A method for detecting the attitude determination state of a spacecraft based on constructing an attitude regression model

Abstract

The present invention relates to a method for detecting the attitude determination state of a spacecraft based on constructing an attitude regression model. The method includes: determining the attitude control actuators and attitude control measurement units of the spacecraft, and obtaining the corresponding parameters to be detected and historical parameters; generating set data through normalization processing; splitting the set data into a training set and training it according to a decision tree model, verifying the training result with a test set, and using the trained decision tree model as the attitude regression model corresponding to the spacecraft; generating data to be detected through normalization processing; substituting the data to be detected into the attitude regression model and detecting the attitude determination state of the spacecraft. By preprocessing the parameters of the spacecraft and specifically constructing a suitable attitude regression model, the present invention enables the detection of the attitude determination state to be applicable to spacecraft using different attitude control actuators and attitude control measurement units; and improves the versatility of the detection of the attitude determination state on the basis of high accuracy.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of data processing in spacecraft TT&C management, and in particular, to a method for detecting the attitude determination state of a spacecraft based on constructing an attitude regression model. Background Art

[0002] When a spacecraft enters a predetermined orbit for operation, sometimes attitude adjustment is required to meet the pointing requirements of payload work. When the satellite attitude makes a small-range adjustment, it is called satellite attitude adjustment, and when the satellite attitude makes a large-range maneuver, it is called satellite attitude maneuver. However, whether it is attitude adjustment or attitude maneuver, it reflects the corresponding relationship between the control amount applied to the attitude control actuator and the output value of the control result, and this corresponding relationship is mainly closely related to the execution ability of the actuator and the attitude dynamics of the satellite in the space environment. However, attitude adjustment / maneuver is a complex mechanism process, which requires integrating the telemetry data of control amounts such as the current spacecraft's space attitude sensors, controllers, and actuators, and through a series of calculations, the data of attitude control single machines such as star sensors, sun sensors, and gyroscopes on the satellite are obtained.

[0003] Regarding the above technical solution, the inventors found that at least the following technical problems exist:

[0004] It is difficult to detect abnormal spacecraft attitudes by establishing a model of the attitude adjustment / maneuver mechanism. Due to the different design mechanisms of spacecraft, the universality of the mechanism model is not strong.

[0005] Therefore, it is necessary to improve one or more problems existing in the above related technical solutions.

[0006] It should be noted that this part aims to provide background or context for the technical solution of the present invention stated in the claims. The description herein is not admitted to be prior art just because it is included in this part. Summary of the Invention

[0007] The purpose of the present invention is to provide a method for detecting the attitude determination state of a spacecraft based on constructing an attitude regression model, so as to at least to a certain extent solve one or more problems caused by the limitations and defects of the related art.

[0008] The present invention provides a method for detecting the attitude determination state of a spacecraft based on constructing an attitude regression model, including:

[0009] Determine the attitude control actuator and attitude control measurement single machine of the spacecraft, and obtain the corresponding parameters to be detected and historical parameters;

[0010] Use the historical parameters of the attitude control actuator as input and the historical parameters of the attitude control measurement single machine as output, and generate set data through normalization processing;

[0011] Split a part of the set data as the training set and train it according to the decision tree model. Then, use another part of the split set data as the test set to verify the training result, and use the trained decision tree model as the attitude regression model corresponding to the spacecraft.

[0012] Use the parameters to be detected of the attitude control actuator as the input and the parameters to be detected of the attitude control measurement unit as the output, and generate the data to be detected through normalization processing.

[0013] Substitute the data to be detected into the attitude regression model and detect the attitude stabilization state of the spacecraft.

[0014] Optionally, the step of using the historical parameters of the attitude control actuator as the input and the historical parameters of the attitude control measurement unit as the output, and generating the set data through normalization processing includes:

[0015] Use the input historical parameters and the output historical parameters in one-to-one correspondence as sample data, perform data outlier rejection according to the sample standard deviation and the sample mean, and perform normalization processing on the outlier-rejected sample data to generate the set data.

[0016] Optionally, the step of splitting a part of the set data as the training set and training it according to the decision tree model includes:

[0017] Define the loss function through the mean squared error, update the training data according to the gradient of the loss function, and substitute the updated training data into the decision tree model for training, where the mean of the samples is used as the initial value.

[0018] Optionally, the step of then using another part of the split set data as the test set to verify the training result includes:

[0019] Use the test set to evaluate the model accuracy of the trained decision tree model, and calculate the determination coefficient of the decision tree model; if the determination coefficient is less than the preset threshold, retrain; if the determination coefficient is greater than or equal to the preset threshold, use the trained decision tree model as the attitude regression model corresponding to the spacecraft.

[0020] Optionally, the step of splitting a part of the set data as the training set and training it according to the decision tree model, then using another part of the split set data as the test set to verify the training result, and using the trained decision tree model as the attitude regression model corresponding to the spacecraft includes:

[0021] Split the set data into the training set and the test set according to a ratio of 7:3.

[0022] Optionally, the steps of splitting a part of the set data as a training set and training it according to a decision tree model, and then using another part of the split set data as a test set to verify the training result, and using the trained decision tree model as the attitude regression model corresponding to the spacecraft include:

[0023] Split the set data into the training set, the test set and the validation set according to the ratio of 6:2:2;

[0024] During the training according to the decision tree model, verify the loss function threshold according to the validation set every time a training is completed, and complete the training when the loss function is less than the validation threshold.

[0025] Optionally, the steps of substituting the data to be detected into the attitude regression model and detecting the attitude determination state of the spacecraft include:

[0026] Set a detection threshold and a detection step size;

[0027] Input the input data of the data to be detected into the attitude regression model to generate prediction data;

[0028] Compare the output data of the data to be detected with the prediction data;

[0029] When the comparison result exceeds the detection threshold and meets the detection step size, it is determined that the attitude determination is abnormal.

[0030] Optionally, the steps of substituting the data to be detected into the attitude regression model and detecting the attitude determination state of the spacecraft include:

[0031] If successive

[0032]

[0033] where is the k-th component of the prediction data generated by the attitude regression model ; is the k-th component of the output data of the data to be detected ; is the feature dimension of the output data; is the detection threshold; is the detection step size.

[0034] Optionally, the steps of substituting the data to be detected into the attitude regression model and detecting the attitude determination state of the spacecraft include:

[0035] Evaluate the detection accuracy of the attitude determination model based on the data to be detected and the predicted data; the attitude determination model is used to control the attitude state of the spacecraft.

[0036] Optionally, the steps of determining the attitude control actuator and the attitude control measurement unit of the spacecraft and obtaining the corresponding parameters to be detected and historical parameters include:

[0037] The parameters of the attitude control actuator are at least one of the following: reaction wheel, magnetic torque actuator, thruster, balance wheel, orbit control subsystem, on-board time or orbit parameters;

[0038] The parameters of the attitude control measurement unit are at least one of the following: sun sensor, star sensor, gyroscope or magnetometer parameters.

[0039] The technical solution provided by the present invention may include the following beneficial effects:

[0040] In the present invention, by preprocessing the parameters of the spacecraft and constructing an adapted attitude regression model accordingly, the detection of the attitude determination state can be applicable to spacecrafts using different attitude control actuators and attitude control measurement units; the detection of the attitude determination state improves the generality on the basis of high accuracy. Description of the Drawings

[0041] The drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0042] Figure 1 A flowchart showing the method for detecting the attitude determination state of a spacecraft based on constructing an attitude regression model in an exemplary embodiment of the present invention;

[0043] Figure 2 A schematic diagram showing the judgment logic of the method for detecting the attitude determination state of a spacecraft based on constructing an attitude regression model in an exemplary embodiment of the present invention;

[0044] Figure 3 A schematic diagram showing the abnormal detection result of star sensor quaternion in an exemplary embodiment of the present invention. Detailed Embodiments

[0045] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art. The features, structures, or characteristics described may be combined in any suitable manner in one or more embodiments.

[0046] In addition, the accompanying drawings are only schematic illustrations of the embodiments of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

[0047] The present invention provides a method for detecting the attitude determination state of a spacecraft based on constructing an attitude regression model, as shown in Figure 1 the following steps are included:

[0048] Step S100: Determine the attitude control actuators and attitude control measurement units of the spacecraft, and obtain the corresponding parameters to be detected and historical parameters.

[0049] Step S200: Use the historical parameters of the attitude control actuators as inputs and the historical parameters of the attitude control measurement units as outputs, and generate set data through normalization processing.

[0050] Step S300: Split part of the set data as a training set and train it according to the decision tree model, then use the other part of the split set data as a test set to verify the training results, and use the trained decision tree model as the attitude regression model corresponding to the spacecraft.

[0051] Step S400: Use the parameters to be detected of the attitude control actuators as inputs and the parameters to be detected of the attitude control measurement units as outputs, and generate data to be detected through normalization processing.

[0052] Step S500: Substitute the data to be detected into the attitude regression model and detect the attitude determination state of the spacecraft.

[0053] It should be understood that, from the perspective of data analysis, according to the actual data characteristics, a mathematical model for attitude determination regression based on a general process is established, and through data-driven training, a model that fits the attitude determination mechanism can provide technical support for the detection of spacecraft attitude anomalies, which has important practical significance for ensuring the normal operation of the spacecraft in orbit.

[0054] It should also be understood that the normalization process of the data to be detected is the same as that of the set data.

[0055] It should also be understood that the parameters to be detected and the historical parameters corresponding to the attitude control measurement unit refer to the relevant measurement values and their increment values of the attitude control measurement unit.

[0056] It should also be understood that after determining the input data and output data, it is necessary to preprocess the input data and output data to generate set data, and this data preprocessing includes data selection, data wild value elimination, data normalization, and training verification test data division.

[0057] It should also be understood that after the calculation is completed, the input and output parameters are restored to the original range. The inverse normalization method for the input data is ; the processing process of the output data is similar. The input parameter , where represents the telemetry value of the i-th parameter in the input quantity at time t, and the output sample , where, represents the telemetry value of the j-th parameter in the output quantity at time t. After inverse normalizing the input and output data, the inverse-normalized data (that is, the data restored to the original scale range of the parameter) is compared with the data to be detected, and further anomaly detection is performed. N, R, M, and T in each sample of this application all represent the total quantity in the corresponding sample.

[0058] Adopting the above-mentioned method for detecting the attitude determination state of a spacecraft based on constructing an attitude regression model, by preprocessing the parameters of the spacecraft and specifically constructing a suitable attitude regression model, the detection of the attitude determination state can be applicable to spacecrafts using different attitude control actuators and attitude control measurement units; the detection of the attitude determination state improves the versatility on the basis of high accuracy.

[0059] Next, reference will be made to Figures 1 to 2 to describe each step of the above-mentioned method for detecting the attitude determination state of a spacecraft based on constructing an attitude regression model in this exemplary embodiment in more detail.

[0060] In some embodiments, as shown in reference Figure 2 , step S100 includes:

[0061] The parameters of the attitude control actuator are at least one of the following: reaction wheel, magnetic torquer, thruster, balance wheel, orbit control subsystem, on-board time or orbit parameters;

[0062] The parameters of the attitude control measurement unit are at least one of the following: parameters of sun sensor, star sensor, gyroscope or magnetometer.

[0063] It should be understood that when the attitude regression model is trained and applied, the inputs to the attitude control execution mechanisms, including reaction wheels, magnetic torque actuators, thrusters, balance wheels, orbit control subsystems, on-board time, orbit, etc., are used as the inputs to the attitude regression model; the outputs of the attitude regression model are the relevant measurement values and their increment values of the attitude control measurement unit, including sun sensors, star sensors, gyroscopes, magnetometers, etc. The input and output data used to train the attitude regression model are selected from normal state data for training. The types of the above inputs and outputs can be adaptively selected according to the actual situation.

[0064] In some embodiments, referring to Figure 2 shown in, step S200 includes:

[0065] Taking the input historical parameters and the output historical parameters in one-to-one correspondence as sample data, performing data wild value rejection according to the sample standard deviation and the sample mean, and normalizing the sample data after wild value rejection to generate set data.

[0066] It should be understood that the above input parameters and output parameters are selected under normal spacecraft attitude maneuvers; the input parameter , where represents the telemetry value of the i-th parameter in the input quantity at time t, and the output sample , where represents the telemetry value of the j-th parameter in the output quantity at time t.

[0067] For data wild value rejection, use for wild value rejection, where is the sample standard deviation, is a positive integer, and generally use for wild value rejection, that is, reject the data falling outside the region, is the sample mean, and fill it with the mean of the two adjacent points in the adjacent region.

[0068] For the normalization of input and output data, normalize the data to between [0,1]. The normalization method for input data is , is the column feature vector of all input samples, is the parameter 's minimum value, is the parameter 's maximum value; the processing process for output data is similar. After the calculation, restore it to the original range. The inverse normalization method for input data is ; the processing process for output data is similar. is the row feature vector of all input samples, max( ) is 's maximum value, min( ) is The minimum value.

[0069] In some embodiments, referring to Figure 2 as shown in, step S300 includes:

[0070] Define a loss function through the mean square error, update the training data according to the gradient of the loss function, and substitute the updated training data into the decision tree model for training, where the mean of the samples is used as the initial value.

[0071] It should be understood that for the training set samples , the process of constructing the pose regression model is as follows:

[0072] (1) Initialize the weak learner , is the mean of the prediction samples .

[0073] (2) For sample i, define the loss function of the pose regression model at sample i as , is the predicted value of the pose regression model at sample i, is the true value of the training set sample.

[0074] (3) According to the gradient of the loss function, obtain the new training data , and use the new training data to obtain the decision tree model .

[0075] (4) Update the decision tree model to obtain , where the parameter represents the update strength of the decision tree model. The calculation formula is:

[0076] . Among them, is the initial value of the update strength of the decision tree model. Specifically, it can take the value = 0.01.

[0077] (5) Repeat the above steps (3) and (4) to obtain the decision tree model update formula

[0078]

[0079] where the decision tree the training data of is ; among them, .

[0080] Select the depth m of the gradient boosting decision tree, that is, use m decision trees in the decision tree model. The decision tree model can be expressed as:

[0081]

[0082] Define the loss function as the mean squared error: 。

[0083] Obtain the initial attitude regression model and the parameters of the attitude regression model.

[0084] In some embodiments, as shown in Figure 2 Step S300 includes:

[0085] Use the test set to evaluate the accuracy of the trained decision tree model, and calculate the determination coefficient of the decision tree model; if the determination coefficient is less than the preset threshold, retrain; if the determination coefficient is greater than or equal to the preset threshold, then use the trained decision tree model as the attitude regression model corresponding to the spacecraft.

[0086] It should be understood that the model obtained after training the decision tree model is a to-be-determined attitude regression model, and test data needs to be introduced for testing. Calculate the error between the output value of the test data and the predicted value of the to-be-determined attitude regression model. If the mean value of the loss function L of all test data is greater than or equal to the threshold (for example, the threshold is equal to 0.001), then output it as the final attitude regression model, and count the loss function L value of the attitude regression model for each test sample; otherwise, return for retraining. If the determination coefficient is greater than or equal to the preset threshold, then output the attitude regression model and the parameters of the attitude regression model. Store data such as the parameters, weights, and thresholds of the attitude regression model for detection applications.

[0087] In some embodiments, as shown in Figure 2 Step S300 includes:

[0088] Split the set data into the training set and the test set according to a ratio of 7:3.

[0089] It should be understood that the above processed data is split into a training set and a test set according to a ratio of 7:3 in terms of the sample size, where the training data is used to train the attitude regression model, and the test data is used to verify the accuracy of the attitude regression model.

[0090] In some embodiments, as shown in Figure 2 Step S300 includes:

[0091] Split the set data into the training set, the test set, and the validation set according to a ratio of 6:2:2.

[0092] During the training based on the decision tree model, verify the loss function threshold according to the validation set every time a training is completed, and complete the training when the loss function is less than the validation threshold.

[0093] It should be understood that the processed set data is split into training data , validation data and test data . T, D, and R are the numbers of the three types of samples respectively, and their ratio is 6:2:2, and M = T + D + R. The training data is used to train the pose regression model, the validation data is used to prevent overfitting, and the test data is used to verify the accuracy of the pose regression model.

[0094] After each training with the training data, it is verified with the validation data. If the value of the loss function L is less than the threshold (for example, 0.001), in order to prevent overfitting, the algorithm iteration is aborted, and the trained network is the pending pose regression model; at the same time, the maximum number of iterations is set to 100 times. When the maximum number is reached, the iteration is also aborted, and the pending pose regression model is output.

[0095] In some embodiments, as shown in reference Figure 2 , step S500 includes:

[0096] Set the detection threshold and the detection step size.

[0097] Input the input data of the data to be detected into the pose regression model to generate prediction data.

[0098] Compare the output data of the data to be detected with the prediction data.

[0099] When the comparison result exceeds the detection threshold and meets the detection step size, it is judged as a pose determination anomaly.

[0100] It should be understood that for the detection rule setting, according to the set of loss function L values output , set the detection threshold = 0.001 and the detection step size , that is, if consecutive data points to be detected exceed the threshold , it is judged as an anomaly; in order to improve the detection sensitivity, take the upper quartile of the set .

[0101] In some embodiments, as shown in reference Figure 2 , step S500 includes:

[0102] If consecutive data points to be detected satisfy the following formula, it is judged as a pose determination anomaly:

[0103]

[0104] Wherein, generates prediction data for the pose regression model The k-th component of; is the output data of the data to be detected The k-th component of, is the feature dimension of the output data; is the detection threshold; is the detection step size.

[0105] It should be understood that the predicted data generated by the attitude regression model is the data after inverse normalization of the output data, that is, the data restored to the original scale range of the parameters after the attitude regression model is generated. Substitute the input data of the data to be detected into the above formula one by one for calculation. If successive data points to be detected satisfy the above formula, it is determined that the attitude determination is abnormal.

[0106] In some embodiments, referring to Figure 2 shown in, step S500 includes:

[0107] Evaluate the detection accuracy of the attitude determination model through the data to be detected and the predicted data; the attitude determination model is a model for controlling the attitude state of the spacecraft.

[0108] It should be understood that the detection accuracy evaluation. For the sequence to be detected , is the total number of sequences, and the output sequence of the attitude regression model is , and the following calculation formula is used to evaluate the detection accuracy of the attitude determination model:

[0109]

[0110] where, is the mean value of the j-th component. The value range of is 0 to 1, and the better the attitude model effect of the attitude determination model, the larger the value. It should be understood that: the output sequence of the attitude regression model is also the data after inverse normalization processing of the data.

[0111] Combined with Figure 3 and the above embodiments, it will be described below through specific examples.

[0112] Step 1: Determine the input and output parameters of the attitude regression model. When the attitude regression model is trained and applied, it mainly takes the parameters applied to the attitude control actuator, including reaction wheels, magnetic torquers, thrusters, balance wheels, orbit control subsystems, on-board time, orbit, etc. as the input of the attitude regression model; the output of the attitude regression model is the relevant measurement values and their increment values of the attitude control measurement unit, including sun sensors, star sensors, gyroscopes, magnetometers, etc.;

[0113] Step 2: Divide the given set of data into input samples and output samples , where M and N are the corresponding sample numbers.

[0114] Step 3: Preprocess the input and output parameters, including data outlier removal, normalization, data alignment, and padding.

[0115] Step 4: Data outlier removal, using for outlier removal, where is the sample standard deviation, is a positive integer, and is used for outlier removal, that is, data falling outside the region is removed, is the sample mean, and the data is filled with the mean of the two adjacent points in the adjacent region.

[0116] Step 5: Normalization processing, normalize the data to between [0, 1]. The normalization method for the input data is , is the column feature vector of all input samples in Step 1; after calculation, it is restored to the original range. The inverse normalization method for the input data is ; the processing process of the output data is similar.

[0117] Step 6: Data alignment and padding, align the data by time point in units of days. In addition, delete the time points with large sparsity in the data, and fill the time points with small sparsity by moving average.

[0118] Step 7: Split the processed data into training data , validation data , and test data , where T, D, and R are the numbers of the three types of samples, and their ratio is 6:2:2, and M = T + D + R. The training data is used to train the pose regression model, the validation data is used to prevent overfitting, and the test data is used to verify the accuracy of the pose regression model.

[0119] Step 8: Select the random forest algorithm to construct the initial pose regression model.

[0120] Step 9: Use the training data to train the initial pose regression model to obtain the trained initial pose regression model, and verify the trained pose regression model through the validation data to obtain the pose regression model verification index (coefficient of determination).

[0121] Step 10: If the coefficient of determination is greater than the threshold, output the pose regression model and the loss function L.

[0122] Step 11: Prepare the data to be detected, and prepare the input and output of the data to be detected according to Steps 1 to 6.

[0123] Step 12: Import the pose regression model, and import the pose regression model output in Step 10;

[0124] Step 13: Set the detection rules. According to the set of loss function L values output in Step 10 , set the detection threshold =0.001 and the detection step , that is, if consecutive points to be detected exceed the threshold , it is judged as abnormal; to improve the detection sensitivity, take the upper quartile of the set ;

[0125] Step 14: Input the data to be detected one by one into the pose regression model for calculation. If consecutive points to be detected satisfy the following formula, it is judged as abnormal in pose determination:

[0126]

[0127] In the above formula, , are the components of the predicted value vector of the pose regression model and the true value vector of the data to be detected respectively, and is the feature dimension of the output data.

[0128] Step 15: Evaluate the detection accuracy. For the sequence to be detected , is the total number of sequences, and the output sequence of the pose regression model is . The following calculation formula is used to evaluate the detection accuracy of the pose determination model:

[0129]

[0130] is the mean value of the jth component. ranges from 0 to 1, and the better the model effect, the larger the value.

[0131] Refer to Figure 3 as shown. Figure 3 In

[0132] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine the different embodiments or examples described in this specification.

[0133] After considering the specification and practicing the invention disclosed herein, those skilled in the art will readily conceive of other embodiments of the present invention. The present invention is intended to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include the common general knowledge or conventional technical means in the technical field not disclosed in the present invention. The specification and examples are only regarded as exemplary, and the true scope and spirit of the present invention are pointed out by the appended claims.

Claims

1. A method for detecting the attitude state of a spacecraft based on constructing an attitude regression model, characterized in that: include: Determine the attitude control actuator and attitude control measurement unit of the spacecraft, and obtain the corresponding parameters to be tested and historical parameters; Taking the historical parameters of the attitude control actuator as input and the historical parameters of the attitude control measurement unit as output, generating aggregate data through normalization processing; Splitting part of the set data as a training set and training according to a decision tree model, then splitting another part of the set data as a test set to verify the training result, and using the trained decision tree model as an attitude regression model corresponding to the spacecraft; Taking the parameters to be detected of the attitude control actuator as input and the parameters to be detected of the attitude control measurement unit as output, and generating the data to be detected through normalization processing; Substituting the data to be detected into the attitude regression model and detecting the attitude state of the spacecraft; The loss function is defined by the mean square error, and the training data is updated according to the gradient of the loss function. The updated training data is substituted into the decision tree model for training. ,The posture regression model construction process is as follows: (1) Initialize weak learners , For the prediction sample The mean of (2) For sample i, the loss function of the posture regression model on sample i is defined as , is the predicted value of the posture regression model in sample i, is the true value of the training set sample; (3) Obtain new training data based on the gradient of the loss function , and use the new training data to get the decision tree model ; (4) Update the decision tree model to obtain , where the parameters Indicates the strength of decision tree model update; The calculation formula is: ; in, Update the initial value of the strength for the decision tree model; (5) Repeat the above steps (3) and (4) to obtain the decision tree model update formula ; Among them, the decision tree The training data is ;in, ; The depth m of the gradient boosting decision tree is selected, that is, m decision trees are used in the decision tree model. The decision tree model can be expressed as: ; The loss function is defined as mean square error: ; An initial posture regression model and parameters of the posture regression model are obtained.

2. The method for detecting the attitude state of a spacecraft based on constructing an attitude regression model according to claim 1, characterized in that: The step of taking the historical parameters of the attitude control actuator as input and the historical parameters of the attitude control measurement unit as output, and generating the set data through normalization processing comprises: The input historical parameters and the output historical parameters are used as sample data in a one-to-one correspondence. The data is eliminated according to the sample standard deviation and the sample mean. The sample data after elimination is normalized to generate set data.

3. The method for detecting the attitude state of a spacecraft based on constructing an attitude regression model according to claim 2, characterized in that: The step of splitting another part of the set data as a test set to verify the training result comprises: The test set is used to evaluate the model accuracy of the trained decision tree model and calculate the determination coefficient of the decision tree model; if the determination coefficient is less than a preset threshold, retraining is performed; if the determination coefficient is greater than or equal to the preset threshold, the trained decision tree model is used as the attitude regression model corresponding to the spacecraft.

4. The method for detecting the attitude state of a spacecraft based on constructing an attitude regression model according to claim 3, characterized in that: The steps of splitting part of the data of the set data as a training set and training according to a decision tree model, and then using another part of the data of the set data as a test set to verify the training result, and using the trained decision tree model as the attitude regression model corresponding to the spacecraft include: The set data is split into the training set and the test set in a ratio of 7:

3.

5. The method for detecting the attitude state of a spacecraft based on constructing an attitude regression model according to claim 3, characterized in that: The steps of splitting part of the data of the set data as a training set and training according to a decision tree model, and then using another part of the data of the set data as a test set to verify the training result, and using the trained decision tree model as the attitude regression model corresponding to the spacecraft include: Splitting the set data into the training set, the test set and the validation set in a ratio of 6:2:2; In the training based on the decision tree model, each time the training is completed, the loss function threshold is verified according to the verification set, and the training is completed when the loss function is less than the verification threshold.

6. The method for detecting the attitude state of a spacecraft based on constructing an attitude regression model according to claim 1, characterized in that: The step of substituting the data to be detected into the attitude regression model and detecting the attitude state of the spacecraft comprises: Set the detection threshold and detection step size; Inputting the input data of the data to be detected into the posture regression model to generate prediction data; Comparing the output data of the data to be detected with the predicted data; When the comparison result exceeds the detection threshold and meets the detection step length, it is judged as posture abnormality.

7. The method for detecting the attitude state of a spacecraft based on constructing an attitude regression model according to claim 6, characterized in that: The step of substituting the data to be detected into the attitude regression model and detecting the attitude state of the spacecraft comprises: If continuous If the points to be detected satisfy the following formula, it is judged as abnormal posture: in, Generate prediction data for the posture regression model The kth component of ; The output data of the data to be detected The kth component of is the characteristic dimension of the output data; is the detection threshold; is the detection step length.

8. The method for detecting the attitude state of a spacecraft based on constructing an attitude regression model according to claim 7, characterized in that: The step of substituting the data to be detected into the attitude regression model and detecting the attitude state of the spacecraft comprises: The detection accuracy of the attitude model is evaluated by the data to be detected and the predicted data; the attitude model is a model for controlling the attitude state of the spacecraft.

9. The method for detecting the attitude state of a spacecraft based on constructing an attitude regression model according to any one of claims 1 to 8, characterized in that: The steps of determining the attitude control actuator and attitude control measurement unit of the spacecraft and obtaining the corresponding parameters to be detected and historical parameters include: The parameters of the attitude control actuator are at least one of the following: parameters of a reaction flywheel, a magnetic torquer, a thruster, a balance wheel, a track control subsystem, on-board time or an orbit; The parameters of the attitude control measurement unit are at least one of the following: parameters of a thermosensitive meter, a star sensitive meter, a gyroscope or a magnetometer.

Images