Space object observation quality control method and system based on parameter adaptive estimation

Abstract

The application relates to the field of aerospace measurement technology and discloses a space target observation quality control method and system based on parameter adaptive estimation. The method comprises the following steps: obtaining observation data of a space target, establishing a mixed distribution observation model containing potential abnormal errors; constructing a latent variable structure for describing the existence state and distribution characteristics of the abnormal errors based on the mixed distribution observation model; iteratively estimating model parameters by using an expectation maximization algorithm, jointly estimating state parameters in the model and distribution parameters representing statistical characteristics of the abnormal errors by executing an expectation step and a maximization step through iteration; classifying and weighting the observation data according to the obtained abnormal posterior probability; performing parameter adaptive estimation based on the results of the classification and weighting processing, and outputting corrected space target state parameters, which are used for precise orbit determination or state estimation of the space target. The above method significantly improves the accuracy and stability of the precise orbit determination of the space target in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of aerospace measurement technology, and in particular to a method and system for quality control of space target observation based on parameter adaptive estimation. Background Technology

[0002] In critical missions such as space target surveillance, precise satellite orbit determination, and space debris cataloging, the quality of observation data directly determines the final accuracy of state estimation. Currently, least squares estimation based on Gaussian-Markov models remains the mainstream method, but its optimality heavily relies on the ideal assumption that observation noise follows a zero-mean Gaussian distribution. However, the real-world space observation environment is exceptionally complex, often causing this assumption to fail. Two main types of destructive errors exist: first, sudden gross errors, stemming from factors such as signal obstruction, multipath effects, and transient receiver malfunctions, manifesting as discrete, large-amplitude outliers; second, unmodeled systematic biases, such as residual atmospheric propagation delays, instrument phase center deviations, and dynamic model errors. These errors may have time correlations or specific patterns, easily leading to systematic shifts in the estimation results.

[0003] Traditional quality control methods, such as statistically based data probing and iterative data probing, identify and eliminate gross errors by constructing alternative hypotheses. However, these methods based on generalized hypothesis testing have inherent limitations: they rely on predefined, discrete error pattern hypotheses and select models through hypothesis testing. When error patterns are unknown, complex, or numerous, the number of hypothesis combinations that need to be enumerated explodes, resulting in a heavy computational burden and making it difficult to continuously estimate the statistical characteristics of errors (such as probability of occurrence and magnitude distribution).

[0004] In recent years, another technical approach based on probabilistic generation models and adaptive parameter estimation has demonstrated unique advantages. The core idea of ​​this method is to treat anomalies as part of the data generation process and describe their overall statistical regularity using continuous probability distribution parameters (such as anomaly occurrence rate, mean, and variance). The Bernoulli-Gaussian model is a typical representative of this type of model, modeling the anomaly error of each observation as the product of a Bernoulli random variable and a Gaussian random variable. The expectation-maximization algorithm provides an efficient parameter estimation framework for this mixed model containing latent variables. However, successfully applying this theoretical framework to the dynamic, multi-source, and high-dimensional observation scenarios of space targets still faces challenges in several aspects, including the gap between the model and practical applications, the difficulty of multi-source data fusion, the adaptability of dynamic estimation, and the estimation efficiency and stability. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a parameter adaptive estimation method based on the Bernoulli-Gaussian model and the expectation-maximization algorithm. The aim is to improve the accuracy and robustness of spatial target state estimation through joint adaptive parameter estimation.

[0006] On the one hand, this invention provides a method for quality control of space target observations based on adaptive parameter estimation, comprising the following steps: S1: Acquire observation data of space targets and establish a hybrid distribution observation model that includes potential anomaly errors. The hybrid distribution observation model describes the observation data as a hybrid probability distribution composed of normal components and anomaly components. S2: Based on the hybrid distribution observation model, construct a latent variable structure to describe the existence state and distribution characteristics of abnormal errors; S3: The expectation-maximization algorithm is used to iteratively estimate the model parameters. By iteratively executing the expectation step and the maximization step, the state parameters and the distribution parameters that characterize the statistical properties of abnormal errors in the model are jointly estimated. S4: Based on the abnormal posterior probability obtained after iterative convergence, classify and weight the observed data to suppress the impact of abnormal observations on state estimation. S5: Based on the results of classification and weighted processing, perform adaptive parameter estimation and output the corrected space target state parameters for precise orbit determination or state estimation of space targets.

[0007] Further, in step S1, establishing a mixed distribution observation model that includes potential anomaly errors includes: Establish a mixed distribution observation model ,in For the observation vector of the space target, To design the matrix, For the state parameters to be estimated, Let be a random error vector. This is the abnormal error vector; For the Observations Its abnormal error Follows a Bernoulli-Gaussian distribution: ,in This is a Bernoulli variable, indicating whether an exception has occurred. is a Gaussian variable representing the magnitude of the outlier.

[0008] Further, in step S2, constructing the latent variable structure for describing the existence state and distribution characteristics of abnormal errors specifically includes: The Bernoulli variable Set to obey parameter as The Bernoulli distribution is denoted as . ,in This indicates the estimated rate of occurrence of the anomaly. The Gaussian variable Set to follow the mean variance is The normal distribution is denoted as . ; Therefore, each observation value The probability of an anomaly is controlled by a Bernoulli distribution, and the magnitude of the anomaly is controlled by a Gaussian distribution.

[0009] Furthermore, before performing the iterative estimation of model parameters in step S3, the model parameters are also simplified through engineering processes. For observations of the same type, it is assumed that they have the same anomaly rate, i.e. The variance of random noise is decomposed into known cofactors. With the unit weight variance factor to be estimated The product of, i.e. ; Decompose the variance of the outlier magnitude into known cofactors. With the outlier factor to be estimated The product of, i.e. ,in Can be set to be with equal.

[0010] Further, in step S3, the iterative estimation of model parameters using the expectation-maximization algorithm specifically includes: In the expectation step, each observation is calculated based on the current model parameter estimates. Posterior probability of belonging to an outlier The calculation formula is as follows: , in, For designing a matrix The OK, The probability density function representing the normal distribution; In the maximization step, based on all calculated posterior probabilities... The model parameters, including state parameters, are updated by maximizing the expectation of the full data likelihood function. Unit weight variance factor Abnormality rate Abnormal error mean and the variance factor of outlier ; Repeat the expectation step and the maximization step until the model parameters converge.

[0011] Further, in step S4, the observed data are classified and weighted based on the anomalous posterior probabilities obtained after iterative convergence, including: After the expectation-maximization algorithm converges iteratively, each observation value is obtained. The corresponding final anomaly posterior probability ; The final anomaly posterior probability This serves as the anomaly confidence level for the observation, used for subsequent processing; or a discrimination threshold can be set when... If the value exceeds the threshold, the observation is identified and marked as an abnormal observation. The converged anomalous posterior probability Alternatively, their transformed values ​​can be used as weights to construct a weight matrix, which is used to perform weighted estimation of the original observation equation.

[0012] Further, in step S5, adaptive parameter estimation is performed based on the results of classification and weighting, and the corrected spatial target state parameters are output, including: The converged anomalous posterior probability The parameters are converted into observation weights, and then weighted least squares estimation is performed on the original observation equation using these weights to calculate the corrected space target state parameters. ; Alternatively, the optimal model parameters obtained after the expectation-maximization algorithm converges can be used as the true description of the current dataset. Under this optimal mixture distribution model, the maximum likelihood estimate of the state parameters can be recalculated, and a highly robust solution for the state parameters can be output. .

[0013] Preferably, before performing the iterative estimation of model parameters in step S3, the parameters are initialized, including: Initial values ​​for state parameters are provided using least squares estimation or median estimation, while outlier parameters are set based on prior information or empirical values. The method supports two modes: static batch processing and dynamic sliding window processing. It achieves recursive processing of time series data by constructing an estimation equation for each observation.

[0014] More preferably, the method further includes: By setting independent anomaly parameters for different types of observation data, including anomaly incidence rate Abnormal error mean and abnormal error variance To adapt to multi-source observation systems.

[0015] On the other hand, the present invention also provides a space target observation quality control system based on parameter adaptive estimation, comprising: The data acquisition and modeling module is used to acquire observation data of space targets and establish a hybrid distribution observation model that includes potential anomaly errors. The hybrid distribution observation model describes the observation data as a hybrid probability distribution composed of normal components and anomaly components. The latent variable construction module is used to construct a latent variable structure that describes the existence state and distribution characteristics of abnormal errors based on the hybrid distribution observation model. The joint parameter estimation module is used to iteratively estimate model parameters using the expectation-maximization algorithm. By iteratively executing the expectation step and the maximization step, it jointly estimates the state parameters and the distribution parameters that characterize the statistical properties of abnormal errors in the model. The anomaly handling module is used to classify and weight the observation data based on the anomaly posterior probability obtained after iterative convergence, so as to suppress the impact of anomaly observations on state estimation. The state calculation and output module is used to perform adaptive parameter estimation based on the results of classification and weighted processing, and output the corrected space target state parameters for precise orbit determination or state estimation of space targets.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention uses the expectation-maximization algorithm to learn the probability and size distribution of anomalies online, which can adapt to complex and ever-changing observation environments without pre-specifying the anomaly pattern or number. This invention uses a soft-weighting method to process suspicious observations, which retains effective information while suppressing the impact of anomalies. Compared with the traditional hard-rejection method, it can achieve better state estimation accuracy. In addition to outputting high-precision state parameters, this invention also provides the posterior probability of anomalies for each observation, providing a diagnostic basis for data quality assessment and system health monitoring; This invention is based on single-observation modeling and can be flexibly applied to static batch processing and dynamic recursive processing to meet different needs for precise orbit determination and real-time state estimation of space targets. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a space target observation quality control method based on parameter adaptive estimation according to the present invention; Figure 2 A comparison of the fitting effects of the Gaussian model and the mixed distribution observation model of this invention on different samples; Figure 3This is a comparison chart showing the true and estimated values ​​of parameters for different samples obtained using the LS least squares method, IDS iterative data probing method, and EM expectation-maximization algorithm. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The specific embodiments of the present invention will be described below with reference to the accompanying drawings and examples.

[0020] Example 1 Please see Figure 1 This embodiment provides a technical solution for a space target observation quality control method based on parameter adaptive estimation. Specifically, it provides a complete and operable implementation process for a space target observation data quality analysis method based on the Bernoulli-Gaussian (BG) model and the expectation-maximization (EM) algorithm, including the following steps: S1: Acquire observation data of space targets and establish a hybrid distribution observation model that includes potential anomaly errors. The hybrid distribution observation model describes the observation data as a hybrid probability distribution composed of normal components and anomaly components. S2: Based on the hybrid distribution observation model, construct a latent variable structure to describe the existence state and distribution characteristics of abnormal errors; S3: The expectation-maximization algorithm is used to iteratively estimate the model parameters. By iteratively executing the expectation step and the maximization step, the state parameters and the distribution parameters that characterize the statistical properties of abnormal errors in the model are jointly estimated. S4: Based on the abnormal posterior probability obtained after iterative convergence, classify and weight the observed data to suppress the impact of abnormal observations on state estimation. S5: Based on the results of classification and weighted processing, perform adaptive parameter estimation and output the corrected space target state parameters for precise orbit determination or state estimation of space targets.

[0021] The first step is the acquisition and preprocessing of observation data in step S1. Specifically, in this embodiment, GNSS observation data files (sampling rate 30 seconds) for a specific low-Earth orbit (LRO) satellite for a given 24 hours on a given day, along with precise ephemeris and clock bias products, are obtained from the International GNSS Service (IGS) data center. Through data preprocessing, invalid epochs and satellites are removed, and the prior residuals for each observation (such as GPS L1C / A code pseudorange) are calculated to preliminarily identify possible cycle slips and gross errors. Ultimately, a data set containing... Each valid epoch contains 1 epoch, and each epoch contains 1 epoch. Observation vector of satellite observation And construct a design matrix based on satellite dynamics and observation geometry. State parameters to be estimated This includes the satellite's three-dimensional position, velocity, and receiver clock bias parameters.

[0022] Based on the preprocessed data, we construct a Bernoulli-Gaussian mixture distribution observation model, including: Establish a mixed distribution observation model ,in For the observation vector of the space target, To design the matrix, For the state parameters to be estimated, Let the random error vector follow a zero-mean Gaussian distribution. For the anomaly error vector, the key step here is to introduce a Bernoulli-Gaussian model to describe the anomaly error. ; For the Observations Its abnormal error Follows a Bernoulli-Gaussian distribution: ,in This is a Bernoulli variable, indicating whether an exception has occurred. is a Gaussian variable representing the magnitude of the outlier.

[0023] Based on this mixed distribution observation model, we construct the latent variable structure for step S2 to describe the existence state and distribution characteristics of anomalous errors, specifically including: The Bernoulli variable Set to obey parameter as The Bernoulli distribution is denoted as . ,in This indicates the estimated rate of occurrence of the anomaly. The Gaussian variable Set to follow the mean variance is The normal distribution is denoted as . , and The mean and variance of the outlier to be estimated are used to describe the statistical characteristics of the outlier error.

[0024] Therefore, each observation value The probability of an anomaly is controlled by a Bernoulli distribution, and the magnitude of the anomaly is controlled by a Gaussian distribution.

[0025] In this embodiment, when When, it indicates that the observation has been affected by an anomaly; when When this time, it indicates that the observation is normal. The model treats the observed data as a mixture of two Gaussian components: one being... For the expected value and variance are... The normal component; the other is... For the expected value and variance are... The abnormal components, the mixing weights are determined by Decide.

[0026] Next, to improve the model's estimability and computational efficiency, before performing the iterative estimation of model parameters in step S3, an engineering simplification of the model parameters is also included: For observations of the same type, it is assumed that they have the same anomaly rate (in this case, all GPS L1 pseudoranges), i.e. The variance of random noise is decomposed into known cofactors. With the unit weight variance factor to be estimated The product of, i.e. ; Decompose the variance of the outlier magnitude into known cofactors. With the outlier factor to be estimated The product of, i.e. ,in Can be set to be with equal.

[0027] Simultaneously, before iteratively estimating the model parameters in step S3, the parameters are initialized, including: Initial values ​​for state parameters are provided using least squares estimation or median estimation, while outlier parameters are set based on prior information or empirical values. The method supports two modes: static batch processing and dynamic sliding window processing. It achieves recursive processing of time series data by constructing an estimation equation for each observation.

[0028] Specifically, the following strategy is used to initialize the parameters: state parameters The standard least squares method was used to perform preliminary calculations on the original observations, and the results were used as... Variance factor: Let (Smaller value) (Larger values) enable the algorithm to attribute large residuals to anomalies in the early stages; anomaly parameter: let (Assuming the anomaly rate is low) (Assuming the anomaly is unbiased); Convergence settings: Set the maximum number of iterations. Convergence threshold .

[0029] Based on this, the adaptive iterative estimation of parameters for the expectation-maximization algorithm (hereinafter referred to as the EW algorithm) in step S3 is performed, specifically including: In the expectation step, each observation is calculated based on the current model parameter estimates. Posterior probability of belonging to an outlier The calculation formula is as follows: , in, For designing a matrix The OK, The probability density function representing the normal distribution; In the maximization step, based on all calculated posterior probabilities... The model parameters, including state parameters, are updated by maximizing the expectation of the full data likelihood function. Unit weight variance factor Abnormality rate Abnormal error mean and the variance factor of outlier ; Repeat the expectation step and the maximization step until the model parameters converge.

[0030] Specifically, we set independent anomaly parameters for different types of observation data, including anomaly occurrence rate. Abnormal error mean and abnormal error variance To adapt to multi-source observation systems.

[0031] Wherein, the E-step (expectation step): estimated based on the current parameters. Calculate each observation value Posterior probability of belonging to an outlier The calculation formula is based on Bayes' theorem, with the numerator being the weighted likelihood of the outlier components and the denominator being the total likelihood (the sum of the normal and outlier components). The value, between 0 and 1, quantifies the probability that the observation is an anomaly; M-step (maximization step): using the calculated... As weights, update all model parameters: state parameters By solving a problem... and The equation for the weighted design matrix is ​​used to update the variance factor. and Updated based on the weighted sum of squared residuals of normal and outlier components respectively. Outlier occurrence rate. Update to all The arithmetic mean.

[0032] Next, based on the anomalous posterior probabilities obtained after iterative convergence in step S3, step S4 involves classifying and weighting the observed data, including: After the expectation-maximization algorithm converges iteratively, each observation value is obtained. The corresponding final anomaly posterior probability ; The final anomaly posterior probability This serves as the anomaly confidence level for the observation, used for subsequent processing; or a discrimination threshold can be set when... If the value exceeds the threshold, the observation is identified and marked as an abnormal observation. The converged anomalous posterior probability Alternatively, their transformed values ​​can be used as weights to construct a weight matrix, which is used to perform weighted estimation of the original observation equation.

[0033] Specifically, the mean anomaly size is calculated from the weighted residuals of the anomaly components; convergence criterion is determined by calculating the parameter update amount. If less than the threshold or reach If the iteration stops, the final parameters are output. and the final anomaly probability for each observation Otherwise, let Return to step E and continue iterating.

[0034] Then, step S5 is performed for data quality control and final state calculation. This involves adaptive parameter estimation based on the results of classification and weighting, outputting the corrected spatial target state parameters, including: The converged anomalous posterior probability The parameters are converted into observation weights, and then weighted least squares estimation is performed on the original observation equation using these weights to calculate the corrected space target state parameters. ; Alternatively, the optimal model parameters obtained after the expectation-maximization algorithm converges can be used as the true description of the current dataset. Under this optimal mixture distribution model, the maximum likelihood estimate of the state parameters can be recalculated, and a highly robust solution for the state parameters can be output. .

[0035] Specifically, in this embodiment, the data quality control and final state calculation include: (1) Anomaly identification and labeling: Set a probability threshold (e.g., 0.5). For each observation, if If an anomaly is detected, it is marked as an "abnormal observation"; otherwise, it is marked as a "normal observation." Simultaneously, an estimated global anomaly rate is output. .

[0036] (2) Adaptive Robust Estimation: Instead of directly removing labeled outlier observations, optimal estimation is performed based on the final mixture model. Specifically, using... Construct a diagonal weight matrix as the weights for each observation. Solve the weighted least squares problem: , This explanation This refers to the precise satellite orbital state parameters obtained after quality control using the BG-EM method, suppressing the effects of anomalies. Final output. And its accuracy assessment information.

[0037] To verify the beneficial effects of the above method, experimental comparison data and analysis are provided between the method of this invention and existing methods. We compare the space target observation quality control method (BG-EM) based on parameter adaptive estimation of this invention with the following two commonly used methods: Traditional Least Squares (LS): Ignores outliers and directly adjusts all observations. Iterative Data Detection (IDS): Typically detects outliers by calculating the standardized residuals of each observation and comparing them to a pre-defined critical value. These outliers are then removed from the dataset (or downweighted), and the remaining (or reweighted) observations are used for a new least squares estimate. This process is repeated iteratively until no new outliers are detected at the current significance level.

[0038] like Figure 2 As shown, when the total error rate and total error size are relatively small, the PDF estimated by both the Gaussian model and the mixture model are close to the histogram, showing satisfactory performance in fitting these samples. As the total error rate and total error size increase (see Samples 2 and 3), the PDF estimated by the Gaussian model plus LS is significantly distorted compared to the histogram. The Gaussian model loses its ability to fit samples contaminated by coarse errors of a large proportion and magnitude. In contrast, the PDF of the mixture model is always close to the histogram and maintains good fitting performance.

[0039] Furthermore, the following is a comparison between the present invention and a mixture model using three different parameter values, and a conventional Gaussian model and corresponding model parameters used to calculate LS, such as... Figure 3 This provides the true and estimated values ​​of the LS and EM parameters under different conditions. From... Figure 3The results show that when the total error rate and total error size are relatively small (see Sample 1), there is no significant difference between the true values ​​and estimated parameters of LS and EM. As the total error rate and total error size increase (see Samples 2 and 3), the accuracy of LS estimation decreases significantly due to a lack of robustness. Conversely, the parameters estimated by EM are less affected by outliers and remain stable. In fact, when the proportion of outliers is large, the accuracy of the gross error parameters even improves.

[0040] The comparative experiments in this embodiment fully demonstrate that the data quality analysis method based on the BG model and EM algorithm proposed in this invention has significant advantages over the traditional least squares method and iterative data detection method in terms of state estimation accuracy, anomaly identification accuracy, and learning ability of anomaly statistical characteristics when there are abnormal observations. It can provide a better and more reliable data quality control means for precise orbit determination and state estimation of space targets.

[0041] Based on the above-described method of the present invention, a space target observation quality control system based on parameter adaptive estimation is provided, comprising: The data acquisition and modeling module is used to acquire observation data of space targets and establish a hybrid distribution observation model that includes potential anomaly errors. The hybrid distribution observation model describes the observation data as a hybrid probability distribution composed of normal components and anomaly components. The latent variable construction module is used to construct a latent variable structure that describes the existence state and distribution characteristics of abnormal errors based on the hybrid distribution observation model. The joint parameter estimation module is used to iteratively estimate model parameters using the expectation-maximization algorithm. By iteratively executing the expectation step and the maximization step, it jointly estimates the state parameters and the distribution parameters that characterize the statistical properties of abnormal errors in the model. The anomaly handling module is used to classify and weight the observation data based on the anomaly posterior probability obtained after iterative convergence, so as to suppress the impact of anomaly observations on state estimation. The state calculation and output module is used to perform adaptive parameter estimation based on the results of classification and weighted processing, and output the corrected space target state parameters for precise orbit determination or state estimation of space targets.

[0042] It should be noted that the steps in the space target observation quality control method based on parameter adaptive estimation provided in this embodiment can be implemented based on the corresponding modules in the space target observation quality control system based on parameter adaptive estimation. Those skilled in the art can refer to the technical solution of the system to implement the steps of the method. That is, the embodiments in the system can be understood as preferred examples of implementing the method, and will not be elaborated here.

[0043] Besides implementing the system and its various devices provided by this invention in purely computer-readable program code, the same functions can be achieved by logically programming the method steps, making the system and its various devices of this invention appear as logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices provided by this invention can be considered as a hardware component, and the devices included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0044] Finally, it should be noted that the above description is only a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be pointed out that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

[0045] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for quality control of space target observations based on adaptive parameter estimation, characterized in that, Includes the following steps: S1: Acquire observation data of space targets and establish a hybrid distribution observation model that includes potential anomaly errors. The hybrid distribution observation model describes the observation data as a hybrid probability distribution composed of normal components and anomaly components. S2: Based on the hybrid distribution observation model, construct a latent variable structure to describe the existence state and distribution characteristics of abnormal errors; S3: The expectation-maximization algorithm is used to iteratively estimate the model parameters. By iteratively executing the expectation step and the maximization step, the state parameters and the distribution parameters that characterize the statistical properties of abnormal errors in the model are jointly estimated. S4: Based on the abnormal posterior probability obtained after iterative convergence, classify and weight the observed data to suppress the impact of abnormal observations on state estimation. S5: Based on the results of classification and weighted processing, perform adaptive parameter estimation and output the corrected space target state parameters for precise orbit determination or state estimation of space targets.

2. The space target observation quality control method based on parameter adaptive estimation according to claim 1, characterized in that, In step S1, establishing a mixed distribution observation model that includes potential anomaly errors includes: Establish a mixed distribution observation model ,in For the observation vector of the space target, To design the matrix, For the state parameters to be estimated, Let be a random error vector. This is the abnormal error vector; For the Observations Its abnormal error Follows a Bernoulli-Gaussian distribution: ,in This is a Bernoulli variable, indicating whether an exception has occurred. is a Gaussian variable representing the magnitude of the outlier.

3. The space target observation quality control method based on parameter adaptive estimation according to claim 2, characterized in that, In step S2, constructing the latent variable structure for describing the existence state and distribution characteristics of abnormal errors specifically includes: The Bernoulli variable Set to obey parameter as The Bernoulli distribution is denoted as . ,in This indicates the estimated rate of occurrence of the anomaly. The Gaussian variable Set to follow the mean variance is The normal distribution is denoted as . ; Therefore, each observation value The probability of an anomaly is controlled by a Bernoulli distribution, and the magnitude of the anomaly is controlled by a Gaussian distribution.

4. The space target observation quality control method based on parameter adaptive estimation according to claim 3, characterized in that, Before performing the iterative estimation of model parameters in step S3, the model parameters are further simplified through engineering: For observations of the same type, it is assumed that they have the same anomaly rate, i.e. The variance of random noise is decomposed into known cofactors. With the unit weight variance factor to be estimated The product of, i.e. ; Decompose the variance of the outlier magnitude into known cofactors. With the outlier factor to be estimated The product of, i.e. ,in Can be set to be with equal.

5. The space target observation quality control method based on parameter adaptive estimation according to claim 4, characterized in that, In step S3, the iterative estimation of model parameters using the expectation-maximization algorithm specifically includes: In the expectation step, each observation is calculated based on the current model parameter estimates. Posterior probability of belonging to an outlier The calculation formula is as follows: ， in, For designing a matrix The OK, The probability density function representing the normal distribution; In the maximization step, based on all calculated posterior probabilities... The model parameters, including state parameters, are updated by maximizing the expectation of the full data likelihood function. Unit weight variance factor Abnormality rate Abnormal error mean and the variance factor of outlier ; Repeat the expectation step and the maximization step until the model parameters converge.

6. The space target observation quality control method based on parameter adaptive estimation according to claim 5, characterized in that, In step S4, the observed data are classified and weighted based on the anomalous posterior probabilities obtained after iterative convergence, including: After the expectation-maximization algorithm converges iteratively, each observation value is obtained. The corresponding final anomaly posterior probability ; The final anomaly posterior probability This serves as the anomaly confidence level for the observation, used for subsequent processing; or a discrimination threshold can be set when... If the value exceeds the threshold, the observation is identified and marked as an abnormal observation. The converged anomalous posterior probability Alternatively, their transformed values ​​can be used as weights to construct a weight matrix, which is used to perform weighted estimation of the original observation equation.

7. The space target observation quality control method based on parameter adaptive estimation according to claim 6, characterized in that, In step S5, adaptive parameter estimation is performed based on the results of classification and weighting, and the corrected spatial target state parameters are output, including: The converged anomalous posterior probability The parameters are converted into observation weights, and then weighted least squares estimation is performed on the original observation equation using these weights to calculate the corrected space target state parameters. ; Alternatively, the optimal model parameters obtained after the expectation-maximization algorithm converges can be used as the true description of the current dataset. Under this optimal mixture distribution model, the maximum likelihood estimate of the state parameters can be recalculated, and a highly robust solution for the state parameters can be output. .

8. The space target observation quality control method based on parameter adaptive estimation according to claim 1, characterized in that, Before performing iterative estimation of model parameters in step S3, the parameters are initialized, including: Initial values ​​for state parameters are provided using least squares estimation or median estimation, while outlier parameters are set based on prior information or empirical values. The method supports two modes: static batch processing and dynamic sliding window processing. It achieves recursive processing of time series data by constructing an estimation equation for each observation.

9. The space target observation quality control method based on parameter adaptive estimation according to claim 1, characterized in that, The method further includes: By setting independent anomaly parameters for different types of observation data, including anomaly incidence rate Abnormal error mean and abnormal error variance To adapt to multi-source observation systems.

10. A space target observation quality control system based on parameter adaptive estimation, characterized in that, include: The data acquisition and modeling module is used to acquire observation data of space targets and establish a hybrid distribution observation model that includes potential anomaly errors. The hybrid distribution observation model describes the observation data as a hybrid probability distribution composed of normal components and anomaly components. The latent variable construction module is used to construct a latent variable structure that describes the existence state and distribution characteristics of abnormal errors based on the hybrid distribution observation model. The joint parameter estimation module is used to iteratively estimate model parameters using the expectation-maximization algorithm. By iteratively executing the expectation step and the maximization step, it jointly estimates the state parameters and the distribution parameters that characterize the statistical properties of abnormal errors in the model. The anomaly handling module is used to classify and weight the observation data based on the anomaly posterior probability obtained after iterative convergence, so as to suppress the impact of anomaly observations on state estimation. The state calculation and output module is used to perform adaptive parameter estimation based on the results of classification and weighted processing, and output the corrected space target state parameters for precise orbit determination or state estimation of space targets.

Images