CN116834977A - A range control method for satellite orbit data - Google Patents

Abstract

The invention relates to the field related to satellite orbit data, specifically a range control method for satellite orbit data. The advanced neural network algorithm of the invention adopts the GPT-3.5 Turbo model, which has extremely high prediction accuracy and generalization ability. The prediction and analysis of orbit data has certain advantages compared with existing technologies; by using advanced neural network algorithms, satellite measurement data can be quickly acquired and accurately processed to achieve real-time control effects. At the same time, this method can reduce the cost of measurement equipment and control equipment. cost and improve control efficiency.

Description

A range control method for satellite orbit data

Technical field

The invention relates to the field related to satellite orbit data, and is specifically a range control method of satellite orbit data.

Background technique

In the application of satellite technology, precise control of satellite orbits is crucial. In order to achieve this goal, satellites need to be equipped with a variety of measurement equipment, including micro thrusters, inter-satellite positioning equipment, etc., to obtain real-time orbit data.

However, due to the different motion states of satellites in orbit and large orbit data errors, traditional static control methods can no longer meet the needs. Therefore, it is necessary to seek a more precise orbit data range control method to ensure the stable operation of satellites.

Contents of the invention

The purpose of the present invention is to provide a range control method for satellite orbit data to solve the problems raised in the above background technology.

In order to achieve the above objects, the present invention provides the following technical solution: a range control method of satellite orbit data, including the following steps:

Step S1: Obtain satellite measurement data;

Step S2: Use advanced neural network algorithms to analyze and process the acquired measurement data, predict the motion patterns of the orbit, and obtain prediction results;

Step S3: Compare the observed values in the satellite measurement data with the prediction results in step S2, correct the data errors, and obtain more accurate orbit data;

Step S4: Transmit the corrected orbit data to the satellite to achieve real-time control.

Preferably, the advanced neural network algorithm used in step S2 is the GPT-3.5 Turbo model, and its specific model processing steps are:

Step a. First collect satellite measurement data, including satellite position, orbit parameters, and motion speed parameters, and preprocess these data to eliminate measurement errors and noise, and convert the data into a digital sequence format;

Step b. After data preprocessing, use the GPT-3.5 Turbo model for training to predict and analyze satellite orbit data. During this process, it is necessary to determine the input and output formats of the model, as well as select parameters and algorithms;

Step c. After the model training is completed, the model needs to be tested and adjusted using cross-validation technology to determine the accuracy and stability of the model;

Step d. After the model is tested and adjusted, the model is applied to the range control of the satellite orbit data. First, the satellite measurement data in step a is input into the model, and after processing and prediction, the range information of the orbit data is obtained. Then, the processed track data is transmitted to the control device for real-time control.

Preferably, the preprocessing of data in step a specifically includes: when obtaining satellite position data, using data validity check to conduct strict data verification to determine whether the data is accurate, complete and consistent; after the data verification is completed Finally, a digital filter is used for filtering processing to reduce the impact of measurement errors and noise on data prediction; when the satellite position data is missing or has outliers, linear interpolation is used for data interpolation to fill in the missing values and eliminate outliers; after the data has completed the above After processing, standard deviation normalization is used to normalize the data to eliminate differences between different data scales.

Preferably, the specific steps of using the GPT-3.5 Turbo model for training in step b are:

Step 1. When predicting satellite position, orbit parameters and speed, satellite measurement data and historical orbit data are used as the input of the model, and the orbit prediction error and orbit data at the next moment are used as the output of the model;

Step 2: After determining the model input and output, input the near-GPT-3.5Turbo model for training. The specific GPT-3.5Turbo model is:

text{MultiHead}(Q,K,V)=text{Concat}(head 1,...,head_h)W^0text{Concat}head_i=text{Attention}(QW_i^Q,KW_i^K,VW_i^V)text {Attention}W_i^Q, W_i^K, W_i^V.

Among them, Q, K, V represent the input query vector, key vector and value vector respectively; head_i represents the output of the i-th head; text is the trainable weight matrix; text(MultiHead} represents the multi-head self-attention mechanism, W_i^Q , W_i^K, W_i^V are trainable parameters in the multi-head attention mechanism; text{Concat} connects the outputs of the multi-heads, and W^0 is the final output layer weight matrix;

Step 3: After model training, evaluate the model using cross-validation to check the accuracy and stability of the model.

Preferably, the data errors are corrected in step S3 to obtain more accurate orbit data. According to the results of data analysis, the input data adjustment method is used to correct the errors of the prediction results. The specific steps are:

Step S31. Use statistical analysis and visualization tools to identify abnormal points and noise points in the satellite measurement data through basic statistical indicators such as the mean, standard deviation or quantile of statistical data to identify and analyze the source of noise or abnormal values;

Step S32: According to the results of data analysis, use the Kalman filter algorithm to filter and replace outliers in the satellite measurement data;

Step S33. After filtering the interference data, reuse the corrected data to perform model training in step S2, and predict the satellite position and speed based on the trained model;

Step S34: After completing the data prediction, compare the prediction results with the satellite measurement data in step S1 to check the accuracy and stability of the data.

Preferably, the specific steps of using the Kalman filter algorithm to filter and replace outliers in the satellite measurement data in step S32 include:

Step S321: First define the satellite's position, velocity and acceleration state quantities as a system state model, and use the system state model to express it as the following linear state equation: x(k+1)=A_kx(k)+B_ku(k)+V (k), where x(k) represents the system state, A_k represents the state transition matrix, B_k represents the control matrix, u(k) represents the control input, and V(k) represents the system noise;

Step S322. The satellite measurement data is used as the observation quantity. The position and velocity vector of the satellite are defined as the observation model. The observation model can be expressed as the following linear observation equation: z(k)=H_kx(k)+w(k), where, z (k) represents the observation value, H_k represents the observation matrix, and w(k) represents the measurement noise;

Step S323: Use variance to describe the system noise and measurement noise, and input the system noise and measurement noise into the system state model, observation model and initial state estimation;

Step S324: Estimate the initialization state of the satellite, where the following initial state estimation is used for satellite position and velocity prediction: x(0)=[pos0, vel0, acc0], where pos0, vel0, acc0 respectively represent the initial position of the satellite. , initial velocity and initial acceleration;

Step S325: Predict the status of the satellite based on the system state model, system noise, measurement noise, and satellite motion rules. At the same time, correct and update the status of the satellite based on the observation model and satellite measurement data, and calculate and update the Kalman gain. To further improve the accuracy and precision of the estimated state, the predicted satellite state and satellite state are corrected and updated using the following Kalman filter algorithm formula:

predict:

Covariance prediction: Pk|k-1=Akpk-1AkT+Qk

Update: Kk＝Pk|k-1HkT(HkPk|k-1HkT+Rk)-1

Correction:

Covariance update: Pk|k=(1-KkHk)Pk|k-1

In the above formula is the optimal state estimate, Pk|k-1 is the covariance matrix of the state estimation error, Kk is the Kalman gain, zk is the observation value, Hk is the observation matrix, Qk is the covariance matrix of the system noise, and Rk is the measurement noise covariance matrix.

Compared with the existing technology, the beneficial effects of the present invention are: the advanced neural network algorithm of the present invention adopts the GPT-3.5 Turbo model, which has extremely high prediction accuracy and generalization ability, and is very effective in the prediction and analysis of orbit data. It has certain advantages over existing technology; by using advanced neural network algorithms, satellite measurement data can be quickly acquired and accurately processed to achieve real-time control effects. At the same time, this method can reduce the cost of measurement equipment and control equipment and improve control efficiency.

Description of the drawings

Figure 1 is a schematic structural diagram of the method flow of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

Please refer to Figure 1. The present invention provides a technical solution: a range control method of satellite orbit data, including the following steps:

Step S1: Obtain satellite measurement data;

Step S2: Use advanced neural network algorithms to analyze and process the acquired measurement data, predict the motion patterns of the orbit, and obtain prediction results;

Step S3: Compare the observed values in the satellite measurement data with the prediction results in step S2, correct the data errors, and obtain more accurate orbit data;

Step S4: Transmit the corrected orbit data to the satellite to achieve real-time control.

Furthermore, the advanced neural network algorithm used in step S2 is the GPT-3.5 Turbo model, and its specific model processing steps are:

Step a. First collect satellite measurement data, including satellite position, orbit parameters, and motion speed parameters, and preprocess these data to eliminate measurement errors and noise, and convert the data into a digital sequence format;

Step b. After data preprocessing, use the GPT-3.5 Turbo model for training to predict and analyze satellite orbit data. During this process, it is necessary to determine the input and output formats of the model, as well as select parameters and algorithms;

Step c. After the model training is completed, the model needs to be tested and adjusted using cross-validation technology to determine the accuracy and stability of the model;

Step d. After the model is tested and adjusted, the model is applied to the range control of the satellite orbit data. First, the satellite measurement data in step a is input into the model, and after processing and prediction, the range information of the orbit data is obtained. Then, the processed track data is transmitted to the control device for real-time control.

Further, the preprocessing of data in step a specifically includes: when obtaining satellite position data, use data validity check to conduct strict data verification to determine whether the data is accurate, complete and consistent; after the data verification is completed, use The digital filter performs filtering processing to reduce the impact of measurement errors and noise on data prediction; when satellite position data is missing or has outliers, linear interpolation is used for data interpolation to fill in missing values and eliminate outliers; after the data has completed the above processing , use standard deviation normalization to normalize the data and eliminate the differences between different data scales.

Further, the specific steps of using the GPT-3.5 Turbo model for training in step b are:

Step 1. When predicting satellite position, orbit parameters and speed, satellite measurement data and historical orbit data are used as the input of the model, and the orbit prediction error and orbit data at the next moment are used as the output of the model;

Step 2: After determining the model input and output, input the near-GPT-3.5Turbo model for training. The specific GPT-3.5Turbo model is:

text{MultiHead}(Q,K,V)=text{Concat}(head_1,...,head_h)W^0text{Concat}head_i=text{Attention}(QW_i^Q,KW_i^K,VW_i^V)text{ Attention}W_i^Q, W_i^K, W_i^V.

Among them, Q, K, V represent the input query vector, key vector and value vector respectively; head_i represents the output of the i-th head; text is the trainable weight matrix; text{MultiHead} represents the multi-head self-attention mechanism, W i^ Q, W i^K, W i^V are trainable parameters in the multi-head attention mechanism; text{Concat} connects the outputs of the multi-heads, and W^0 is the final output layer weight matrix;

Step 3: After model training, evaluate the model using cross-validation to check the accuracy and stability of the model.

Further, in step S3, the data error is corrected to obtain more accurate orbit data. According to the results of data analysis, the input data adjustment method is used to correct the error of the prediction result. The specific steps are:

Step S31. Use statistical analysis and visualization tools to identify abnormal points and noise points in the satellite measurement data through basic statistical indicators such as the mean, standard deviation or quantile of statistical data to identify and analyze the source of noise or abnormal values;

Step S32: According to the results of data analysis, use the Kalman filter algorithm to filter and replace outliers in the satellite measurement data;

Step S33. After filtering the interference data, reuse the corrected data to perform model training in step S2, and predict the satellite position and speed based on the trained model;

Step S34: After completing the data prediction, compare the prediction results with the satellite measurement data in step S1 to check the accuracy and stability of the data.

Further, in step S32, the Kalman filter algorithm is used to filter and replace outliers in the satellite measurement data. The specific steps include:

Step S321: First define the satellite's position, velocity and acceleration state quantities as a system state model, and use the system state model to express it as the following linear state equation: x(k+1)=A_kx(k)+B_ku(k)+V (k), where x(k) represents the system state, A_k represents the state transition matrix, B_k represents the control matrix, u(k) represents the control input, and V(k) represents the system noise;

Step S322. The satellite measurement data is used as the observation quantity. The position and velocity vector of the satellite are defined as the observation model. The observation model can be expressed as the following linear observation equation: z(k)=H_kx(k)+w(k), where, z (k) represents the observation value, H_k represents the observation matrix, and w(k) represents the measurement noise;

Step S323: Use variance to describe the system noise and measurement noise, and input the system noise and measurement noise into the system state model, observation model and initial state estimation;

Step S324: Estimate the initialization state of the satellite, where the following initial state estimation is used for satellite position and velocity prediction: x(0)=[pos0, vel0, acc0], where pos0, vel0, acc0 respectively represent the initial position of the satellite. , initial velocity and initial acceleration;

Step S325: Predict the status of the satellite based on the system state model, system noise, measurement noise, and satellite motion rules. At the same time, correct and update the status of the satellite based on the observation model and satellite measurement data, and calculate and update the Kalman gain. To further improve the accuracy and precision of the estimated state, the predicted satellite state and satellite state are corrected and updated using the following Kalman filter algorithm formula:

predict:

Covariance prediction: Pk|k-1=Akpk-1AkT+Qk

Update: Kk＝Pk|k-1HkT(HkPk|k-1HkT+Rk)-1

Correction:

Covariance update: Pk|k=(1-KkHk)Pk|k-1

In the above formula is the optimal state estimate, Pk|k-1 is the covariance matrix of the state estimation error, Kk is the Kalman gain, zk is the observation value, Hk is the observation matrix, Qk is the covariance matrix of the system noise, and Rk is the measurement noise covariance matrix.

The advanced neural network algorithm of the present invention adopts the GPT-3.5 Turbo model, which has extremely high prediction accuracy and generalization ability. It has certain advantages over existing technologies in the prediction and analysis of orbit data; by utilizing the advanced neural network The algorithm can quickly acquire and accurately process satellite measurement data to achieve real-time control. At the same time, this method can reduce the cost of measurement equipment and control equipment and improve control efficiency.

Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art will understand that various changes, modifications, and substitutions can be made to these embodiments without departing from the principles and spirit of the invention. and modifications, the scope of the invention is defined by the appended claims and their equivalents.

Claims (6)

1. A method for controlling the range of satellite orbit data, comprising the steps of:

s1, acquiring satellite measurement data;

s2, analyzing and processing the acquired measurement data by utilizing an algorithm of the advanced neural network, and predicting a motion rule of the orbit to obtain a prediction result;

s3, comparing the observed value in the satellite measurement data with the predicted result in the step S2, and correcting the data error to obtain more accurate orbit data;

and S4, transmitting the corrected orbit data to a satellite to realize instant control.

2. The method for controlling the range of satellite orbit data according to claim 1, wherein: the algorithm of the advanced neural network used in the step S2 is a GPT-3.5Turbo model, and the specific model processing steps are as follows:

step a, firstly, collecting satellite measurement data including satellite position, orbit parameters and motion speed parameters, preprocessing the data, eliminating measurement errors and noise, and converting the data into a digital sequence format;

step b, training by using a GPT-3.5Turbo model after data preprocessing to predict and analyze satellite orbit data, wherein in the process, input and output formats of the model need to be determined, and parameters and algorithms need to be selected;

step c, after model training is completed, testing and adjusting the model by adopting a cross-validation technology to determine the accuracy and stability of the model;

and d, after model testing and adjustment, applying the model to range control of satellite orbit data, firstly, inputting the satellite measurement data in the step a into the model, processing and predicting to obtain range information of the orbit data, and then, transmitting the processed orbit data to control equipment for real-time control.

3. A method of controlling the range of satellite orbit data according to claim 2, wherein: the preprocessing of the data in the step a specifically comprises the following steps: when satellite position data is acquired, data validity check is used for carrying out strict data check so as to judge whether the data are accurate, complete and consistent; after the data verification is finished, a digital filter is adopted for filtering treatment, so that the influence of measurement errors and noise on data prediction is reduced; when the satellite position data has a missing or abnormal value, performing data interpolation by adopting linear interpolation, filling the missing value and eliminating the abnormal value; after the data is processed, the data is normalized by using standard deviation normalization, and the difference between different data scales is eliminated.

4. A method of controlling the range of satellite orbit data according to claim 2, wherein: the training specific steps of the step b by using the GPT-3.5Turbo model are as follows:

step one, when predicting satellite positions, orbit parameters and speeds, satellite measurement data and historical orbit data are used as input of a model, orbit prediction errors, orbit data at the next moment and the like are used as output of the model;

step two, after the input and the output of the model are determined, a near GPT-3.5Turbo model is input for training, wherein the GPT-3.5Turbo model specifically comprises:

text{MultiHead}(Q,K,V)＝text{Concat}(head_1,…,head_h)W^Otext{Concat}head_i＝text{Attention}(QW_i^Q,KW_i^K,VW_i^V)

text{Attention}W_i^Q,W_i^K,W_i^V。

wherein Q, K, V represents an input query vector, a key vector, and a value vector, respectively; head_i represents the output of the i-th header; text is a trainable weight matrix; text { MultiHead } represents a multi-headed self-attention mechanism, W_i≡Q, W_i≡K, W_i≡V being a trainable parameter in the multi-headed attention mechanism; text { Concat } connects the outputs of multiple heads, W≡O being the final output layer weight matrix;

and thirdly, after model training, evaluating the model by adopting cross verification, and checking the accuracy and stability of the model.

5. The method for controlling the range of satellite orbit data according to claim 1, wherein: in the step S3, the data error is corrected to obtain more accurate track data, wherein, according to the result of data analysis, a method of adjusting input data is adopted to correct the error of the predicted result, and the specific steps are as follows:

step S31, identifying abnormal points and noise points in satellite measurement data through statistical analysis and visualization tools and basic statistical indexes such as mean value, standard deviation or quantile of the statistical data so as to identify and analyze sources of noise or abnormal values;

step S32, filtering and replacing abnormal values in satellite measurement data by using a Kalman filtering algorithm according to the data analysis result;

step S33, after filtering the interference data, re-adopting the corrected data to perform model training in step S2, and predicting the satellite position and speed according to the trained model;

and step S34, after the data prediction is completed, comparing the prediction result with the satellite measurement data in the step S1 to test the accuracy and stability of the data.

6. The method for controlling the range of satellite orbit data according to claim 1, wherein: the step S32 of filtering and replacing the outlier in the satellite measurement data by using the kalman filter algorithm specifically includes:

step S321, first, the satellite position, velocity and acceleration state quantity are defined as a system state model, and the system state model is expressed as the following linear state equation: x (k+1) =a_kx (k) +b_ku (k) +v (k), where x (k) represents a system state, a_k represents a state transition matrix, b_k represents a control matrix, u (k) represents a control input, and v (k) represents system noise;

in step S322, the satellite measurement data is used as an observed quantity, and the position and velocity vectors of the satellite are defined as an observation model, which can be expressed as a linear observation equation as follows: z (k) =h_kx (k) +w (k), where z (k) represents the observed value, h_k represents the observed matrix, and w (k) represents the measurement noise;

step S323, describing system noise and measurement noise by using variances, and inputting the system noise and the measurement noise in a system state model, an observation model and initial state estimation;

step S324, estimating an initialized state of the satellite, wherein the following initial state estimation is used for satellite position and velocity prediction: x (0) = [ pos0, vel0, acc0], wherein pos0, vel0, acc0 represent an initial position, an initial velocity, and an initial acceleration of the satellite, respectively;

step S325, predicting the satellite state according to the system state model, the system noise, the measurement noise and the satellite motion law, and correcting and updating the satellite state according to the observation model and the satellite measurement data, and calculating and updating the Kalman gain, so as to further improve the accuracy and precision of the estimated state, wherein the predicted satellite state and the satellite state are corrected and updated by using the following Kalman filtering algorithm formula:

and (3) predicting:

covariance prediction: pk|k-1=akpk-1akt+qk

Updating: kk=Pk|k-1 HkT (HkPk|k-1 HkT+Rk) -1

And (3) correction:

covariance update: pk|k= (1-KkHk) Pk|k-1

Wherein in the above formulaIs the optimal state estimation value, pk|k-1 is the covariance matrix of the state estimation error, kk is the kalman gain, zk is the observed value, hk is the observed matrix, qk is the covariance matrix of the system noise, and Rk is the covariance matrix of the measurement noise.