Satellite solar cell array data driving modeling method based on working condition identification

Abstract

The invention provides a satellite solar cell array data-driven modeling method based on working condition recognition, belongs to the technical field of aerospace, and solves the problems that a solar cell array is difficult to manage and maintain, and the defects of a mechanism modeling method and a traditional data-driven modeling method exist. The modeling step comprises the following steps: collecting telemetry data of an on-orbit satellite and preprocessing the telemetry data; adding a label to the preprocessed telemetry data, and dividing the telemetry data added with the label into a training set, a verification set and a test set; establishing a data-driven model, training, verifying and testing a CNN (Convolutional Neural Network) in the data-driven model, and outputting a working condition identification result; redividing the telemetry data after feature selection and data cleaning into data sets under five working conditions according to labels, and training, verifying and testing the LSTM networks under different working conditions in the data-driven model; and inputting a working condition identification result output by the CNN network into a corresponding LSTM network to obtain a predicted output parameter of the solar cell.

Description

Technical Field

[0001] The invention relates to a satellite solar array data-driven modeling method based on working condition identification, and belongs to the technical field of aerospace. Background Art

[0002] With the rapid development of aerospace technology, satellites play an important role in communications, weather forecasting, geological monitoring and deep space exploration. Most satellites use a solar array-battery combined power supply to power their payloads. The solar array is the only energy source for satellites in orbit, making it one of the most important submodules of the satellite. Since the solar array sail panels are exposed to a large amount of space radiation and drastic temperature changes during their in-orbit operation, their stability decreases. In order to ensure the stable operation of satellites in orbit and avoid the failure of the entire satellite mission due to the failure of the solar array, it is necessary to establish a model for the solar array to identify its operating conditions, analyze and predict its output, and achieve predictive management and maintenance.

[0003] At present, the methods of solar array modeling include mechanism modeling and data-driven modeling. Among them, the mechanism modeling method mainly includes photovoltaic equations and volt-ampere characteristic curve models. The photovoltaic equation involves many parameters and complex calculations. Parameters such as ultraviolet loss factor, combined loss factor, irradiation loss factor, and temperature coefficient need to be determined by referring to relevant literature. The accuracy depends on the acquisition of parameters and there are large errors. The volt-ampere characteristic curve model is an implicit function that cannot be made explicit. The model analysis is difficult and usually requires the use of computers for iterative derivation, which requires high computing power. Traditional data-driven modeling methods mainly include decision trees, vector machines, or BP neural networks. Although there is no need to consider complex model mechanisms, if a single model is established for the dynamic and changeable solar array working conditions, the model accuracy will decrease. Especially for the solar array shadow state and shadow state, since the above states experience a short time and the amount of data is much smaller than other states, it leads to serious data class imbalance problems, making it difficult to achieve effective management and maintenance of solar arrays. Summary of the invention

[0004] The present invention aims to solve the difficulties in the management and maintenance of solar arrays. The solar array mechanism modeling method has many parameters and is difficult to obtain accurately. The volt-ampere characteristic equation involves the difficulty of analyzing implicit functions. In addition, the traditional data-driven modeling method has the problems of poor single model accuracy and unbalanced data classes. A satellite solar array data-driven modeling method based on working condition identification is proposed. The steps include:

[0005] Step 1: Collect telemetry data from satellites in orbit;

[0006] Step 2: Perform feature selection, data cleaning and data sampling on the collected telemetry data to complete data preprocessing;

[0007] Step 3: Label the preprocessed telemetry data and divide the labeled telemetry data into training set, validation set, and test set;

[0008] Step 4: Establish a data-driven model, use the training set, validation set, and test set to train, validate, and test the CNN network in the data-driven model, and output the working condition recognition results;

[0009] Step 5: Re-divide the telemetry data after feature selection and data cleaning into data sets under five working conditions according to the labels, divide the data set of each working condition into training set, validation set and test set in proportion, and train, validate and test the LSTM network under different working conditions in the data-driven model respectively;

[0010] Step 6: Input the working condition recognition results output by the CNN network into the corresponding LSTM network to obtain the predicted output parameters of the solar array.

[0011] Preferably, the telemetry data of the on-orbit satellite collected in step 1 specifically includes illumination status data, ground shadow status data, shadow entry status data, shadow exit status data and position holding status data.

[0012] Preferably, the steps of performing feature selection, data cleaning and data sampling on the collected telemetry data in step 2 include:

[0013] Step 2.1: Extract features related to the operation status of the solar array from the collected telemetry data of the on-orbit satellite as training features of the model, wherein the features related to the operation status of the solar array include but are not limited to the temperature of the solar array, the current of the simulated solar sensor, the voltage of the solar subarray, and the state of illumination / ground shadow;

[0014] Step 2.2: Remove missing values ​​from the telemetry data after feature selection, introduce special labels for the ground shadow state data, and use the Z-score method to process them separately. Use the Z-score method to process outliers for data without special labels, and replace the outliers with upper and lower thresholds of 3 times the standard deviation to complete data cleaning. Among them, the data without special labels include illumination state data, shadow state data, shadow state data, and position holding state data;

[0015] Step 2.3: Taking the shadow state data as the benchmark, the amount of shadow state data is n, the amount of equally spaced shadow state and shadow exit state data is 2n, the amount of illumination state data before sampling the shadow state is n, and the amount of illumination state data after sampling the shadow exit state is n. The position holding state data are sampled one by one to complete data preprocessing. The preprocessed telemetry data is saved according to the time dimension and exported as a time series data set.

[0016] Preferably, adding labels to the preprocessed telemetry data in step 3 specifically includes: obtaining labels according to the "lighting / ground shadow state" feature, when the lighting / ground shadow state is 0, the label is the lighting state; when the lighting / ground shadow state is 1, the label is the ground shadow state; the data label before the ground shadow state and after the lighting state is the shadow entry state; the data label after the ground shadow state and before the lighting state is the shadow exit state; and the label of the position holding mode is obtained according to the characteristics of the attitude and orbit control working mode.

[0017] Preferably, the CNN network in step 4 is composed of 4-6 convolutional layer structures and 1-2 fully connected layer structures, each convolutional layer is composed of a one-dimensional convolutional layer, a batch normalization layer, a nonlinear activation function layer and a maximum pooling layer, each convolutional layer structure is connected in sequence, and the fully connected layer is connected after the last convolutional layer structure to convert the data into the final number of classification categories.

[0018] Preferably, in step 4, training, verifying and testing the CNN network in the data-driven model using the training set, the validation set and the test set specifically includes:

[0019] Use the training set to train the CNN network, use the validation set to verify the structural parameters, weights and biases of the CNN network, optimize the network parameters and save them, and use the test set to test the accuracy of the solar array working condition classification and recognition. If the accuracy is lower than the preset value, retrain the model until the accuracy is greater than the preset value.

[0020] The expression of the accuracy of solar array working condition classification and identification is:

[0021]

[0022] In formula (1), TP i is the number of samples correctly predicted as category i, C is the number of working conditions, and N is the total number of samples.

[0023] Preferably, in step 5, the step of re-dividing the telemetry data after feature selection and data cleaning into data sets of five working conditions according to labels includes:

[0024] Step 5.1: Divide the time series data set into data sets under five working conditions according to the labels of the telemetry data, "illumination state, ground shadow state, shadow entry state, shadow exit state, and position keeping state", wherein the data sets under the five working conditions include an illumination state data set, a ground shadow state data set, a shadow entry state data set, a shadow exit state data set, and a position keeping state data set;

[0025] Step 5.2: Determine the time step tw of the data set corresponding to each working condition, and use the sliding window method to generate the input sequence and the corresponding target value, where the input sequence is the first tw time step data, and the target value is the tw+1th data. Starting from the starting position of the time series, slide forward one time step each time until the end of the sequence to complete the creation of fixed-length input and output sequences.

[0026] Preferably, the feature dimensions of the LSTM network in step 5 include but are not limited to solar array temperature, analog solar sensor current, solar sub-array voltage, and light / shadow status. The target feature of the LSTM network is the solar array current. Each LSTM network is composed of 2-6 layers of LSTM sub-network layer structures and one fully connected layer, wherein the LSTM sub-network layer structure includes a number of hidden layer units and nonlinear activation functions, the LSTM sub-network layer structures are connected in sequence, the fully connected layer structure includes forward propagation and reverse propagation processes, the fully connected layer is connected after the last layer of LSTM sub-network layer structure, and the output size of the fully connected layer is 1, which is used to fit the output parameters of the solar array.

[0027] Preferably, the LSTM networks under the five working conditions in step 5 include: an LSTM network under an illumination state condition, an LSTM network under an earth shadow state condition, an LSTM network under an entry shadow state condition, an LSTM network under an exit shadow state condition, and an LSTM network under a position holding state condition. Training, verifying, and testing the LSTM networks under the five working conditions in the data-driven model specifically include:

[0028] Use the illumination state training set to train the LSTM network under illumination state conditions, use the illumination state verification set to verify the structural parameters, weights and biases of the LSTM network under illumination state conditions, optimize the network parameters and save them, and use the illumination state test set to verify the fitting effect of the illumination state selection solar array target features;

[0029] Use the earth shadow state training set to train the LSTM network under the earth shadow state condition, use the earth shadow state verification set to verify the structural parameters, weights and biases of the LSTM network under the earth shadow state condition, optimize the network parameters and save them, and use the earth shadow state test set to verify the fitting effect of the earth shadow state to select the target characteristics of the solar array;

[0030] The shadow state training set is used to train the LSTM network under the shadow state working condition, and the shadow state verification set is used to verify the structural parameters, weights and biases of the LSTM network under the shadow state working condition. The network parameters are optimized and saved, and the shadow state test set is used to test the fitting effect of the target characteristics of the solar array selected under the shadow state.

[0031] The LSTM network under the shadow state working condition is trained using the shadow state training set, and the structural parameters, weights and biases of the LSTM network under the shadow state working condition are verified using the shadow state verification set. The network parameters are optimized and saved, and the shadow state test set is used to test the fitting effect of the target characteristics of the solar array selected under the shadow state.

[0032] The LSTM network under the position-holding state working condition is trained using the position-holding state training set, and the structural parameters, weights and biases of the LSTM network under the position-holding state working condition are verified using the position-holding state verification set. The network parameters are optimized and saved, and the position-holding state test set is used to verify the fitting effect of the target features of the solar array selected in the position-holding state.

[0033] Preferably, obtaining the predicted output parameters of the solar array in step 6 specifically includes:

[0034] According to the working condition recognition result of the CNN network, the corresponding LSTM network is selected for prediction. The predicted future output of the solar array is 5-20 time steps, and the error index of the data-driven model is calculated. If the error index is greater than the preset threshold, the CNN network and LSTM network are retrained until the error index is ≤ the preset threshold, and the predicted output of the solar array is obtained;

[0035] The expression of the error index is:

[0036]

[0037] In formulas (2) and (3), N is the total number of samples, y i is the true value of the i-th sample, is the model output value of the i-th sample, MAE is the mean absolute error, and MSE is the mean square error.

[0038] The beneficial effects of the present invention are:

[0039] (1) The present invention proposes a method for identifying the working condition of a satellite solar array, which can achieve accurate working condition identification through a simple network structure. It can be used for status monitoring and early diagnosis of faults of satellite solar arrays, and plays an important role in adjusting data-driven decisions and optimizing performance.

[0040] (2) The present invention predicts output parameters based on the identification of solar array working conditions, and uses multiple models to fit the data features under different working conditions. Compared with directly regressing and fitting the output parameters, the complexity of the model is effectively reduced and the accuracy of the model is increased. Although the output parameters under different working conditions are obtained by different models, the output results of the model can be recombined into a time series according to the original time sequence of the data to achieve on-orbit synchronization and perception of the satellite solar array.

[0041] (3) The present invention establishes a model based on a data-driven method, which does not involve the working mechanism of the solar array and is different from the traditional data method. It only needs to input the collected original telemetry data into the algorithm, and the functions of data preprocessing, calculation, output, comparison, etc. can be automatically realized by the algorithm, and no relevant professional knowledge is required for ground operators. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic flow chart of a satellite solar array data-driven modeling method based on working condition identification provided by the present invention;

[0043] Figure 2 A principle block diagram of a satellite solar array data-driven modeling method based on working condition identification provided by the present invention;

[0044] Figure 3 A schematic diagram of the structure of the CNN network provided by the present invention. DETAILED DESCRIPTION

[0045] Combination Figure 1-3 To illustrate this embodiment, Figure 1 and Figure 2 As shown, the steps of a satellite solar array data-driven modeling method based on working condition identification described in this embodiment include:

[0046] S1: collect telemetry data for data preprocessing;

[0047] This implementation uses the solar array telemetry data of a certain type of high-orbit satellite from March 19, 2023 to December 20, 2023 as the collected data. The collected telemetry data includes the vernal equinox, summer solstice, autumnal equinox, winter solstice and other dates or time periods. The data preprocessing specifically includes:

[0048] S101: Feature selection is performed on the telemetry data. The training features include the solar array temperature at different sampling points, the state of the illuminated area / shadow area, the voltage of all solar arrays, and the simulated solar sensor current at different sampling points. The target feature is the manually marked label.

[0049] S102: Remove missing values ​​from telemetry data. Due to the long distance of the transmission link and certain electromagnetic interference, a few training features of telemetry data have outliers. The outliers are detected using the Z-score method and replaced with upper and lower thresholds of 3 times the standard deviation. The Z-score (standard score) method is a data standardization method that is mainly used to measure the distance of a data point from the standard deviation of its mean. By converting the data into a standard normal distribution (mean is 0 and standard deviation is 1), the dimensional differences between different data sets are eliminated to facilitate comparison or subsequent analysis.

[0050] S103: Sample the data of each day independently. For example, if the telemetry data of the autumnal equinox contains 61 time steps of the shadow state, then the shadow state and the shadow state are sampled with 122 time steps at equal intervals, and the illumination state is sampled with 61 time steps at equal intervals before the shadow state and after the shadow state. After sampling the telemetry data of all time periods, save them according to the time dimension and export them as a new time series data set;

[0051] S2: Add labels and divide into training set, validation set and test set;

[0052] In the preprocessed data set, the illumination state is labeled 0, the ground shadow state is labeled 1, the entry shadow state is labeled 2, the exit shadow state is labeled 3, and the position retention state is labeled 4. The labeled telemetry data set is divided into training set, validation set and test set in a ratio of 14:3:3.

[0053] S3: Build a data-driven model, train the CNN network, and verify the network's structural parameters, weights, and biases;

[0054] The CNN network in this embodiment is a convolutional neural network, which consists of a 5-layer convolution layer structure and a 2-layer fully connected layer structure. The number of input channels of the network is 1, the number of output channels is 5, the 1st to 3rd convolution layer structure consists of a one-dimensional convolution layer, a batch normalization layer, a nonlinear activation function ReLU layer and a maximum pooling layer, and the 4th to 5th convolution layer structure consists of a one-dimensional convolution layer, a batch normalization layer, and a nonlinear activation function ReLU layer; the number of output channels of the mth convolution layer is consistent with the number of input channels of the m+1th convolution layer; the number of batch normalization channels of the mth layer is consistent with the number of output channels of the mth convolution layer; the convolution kernel size, step size and padding size can be designed according to the data dimension; the network structure of CNN is as follows Figure 3 As shown in the figure, the inner structure of the layer is represented by a cube block, where the width of the cube represents the number of input channels, and the height of the cube represents the sequence length; the input dimensions of the model are batch size, number of input channels, and sequence length). After being processed by the convolutional layer structure and the fully connected layer structure, the output of the model is the batch size and the number of classifications;

[0055] Training the CNN network and verifying the network's structural parameters, weights and biases specifically include: using the training set to train the CNN network, using the verification set to verify the CNN network's structural parameters, weights and biases, optimizing the network parameters and saving them.

[0056] S4: Check and output the identification result of the solar array working condition;

[0057] This implementation uses a test set to test the accuracy of solar array operating condition classification and identification. If the accuracy is lower than a preset value, the model is retrained until the accuracy is greater than the preset value.

[0058] The expression of the accuracy of solar array working condition classification and identification is:

[0059]

[0060] In formula (1), TP i is the number of samples correctly predicted as category i, C is the number of working conditions, and N is the total number of samples.

[0061] S5: Create fixed-length input and output sequences from time series data according to five different working conditions, and divide them into five sets of training sets, validation sets, and test sets respectively;

[0062] This implementation method re-divides the data after feature selection and data cleaning into five new data sets according to the labels of the telemetry data, "illumination state, ground shadow state, shadow entry state, shadow exit state, and position holding state". For each data set, an input and output sequence is created for the LSTM network to facilitate the training of time series data, wherein the LSTM network is a recurrent neural network; the method of creating the input and output sequence specifically includes:

[0063] S501: Extract continuous time steps from the input data to form a sequence of a fixed time window tw as input, that is, the historical data required by the model. In this embodiment, the time window is 100.

[0064] S502: Determine the number of future steps that the model needs to predict. The prediction length in this embodiment is 10 time steps.

[0065] S503: Loop through the input data segmentation sequence, starting from the start position of the time series, and slide the index i forward one time step each time until L-tw, where L is the sequence length of the data set.

[0066] S504: In each iteration, extract an input feature sequence of length tw from the current index i.

[0067] S505: Extract the portion of the target feature from the end of the current window to the predicted length as an output feature sequence.

[0068] S506: The input sequence and output sequence under each working condition are divided into a training set, a validation set and a test set in a ratio of 14:3:3, and min-max normalization is performed. The training features are consistent with the training set of the CNN network, and the target feature is the solar array current.

[0069] S6: Train five LSTM networks respectively to verify the network's structural parameters, weights and biases;

[0070] In this embodiment, the LSTM network consists of a 3-layer LSTM sub-network layer structure and a fully connected layer structure, and uses multivariate feature prediction unit features. Among them, the LSTM sub-network layer structure contains 50 hidden layer units. The training process of the LSTM sub-network layer structure follows the following formula:

[0071]

[0072] h (t) =o (t) ·tanh(c (t) ) (4);

[0073] In formulas (2)-(4), i (t) is the input gate, f (t) For the forget gate, (t) is the output gate, is the new information input, σ is the nonlinear activation function sigmoid, W is the weight matrix, h (t-1) is the output information at time t-1. (t-1) is the state of information at time t-1, c (t) is the information state at time t. In the iterative update process of the model, the input is the input information x at time t (t) , the output is the output information h at time t (t) , and c (t-1) and c (t) As historical status information, it will participate in the iteration of the model and determine the model's allocation of the data at time t-1 and the data at time t;

[0074] The corresponding training sets are used to train the LSTM networks under five working conditions respectively. The validation set data is used to verify the structural parameters, weights and bias of the network. The network parameters are optimized and saved. Finally, the test set is used to verify the fitting effect of the target characteristics of the solar array.

[0075] S7: Check and predict the output parameters of the solar array;

[0076] For the characteristics of different working conditions, the corresponding trained LSTM network is used for prediction. The trained LSTM model is loaded, and the test set sequence of the corresponding working condition is input. The output is generated and inverse normalized. The mean absolute error and mean square error are used to evaluate the performance of the model.

[0077] The expression of the error index is:

[0078]

[0079] In formulas (5) and (6), N is the total number of samples, y i is the true value of the i-th sample, is the model output value of the i-th sample, MAE is the mean absolute error, and MSE is the mean square error.

[0080] The above is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement made to the above embodiments without departing from the content of the technical solution of the present invention, based on the technical essence of the present invention, within the spirit and principles of the present invention, still fall within the protection scope of the technical solution of the present invention.

Claims

1. A satellite solar array data-driven modeling method based on working condition identification, characterized in that: The steps of the satellite solar array data-driven modeling method based on working condition identification include: Step 1: Collect telemetry data from satellites in orbit; Step 2: Perform feature selection, data cleaning and data sampling on the collected telemetry data to complete data preprocessing; Step 3: Label the preprocessed telemetry data and divide the labeled telemetry data into training set, validation set, and test set; Step 4: Establish a data-driven model, use the training set, validation set, and test set to train, validate, and test the CNN network in the data-driven model, and output the working condition recognition results; Step 5: Re-divide the telemetry data after feature selection and data cleaning into data sets under five working conditions according to the labels, divide the data set of each working condition into training set, validation set and test set in proportion, and train, validate and test the LSTM network under different working conditions in the data-driven model respectively; Step 6: Input the working condition recognition results output by the CNN network into the corresponding LSTM network to obtain the predicted output parameters of the solar array.

2. According to the data-driven modeling method of satellite solar array based on working condition identification according to claim 1, it is characterized in that: The telemetry data of the on-orbit satellite collected in step 1 specifically includes illumination status data, ground shadow status data, shadow entry status data, shadow exit status data and position holding status data.

3. According to the data-driven modeling method of satellite solar array based on working condition identification in claim 1, it is characterized in that: The steps of feature selection, data cleaning and data sampling of the collected telemetry data in step 2 include: Step 2.1: Extract features related to the operation status of the solar array from the collected telemetry data of the on-orbit satellite as training features of the model, wherein the features related to the operation status of the solar array include but are not limited to the temperature of the solar array, the current of the simulated solar sensor, the voltage of the solar subarray, and the state of illumination / ground shadow; Step 2.2: Remove missing values ​​from the telemetry data after feature selection, introduce special labels for the ground shadow state data, and use the Z-score method to process them separately. Use the Z-score method to process outliers for data without special labels, and replace the outliers with upper and lower thresholds of 3 times the standard deviation to complete data cleaning. Among them, the data without special labels include illumination state data, shadow state data, shadow state data, and position holding state data; Step 2.3: Taking the shadow state data as the benchmark, the amount of shadow state data is n, the amount of equally spaced shadow state and shadow exit state data is 2n, the amount of illumination state data before sampling the shadow state is n, and the amount of illumination state data after sampling the shadow exit state is n. The position holding state data are sampled one by one to complete data preprocessing. The preprocessed telemetry data is saved according to the time dimension and exported as a time series data set.

4. The satellite solar array data-driven modeling method based on working condition identification according to claim 1 is characterized in that: Adding labels to the preprocessed telemetry data in step 3 specifically includes: obtaining labels according to the "light / ground shadow state" feature, when the light / ground shadow state is 0, the label is the light state; when the light / ground shadow state is 1, the label is the ground shadow state; the data label before the ground shadow state and after the light state is the shadow entry state; the data label after the ground shadow state and before the light state is the shadow exit state; the label of the position holding mode is obtained according to the characteristics of the attitude and orbit control working mode.

5. The satellite solar array data-driven modeling method based on working condition identification according to claim 1 is characterized in that: In step 4, the CNN network consists of 4-6 convolutional layer structures and 1-2 fully connected layer structures. Each convolutional layer consists of a one-dimensional convolutional layer, a batch normalization layer, a nonlinear activation function layer and a maximum pooling layer. Each convolutional layer structure is connected in sequence, and the fully connected layer is connected after the last convolutional layer structure to convert the data into the final number of classification categories.

6. The satellite solar array data-driven modeling method based on working condition identification according to claim 1 is characterized in that: In step 4, the training set, validation set, and test set are used to train, validate, and test the CNN network in the data-driven model. Specifically, the following steps are performed: The CNN network is trained using the training set, the structural parameters, weights and biases of the CNN network are verified using the verification set, the network parameters are optimized and saved, and the accuracy of the solar array working condition classification and recognition is tested using the test set. If the accuracy is lower than a preset value, the model is retrained until the accuracy is greater than the preset value; The expression of the accuracy of solar array working condition classification and identification is: In formula (1), TP i is the number of samples correctly predicted as category i, C is the number of working conditions, and N is the total number of samples.

7. The satellite solar array data-driven modeling method based on working condition identification according to claim 1 is characterized in that: In step 5, the steps of re-dividing the telemetry data after feature selection and data cleaning into data sets of five working conditions according to labels include: Step 5.1: Divide the time series data set into data sets under five working conditions according to the labels of the telemetry data, "illumination state, ground shadow state, shadow entry state, shadow exit state, and position holding state", wherein the data sets under the five working conditions include an illumination state data set, a ground shadow state data set, a shadow entry state data set, a shadow exit state data set, and a position holding state data set; Step 5.2: Determine the time step tw of the data set corresponding to each working condition, and use the sliding window method to generate the input sequence and the corresponding target value, where the input sequence is the first tw time step data, and the target value is the tw+1th data. Starting from the starting position of the time series, slide forward one time step each time until the end of the sequence to complete the creation of fixed-length input and output sequences.

8. The satellite solar array data-driven modeling method based on working condition identification according to claim 1 is characterized in that: The feature dimensions of the LSTM network in step 5 include but are not limited to solar array temperature, analog solar sensor current, solar sub-array voltage, and light / ground shadow status. The target feature of the LSTM network is the solar array current. Each LSTM network consists of 2-6 LSTM sub-network layer structures and one fully connected layer, wherein the LSTM sub-network layer structure includes a number of hidden layer units and nonlinear activation functions. The LSTM sub-network layer structures are connected in sequence, and the fully connected layer structure includes forward propagation and reverse propagation processes. The fully connected layer is connected after the last LSTM sub-network layer structure, and the output size of the fully connected layer is 1, which is used to fit the output parameters of the solar array.

9. The satellite solar array data-driven modeling method based on working condition identification according to claim 1 is characterized in that: The LSTM networks under the five working conditions in step 5 include: the LSTM network under the illumination state, the LSTM network under the ground shadow state, the LSTM network under the shadow entry state, the LSTM network under the shadow exit state, and the LSTM network under the position keeping state. The training, verification, and testing of the LSTM networks under the five working conditions in the data-driven model specifically include: Use the illumination state training set to train the LSTM network under illumination state conditions, use the illumination state verification set to verify the structural parameters, weights and biases of the LSTM network under illumination state conditions, optimize the network parameters and save them, and use the illumination state test set to verify the fitting effect of the illumination state selection solar array target features; Use the earth shadow state training set to train the LSTM network under the earth shadow state condition, use the earth shadow state verification set to verify the structural parameters, weights and biases of the LSTM network under the earth shadow state condition, optimize the network parameters and save them, and use the earth shadow state test set to verify the fitting effect of the earth shadow state to select the target characteristics of the solar array; The shadow state training set is used to train the LSTM network under the shadow state working condition, and the shadow state verification set is used to verify the structural parameters, weights and biases of the LSTM network under the shadow state working condition. The network parameters are optimized and saved, and the shadow state test set is used to test the fitting effect of the target characteristics of the solar array selected under the shadow state. The LSTM network under the shadow state working condition is trained using the shadow state training set, and the structural parameters, weights and biases of the LSTM network under the shadow state working condition are verified using the shadow state verification set. The network parameters are optimized and saved, and the shadow state test set is used to test the fitting effect of the target characteristics of the solar array selected under the shadow state. The LSTM network under the position-holding state working condition is trained using the position-holding state training set, and the structural parameters, weights and biases of the LSTM network under the position-holding state working condition are verified using the position-holding state verification set. The network parameters are optimized and saved, and the position-holding state test set is used to verify the fitting effect of the target features of the solar array selected in the position-holding state.

10. The satellite solar array data-driven modeling method based on working condition identification according to claim 1, characterized in that: The predicted output parameters of the solar array obtained in step 6 specifically include: According to the working condition recognition result of the CNN network, the corresponding LSTM network is selected for prediction. The predicted future output of the solar array is 5-20 time steps, and the error index of the data-driven model is calculated. If the error index is greater than the preset threshold, the CNN network and LSTM network are retrained until the error index is ≤ the preset threshold, and the predicted output of the solar array is obtained; The expression of the error index is: In formulas (2) and (3), N is the total number of samples, y i is the true value of the i-th sample, is the model output value of the i-th sample, MAE is the mean absolute error, and MSE is the mean square error.

Images