. Prior-guided decoupled joint estimation and adaptive error correction In global centralized fusion, this invention designs a joint estimation strategy based on prior model-guided initiation and closed-loop verification: 1) Prior boot startup: In the initial phase, when the probe approaches the target celestial object and optical conditions are favorable, the system inputs astronomical data (star sensor), inertial navigation data (IMU), and visual data (camera), and introduces a nominal orbital dynamics model (constructed based on low-order prior physical field parameters) as a primary constraint. During this phase, the system assigns the highest weight to geometric astronomical observations, and decouples the probe's initial reference short-arc orbit through local geometric intersection and filtering algorithms.

[0049] 2) Construction and calibration of physical fields: The high-precision reference short-arc orbit obtained during the initial phase is used as a strong prior state for the inversion of physical field parameters. Its nominal values ​​are fixed or its update range is limited during the initial parameter calibration phase, while the orbital state covariance is preserved and its uncertainty is propagated to the physical field parameter estimation process. Measurement data from physical field sensors such as the gravitational gradiometer and magnetometer are incorporated into the system. Under the constraints of the current reference orbit, the error increment of the target astrophysical field is constructed and solved in reverse. and .

[0050] To address the varying observability conditions of physical field parameters, a phased release strategy is implemented: combining the aforementioned Fisher information matrix and observability quantification analysis, estimations are prioritized for physical field parameters with sufficient observability under the current orbital geometry (such as low-order gravitational field perturbation parameters and principal dipole magnetic field parameters); estimations for high-order physical field parameters with insufficient observability are temporarily withheld to preserve prior values ​​and prevent under-observation from causing parameter estimation singularities or filter divergence. Activation and updates are then implemented once subsequent changes in the probe's orbital geometry and spatial position, and the parameter observability reach a preset threshold.

[0051] 3) Model update: Decouple and differentially update the system dynamics and measurement equations based on physical field properties: Gravitational field model update (applies to orbital dynamics prediction): increments the error of the inverted gravitational field parameters. Superimposed on the prior nominal gravitational field model, the orbital dynamics model of the detector is updated, upgrading it from an initial low-order nominal model to an evolutionary high-precision orbital dynamics equation, which significantly improves the accuracy of orbital state extrapolation and prediction during periods without continuous optical observations or astronomical geometric degradation. Magnetic field model update (applied to the measurement prediction model): The error increment of the inverted magnetic field parameters is updated. Updated to the target celestial body's magnetic field evolution model to correct subsequent theoretical magnetic field measurement prediction equations. This will improve the theoretical measurement and prediction accuracy of the magnetometer.

[0052] The updated high-precision orbital dynamics model and the magnetic field measurement prediction model are used together as input benchmarks to participate in subsequent top-level global centralized information fusion and consistency tolerance verification of state and observation data.

[0053] 4) Determining the permissible range of deviation between the physical field model and the observed data: The theoretical measurements obtained by back-calculation from the updated physical field model are compared with the observation data from actual sensors to construct the first... The composite measurement residual vector at time:

[0054] in, and These are the actual measurements from the gravitational gradiometer and magnetometer at the current moment, respectively. and These are the theoretically predicted measurements calculated based on the updated physical field parameters. For the first The augmented state vector estimate at time t.

[0055] To quantitatively determine whether the deviation is within a reasonable tolerance, the system defines the normalized Mahalanobis distance test metric. As a criterion for deviation:

[0056] in, The residual covariance matrix is... For the measurement matrix Jacobian, To predict the covariance matrix in one step, This is the measurement noise covariance matrix.

[0057] System setting allowable threshold (Based on chi-square distribution) Determining the critical value under a pre-set confidence level. (For measuring residual dimensions). The judgment criterion is defined as:

[0058] If the judgment is "yes" (deviation within acceptable range): it is determined that the currently updated physical field parameters are in high agreement with the actual orbital environment, the system converges, and the high-precision navigation status of the detector is directly output. and the final calibrated higher-order physical field parameters ,in, Let the final calibrated target celestial body's gravitational field spherical harmonic coefficient vector be... , The final calibrated target celestial body magnetic field dipole parameter vector and .

[0059] If the result is "No" (deviation exceeds the allowable range): This indicates a gross error in the current physical field measurement, triggering the fault-tolerant isolation mechanism. The system temporarily suspends updating the abnormal parameter in the current cycle, maintaining the converged value from the previous cycle, and prioritizing the continuous and stable output of the navigation status based on the geometric kinematics perception benchmark. Simultaneously, the out-of-range result is fed back to the front end, adaptively expanding the abnormal physical field measurement noise covariance, suppressing its fusion weights, and performing rolling iterative optimization based on subsequent measurement data. A new global joint estimation and consistency check are then performed to obtain a new normalized Mahalanobis distance verification quantity. ;like This indicates that updates will resume once the indicator returns to convergence. If the limit is continuously exceeded, the channel will be cut off and the operation will be downgraded. The object that maintains the convergence value of the previous cycle should be clearly defined as the physical field parameters corresponding to the abnormal channel. The detector position, velocity and attitude still need to be continuously predicted and updated.

[0060] 5. System Composition This invention also provides a deep-space two-level autonomous navigation information fusion system based on target selection, comprising: a multi-source heterogeneous sensor acquisition module, an observation availability determination and target / measurement information selection module, a first-level local fusion module, a second-level global joint estimation module, a physical field parameter estimation and model update module, and a closed-loop consistency verification and navigation output module. Each module is used to execute the method steps described in 1-4 above.

[0061] Example 2 A deep-space autonomous navigation method based on sensor combination optimization includes: Acquire measurement data of a multi-source heterogeneous system and preprocess it, calculate the optimization objective function value of each candidate sensor combination, and determine the candidate sensor combination with the largest optimization objective function value as the sensor combination; The measurement data of the geometry and kinematics sensors and the solar sensor in the sensor array are preprocessed in a distributed low-level manner to obtain the local state estimates and their covariance matrices of each branch. The local state estimates and their covariance matrices of each branch and the measurement data of the space physics field sensor are input into the main filter. Based on the augmented state vector, evolutionary orbital dynamics constraints and strongly coupled measurement model, global elastic fusion is performed in the form of information filtering to perform centralized global joint estimation. Using the initial reference short-arc orbit in the global joint estimation as a priori, the physical field parameters are released and calibrated step by step according to the current observability conditions to obtain the error increments of the gravitational field parameters and the magnetic field parameters. The orbital dynamics model is updated based on the error increments of the gravitational field parameters, and the magnetic field measurement prediction model is updated based on the error increments of the magnetic field parameters to obtain the updated theoretical measurement data. The updated theoretical measurement data and actual measurement data are subjected to consistency verification. When the verification value of consistency verification is less than or equal to the preset tolerance threshold, the detector navigation status and the final calibrated physical field parameters are output. When the verification value of consistency verification is greater than the preset tolerance threshold, the fault tolerance isolation mechanism is triggered, the measurement noise covariance and fusion weight of the current physical field measurement data are adjusted, and global joint estimation and consistency verification are performed based on subsequent measurement data.

[0062] According to the present invention, a deep space autonomous navigation method based on sensor combination optimization is provided, wherein the acquisition and preprocessing of measurement data from a multi-source heterogeneous system are described, and the preprocessing specifically includes: The measurement data of the multi-source heterogeneous system are analyzed using a state-space model, the expression of which is: ; in, The first derivative of the system state vector with respect to time. It is a vector-valued function. Let be the system state vector. For the system input vector, For the first The measurement output of each sensor, It is a nonlinear measurement function. and These are process noise and measurement noise, respectively. It is a time variable.

[0063] According to the deep space autonomous navigation method based on sensor combination optimization provided by the present invention, the calculation of the optimization objective function value of each candidate sensor combination specifically includes: Fisher information matrices for each candidate sensor combination are constructed based on the motion state of the multi-source heterogeneous system, the Fisher information matrices, and the onboard computational and power consumption costs of each candidate sensor. Then, the calculation is performed using an optimization objective function. The expression is: ; in, For the candidate sensor set, This represents the Fisher information matrix for each candidate sensor. For the entire set of available sensors, To call the first The computational cost and power consumption cost of each sensor and These are adaptive weighting coefficients; The expression for the Fisher information matrix of each candidate sensor is as follows: ; in, Let Jacobian matrix be the equation of measurement. To measure the noise covariance matrix, For transpose; The Jacobian matrix is ​​obtained through the nonlinear measurement function. Taking the partial derivative with respect to the state, its expression is: ; in, For the current moment The system state estimate; The nonlinear measurement function is derived from the nonlinear measurement equation. The first-order expression is obtained by performing a Taylor expansion near the current state estimate: .

[0064] According to the deep space autonomous navigation method based on sensor combination optimization provided by the present invention, the distributed low-level preprocessing is completed by several locally distributed filters operating in parallel, wherein the locally distributed filters include: The first local distributed filter performs time integration and dimensionality reduction by pre-integrating the force and angular velocity measurements to obtain the estimated local relative orbital motion state. and the corresponding local covariance matrix ; The second local distributed filter, by establishing an attitude fusion filter loop, estimates and corrects the gyroscope constant bias and integral angle drift online and in real time, thus obtaining the inertial attitude state estimate. and the corresponding local attitude covariance matrix ; The third local distributed filter constructs a deterministic pointing constraint on a specific axis of the spacecraft using the solar direction vector. Combined with the pre-integrated angular velocity, it establishes a dimension-reduced filtering loop to correct the gyroscope's drift error in the direction perpendicular to the solar vector in real time, thus obtaining an estimate of the solar line-of-sight reference attitude state. and the corresponding local covariance matrix .

[0065] According to the deep space autonomous navigation method based on sensor combination optimization provided by the present invention, the centralized global joint estimation specifically includes: Within the main filter, global elastic fusion is performed using information filtering. The update rule for global elastic fusion is as follows: ; in, and These are the estimated local optimal state and covariance matrix after dimensionality reduction, respectively. The prediction information matrix of the master filter before measurement update. The main filter's prediction information vector before measurement updates. The branch numbers of the local filters participating in global fusion. The total number of local filter branches participating in global fusion. Local adaptive fusion weights are assigned based on system observability optimization. For tensor measurement by gravitational gradiometer Vector measurement with magnetometer The global physical field measurement vector constituted and These are the corresponding global measurement Jacobian matrix and measurement noise covariance matrix, respectively; An elastic adjustment factor introduced for global measurement of physical fields. This is a transpose.

[0066] According to the deep space autonomous navigation method based on sensor combination optimization provided by the present invention, the expression of the augmented state vector is as follows: ; in, and These represent the position and velocity of the probe in the inertial frame of the central celestial body, respectively. For attitude quaternions, and These are the zero bias values ​​for the accelerometer and gyroscope, respectively. The increment of the gravitational field parameters of the target celestial body; The increment of the target celestial body's magnetic field parameter error. For transpose; The strongly coupled measurement model is constructed based on a gravitational gradiometer and a magnetometer: The tensor measurement equation of the gravitational gradiometer is: ; in, For the gravitational gradient tensor model of the target celestial body, These are the initial values ​​of the gravitational field parameters. For attitude quaternions, This is the attitude transformation matrix. To measure noise for the gravitational gradiometer; The vector measurement equation of the magnetometer is expressed as: ; in, For the target celestial body's magnetic field evolution vector model, These are the initial values ​​for the magnetic field parameters. This is for the noise in the vector measurement of the magnetometer.

[0067] According to the present invention, a deep space autonomous navigation method based on sensor combination optimization is provided, wherein the step-by-step release calibration of physical field parameters according to the current observability conditions specifically includes: In the initial calibration stage of the physical field parameters, the nominal value of the initial reference short arc segment orbit is fixed or the update range of the initial reference short arc segment orbit is limited, the orbit state covariance is preserved and its uncertainty is propagated to the physical field parameter estimation process. Based on the Fisher information matrix, physical field parameters whose observability meets the preset threshold are released, while physical field parameters whose observability does not meet the preset threshold are kept in the prior value until the observability meets the preset threshold.

[0068] According to the deep space autonomous navigation method based on sensor combination optimization provided by the present invention, the method of updating the orbital dynamics model based on the error increment of gravitational field parameters and updating the magnetic field measurement prediction model based on the error increment of magnetic field parameters specifically includes: The incremental error of the gravitational field parameters is superimposed onto the prior nominal gravitational field model to update the orbital dynamics model; The magnetic field parameter error increment is updated into the target celestial body's magnetic field evolution model, thereby updating the magnetic field measurement prediction model.

[0069] According to the deep space autonomous navigation method based on sensor combination optimization provided by the present invention, the consistency verification includes: The updated theoretical measurement data is compared with the actual sensor observation data to construct the first... Integrated measurement residual vector at time step Based on the comprehensive measurement residual vector Calculate the normalized Mahalanobis distance test value And determine the allowable threshold based on the critical value of the chi-square distribution at the preset confidence level. ; The normalized Mahalanobis distance test value The expression is: ; in, The residual covariance matrix is... For the measurement matrix Jacobian, To predict the covariance matrix in one step, To measure the noise covariance matrix, This is the composite measurement residual vector; The comprehensive measurement residual vector The expression is: ; in, and These are the actual measurements from the gravitational gradiometer and magnetometer at the current moment, respectively. and These are the theoretically predicted measurements calculated based on the updated physical field parameters. For the first The augmented state vector estimate at time step 1; when At that time, output the detector navigation status. and the final calibrated higher-order physical field parameters ,in, This is the vector of spherical harmonic coefficients of the gravitational field of the target celestial body, as determined in the final calibration. This is the final calibrated parameter vector of the target celestial magnetic field dipole; when At this time, the fault-tolerant isolation mechanism is triggered, temporarily suspending the update of abnormal physical field parameters and maintaining the converged value of the previous cycle. The detector navigation state is output based on the local state estimates of the geometric and kinematic sensors. The covariance of the abnormal physical field measurement noise is increased and its fusion weight is applied. Global joint estimation and consistency verification are performed again using subsequent measurement data to obtain a new normalized Mahalanobis distance verification quantity. ; like To restore the update of abnormal physical field parameters, if The measurement channel for abnormal physical fields is cut off and the operation is downgraded.

[0070] Example 3 A deep-space autonomous navigation system based on sensor combination optimization, used to execute the deep-space autonomous navigation method described in any of the above embodiments, includes: The multi-source heterogeneous sensor acquisition module is used to acquire measurement data from the multi-source heterogeneous system and perform preprocessing. The observation availability determination and target / measurement information optimization module is used to construct the Fisher information matrix of each candidate sensor combination. Based on the motion state of the multi-source heterogeneous system, the Fisher information matrix and the on-board computing cost and power consumption cost of each candidate sensor, the module calculates the optimization objective function value of each candidate sensor combination and determines the candidate sensor combination with the largest optimization objective function value as the sensor combination. The first-level local fusion module is used to perform distributed low-level preprocessing on the measurement data of the geometric and kinematic sensors and the measurement data of the solar sensor in the sensor combination to obtain the local state estimates and their covariance matrices of each branch. The second-level global joint estimation module is used to input the local state estimates of each branch and their covariance matrix and the raw data of the space physics field sensor into the main filter. Based on the augmented state vector, evolutionary orbital dynamics constraints and strongly coupled measurement model, it performs global elastic fusion in the form of information filtering to perform centralized global joint estimation. The physical field parameter estimation and model update module is used to take the initial reference short arc orbit in the global joint estimation as a priori, perform step-by-step release calibration of the physical field parameters according to the current observability conditions, obtain the error increments of the gravitational field parameters and the error increments of the magnetic field parameters, update the orbital dynamics model based on the error increments of the gravitational field parameters, update the magnetic field measurement prediction model based on the error increments of the magnetic field parameters, and obtain the updated theoretical measurement data. The closed-loop consistency verification and navigation output module is used to perform consistency verification on the updated theoretical measurement data and actual measurement data. When the verification value of the consistency verification is less than or equal to the allowable threshold, the module outputs the detector navigation status and the final calibrated physical field parameters. When the verification value of the consistency verification is greater than the allowable threshold, the module triggers the fault-tolerant isolation mechanism, adjusts the measurement noise covariance and fusion weight of the current physical field measurement data, and performs global joint estimation and consistency verification based on subsequent measurement data.

[0071] In summary, this invention establishes an optimal observability criterion for multi-source heterogeneous sensors, comprehensively balancing system state observability with onboard computation and power consumption costs, thereby achieving dynamic autonomous optimization of onboard sensor measurement information and combination configurations. A two-level fusion filtering architecture is constructed that decouples geometric kinematic perception and physical field dynamics. The bottom layer performs local high-dimensional kinematic dimensionality reduction and smoothing, while the top layer implements global elastic information fusion, achieving the best balance between spaceborne computing power constraints and global state estimation accuracy. A priori-guided sequential decoupling and closed-loop tolerance verification strategy is introduced. Based on the observability conditions, the physical field parameters are estimated in stages. When the limit is exceeded, elastic degradation and adaptive covariance expansion are implemented to overcome the strong coupling bottleneck of mutual constraints between the state and unknown physical field parameters. This enables high-precision joint fault-tolerant estimation of the deep space navigation state and the target astrophysical field parameters, significantly improving the adaptability, continuity and robustness of autonomous navigation in the extreme and complex environment of deep space.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of this application do not depart from the spirit and scope of the technical solutions of this application, and should all be covered within the scope of the claims of this application.

Claims

1. A deep-space autonomous navigation method based on sensor combination optimization, comprising: Acquire measurement data of a multi-source heterogeneous system and preprocess it, calculate the optimization objective function value of each candidate sensor combination, and determine the candidate sensor combination with the largest optimization objective function value as the sensor combination; The measurement data of the geometry and kinematics sensors and the solar sensor in the sensor array are preprocessed in a distributed low-level manner to obtain the local state estimates and their covariance matrices of each branch. The local state estimates and their covariance matrices of each branch and the measurement data of the space physics field sensor are input into the main filter. Based on the augmented state vector, evolutionary orbital dynamics constraints and strongly coupled measurement model, global elastic fusion is performed in the form of information filtering to perform centralized global joint estimation. Using the initial reference short-arc orbit in the global joint estimation as a priori, the physical field parameters are released and calibrated step by step according to the current observability conditions to obtain the error increments of the gravitational field parameters and the magnetic field parameters. The orbital dynamics model is updated based on the error increments of the gravitational field parameters, and the magnetic field measurement prediction model is updated based on the error increments of the magnetic field parameters to obtain the updated theoretical measurement data. The updated theoretical measurement data and the actual measurement data are subjected to consistency verification. When the verification quantity of the consistency check is less than or equal to the preset allowable threshold, the detector navigation status and the final calibrated physical field parameters are output. When the verification quantity of the consistency check exceeds the preset tolerance threshold, the fault tolerance isolation mechanism is triggered, the measurement noise covariance and fusion weight of the current physical field measurement data are adjusted, and global joint estimation and consistency check are performed based on subsequent measurement data.

2. The deep space autonomous navigation method of claim 1, wherein, The acquisition and preprocessing of measurement data from the multi-source heterogeneous system specifically includes: The measurement data of the multi-source heterogeneous system are analyzed using a state-space model, the expression of which is: ； in, The first derivative of the system state vector with respect to time. It is a vector-valued function. Let be the system state vector. For the system input vector, For the first The measurement output of each sensor, It is a nonlinear measurement function. and These are process noise and measurement noise, respectively. It is a time variable.

3. The deep space autonomous navigation method according to claim 2, characterized in that, The calculation of the optimization objective function value for each candidate sensor combination specifically includes: Fisher information matrices for each candidate sensor combination are constructed based on the motion state of the multi-source heterogeneous system, the Fisher information matrices, and the onboard computational and power consumption costs of each candidate sensor. Then, the calculation is performed using an optimization objective function. The expression is: ； in, For the candidate sensor set, This represents the Fisher information matrix for each candidate sensor. For the entire set of available sensors, To call the first The computational cost and power consumption cost of each sensor and These are adaptive weighting coefficients; The expression for the Fisher information matrix of each candidate sensor is as follows: ； in, Let Jacobian matrix be the equation of measurement. To measure the noise covariance matrix, For transpose; The Jacobian matrix is ​​obtained through the nonlinear measurement function. Taking the partial derivative with respect to the state, its expression is: ； in, For the current moment The system state estimate; The nonlinear measurement function is derived from the nonlinear measurement equation. The first-order expression is obtained by performing a Taylor expansion near the current state estimate: 。 4. The deep space autonomous navigation method according to claim 1, characterized in that, The distributed low-level preprocessing is accomplished through several locally distributed filters operating in parallel, and the locally distributed filters include: The first local distributed filter performs time integration and dimensionality reduction by pre-integrating the force and angular velocity measurements to obtain the estimated local relative orbital motion state. and the corresponding local covariance matrix ; The second local distributed filter, by establishing an attitude fusion filter loop, estimates and corrects the gyroscope constant bias and integral angle drift online and in real time, thus obtaining the inertial attitude state estimate. and the corresponding local attitude covariance matrix ; The third local distributed filter constructs a deterministic pointing constraint on a specific axis of the spacecraft using the solar direction vector. Combined with the pre-integrated angular velocity, it establishes a dimension-reduced filtering loop to correct the gyroscope's drift error in the direction perpendicular to the solar vector in real time, thus obtaining an estimate of the solar line-of-sight reference attitude state. and the corresponding local covariance matrix .

5. The deep space autonomous navigation method according to claim 1, characterized in that, The centralized global joint estimation specifically includes: Within the main filter, global elastic fusion is performed using information filtering. The update rule for global elastic fusion is as follows: ； in, and These are the estimated local optimal state and covariance matrix after dimensionality reduction, respectively. The prediction information matrix of the master filter before measurement update. The main filter's prediction information vector before measurement updates. The branch numbers of the local filters participating in global fusion. The total number of local filter branches participating in global fusion. Local adaptive fusion weights are assigned based on system observability optimization. For tensor measurement by gravitational gradiometer Vector measurement with magnetometer The global physical field measurement vector constituted and These are the corresponding global measurement Jacobian matrix and measurement noise covariance matrix, respectively; An elastic adjustment factor introduced for global measurement of physical fields. This is a transpose.

6. The deep space autonomous navigation method according to claim 1, characterized in that, The expression for the augmented state vector is: ； in, and These represent the position and velocity of the probe in the inertial frame of the central celestial body. For attitude quaternions, and These are the zero bias values ​​for the accelerometer and gyroscope, respectively. The increment of the gravitational field parameters of the target celestial body; The increment of the target celestial body's magnetic field parameter error. For transpose; The strongly coupled measurement model is constructed based on a gravitational gradiometer and a magnetometer: The tensor measurement equation of the gravitational gradiometer is: ； in, For the gravitational gradient tensor model of the target celestial body, These are the initial values ​​of the gravitational field parameters. For attitude quaternions, This is the attitude transformation matrix. To measure noise for the gravitational gradiometer; The vector measurement equation of the magnetometer is expressed as: ； in, For the target celestial body's magnetic field evolution vector model, These are the initial values ​​for the magnetic field parameters. This is for noise in the vector measurement of the magnetometer.

7. The deep space autonomous navigation method according to claim 1, characterized in that, The step-by-step release calibration of physical field parameters based on the current observability conditions specifically includes: In the initial calibration stage of the physical field parameters, the nominal value of the initial reference short arc segment orbit is fixed or the update range of the initial reference short arc segment orbit is limited, the orbit state covariance is preserved and its uncertainty is propagated to the physical field parameter estimation process. Based on the Fisher information matrix, physical field parameters whose observability meets the preset threshold are released, while physical field parameters whose observability does not meet the preset threshold are kept in the prior value until the observability meets the preset threshold.

8. The deep space autonomous navigation method according to claim 1, characterized in that, The update of the orbital dynamics model based on the error increment of gravitational field parameters and the update of the magnetic field measurement prediction model based on the error increment of magnetic field parameters specifically include: The incremental error of the gravitational field parameters is superimposed onto the prior nominal gravitational field model to update the orbital dynamics model; The magnetic field parameter error increment is updated into the target celestial body's magnetic field evolution model, thereby updating the magnetic field measurement prediction model.

9. The deep space autonomous navigation method according to claim 1, characterized in that, The consistency check includes: The updated theoretical measurement data is compared with the actual sensor observation data to construct the first... Integrated measurement residual vector at time step Based on the comprehensive measurement residual vector Calculate the normalized Mahalanobis distance test value And determine the allowable threshold based on the critical value of the chi-square distribution at the preset confidence level. ; The normalized Mahalanobis distance test value The expression is: ； in, The residual covariance matrix is... For the measurement matrix Jacobian, To predict the covariance matrix in one step, To measure the noise covariance matrix, This is the composite measurement residual vector; The comprehensive measurement residual vector The expression is: ； in, and These are the actual measurements from the gravitational gradiometer and magnetometer at the current moment, respectively. and These are the theoretically predicted measurements calculated based on the updated physical field parameters. For the first The augmented state vector estimate at time step 1; when At that time, output the detector navigation status. and the final calibrated higher-order physical field parameters ,in, This is the vector of spherical harmonic coefficients of the gravitational field of the target celestial body, as determined in the final calibration. This is the final calibrated parameter vector of the target celestial magnetic field dipole; when At this time, the fault-tolerant isolation mechanism is triggered, temporarily suspending the update of abnormal physical field parameters and maintaining the converged value of the previous cycle. The detector navigation state is output based on the local state estimates of the geometric and kinematic sensors. The covariance of the abnormal physical field measurement noise is increased and its fusion weight is applied. Global joint estimation and consistency verification are performed again using subsequent measurement data to obtain a new normalized Mahalanobis distance verification quantity. ; like To restore the update of abnormal physical field parameters, if The measurement channel for abnormal physical fields is cut off and the operation is downgraded.

10. A deep-space autonomous navigation system based on sensor combination optimization, used to execute the deep-space autonomous navigation method according to any one of claims 1-9, characterized in that, include: The multi-source heterogeneous sensor acquisition module is used to acquire measurement data from the multi-source heterogeneous system and perform preprocessing. The observation availability determination and target / measurement information optimization module is used to construct the Fisher information matrix of each candidate sensor combination. Based on the motion state of the multi-source heterogeneous system, the Fisher information matrix and the on-board computing cost and power consumption cost of each candidate sensor, the module calculates the optimization objective function value of each candidate sensor combination and determines the candidate sensor combination with the largest optimization objective function value as the sensor combination. The first-level local fusion module is used to perform distributed low-level preprocessing on the measurement data of the geometric and kinematic sensors and the measurement data of the solar sensor in the sensor combination to obtain the local state estimates and their covariance matrices of each branch. The second-level global joint estimation module is used to input the local state estimates of each branch and their covariance matrix and the raw data of the space physics field sensor into the main filter. Based on the augmented state vector, evolutionary orbital dynamics constraints and strongly coupled measurement model, it performs global elastic fusion in the form of information filtering to perform centralized global joint estimation. The physical field parameter estimation and model update module is used to take the initial reference short arc orbit in the global joint estimation as a priori, perform step-by-step release calibration of the physical field parameters according to the current observability conditions, obtain the error increments of the gravitational field parameters and the error increments of the magnetic field parameters, update the orbital dynamics model based on the error increments of the gravitational field parameters, update the magnetic field measurement prediction model based on the error increments of the magnetic field parameters, and obtain the updated theoretical measurement data. and The closed-loop consistency verification and navigation output module is used to verify the consistency between the updated theoretical measurement data and the actual measurement data; when the verification value of the consistency verification is less than or equal to the allowable threshold, the module outputs the detector navigation status and the final calibrated physical field parameters. When the verification quantity of the consistency check exceeds the allowable threshold, the fault-tolerant isolation mechanism is triggered, the measurement noise covariance and fusion weight of the current physical field measurement data are adjusted, and global joint estimation and consistency check are performed based on subsequent measurement data.

Images