Satellite temperature prediction method based on Monte Carlo quantile regression
By combining LSTM and MCQR models, using Monte Carlo quantile regression method to process satellite temperature data, the problem of inaccurate satellite temperature prediction in the prior art is solved, and an effective solution for temperature prediction and uncertainty quantification is achieved.
Patent Information
- Application Number
- CN202411644482.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2025-03-14
AI Technical Summary
It is difficult to build an accurate satellite temperature prediction model in the prior art, and the probability of satellite system failure or sudden failure is high.
Using a Monte Carlo quantile regression method, combined with LSTM and MCQR models, the pre-trained LSTM-MCQR model is input through the splicing of time sequence temperature data and random quantile levels, multiple predicted temperatures are output and the final predicted temperature and uncertainty are calculated.
Accurate prediction and uncertainty quantification of satellite temperature are achieved, the probability of satellite system failure is reduced, and the reliability of temperature data analysis is improved.
Smart Images
Figure CN119622657A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of satellite temperature prediction, and in particular to a satellite temperature prediction method based on Monte Carlo quantile regression. Background Art
[0002] In engineering, satellite on-orbit temperature data usually comes from various temperature sensors. Since temperature data will affect the reliability of satellites to a certain extent, for example, sudden extreme high temperatures may cause unexpected failures of satellites, it is very necessary to effectively analyze temperature data in preparation for further reliability analysis. In order to reduce the probability of satellite system failures or sudden failures caused by abnormal satellite temperature, building an accurate temperature prediction model is an urgent problem to be solved. Summary of the invention
[0003] In order to solve some or all of the technical problems existing in the above-mentioned prior art, the present invention provides a satellite temperature prediction method based on Monte Carlo quantile regression.
[0004] The technical solution of the present invention is as follows:
[0005] A satellite temperature prediction method based on Monte Carlo quantile regression is provided, which includes:
[0006] Obtain satellite temperature data at multiple time points before the time point to be predicted;
[0007] constructing time series temperature data according to the satellite temperature data;
[0008] The time series temperature data are horizontally spliced with a random quantile ranging from 0 to 1 for multiple times to obtain multiple input data;
[0009] The input data are respectively input into the pre-trained LSTM-MCQR model to obtain a plurality of corresponding predicted temperatures, wherein the LSTM-MCQR model comprises an LSTM layer, a flattening layer and a linear layer, the LSTM layer is used to learn and extract the long-term dependency features of the sequence from the input data, the flattening layer is used to flatten and reduce the multi-dimensional features output by the LSTM layer, and the linear layer is used to perform weighted combination of the flattened features to output the predicted temperature;
[0010] The average of the multiple predicted temperatures is used as the final predicted temperature, and the standard deviation of the multiple predicted temperatures is used as the uncertainty.
[0011] In one embodiment of the present invention, the training process of the LSTM-MCQR model is as follows:
[0012] Constructing a data set, each set of data in the data set includes time series temperature data and corresponding temperature labels;
[0013] A random quantile level ranging from 0 to 1 is concatenated to the time series temperature data in each group of data in the data set to construct a training set;
[0014] Build an LSTM-MCQR initial model, which includes an LSTM layer, a flattening layer and a linear layer. The LSTM layer is used to learn and extract the long-term dependency features of the sequence from the input data, the flattening layer is used to flatten and reduce the multidimensional features output by the LSTM layer, and the linear layer is used to perform weighted combination of the flattened features to output the predicted temperature;
[0015] The time series temperature data at the spliced random quantile level is taken as input, and the corresponding temperature label is taken as output. The LSTM-MCQR initial model is trained to update the model parameters. After the training, the LSTM-MCQR model is obtained.
[0016] In one embodiment of the present invention, a method for constructing the time series temperature data and the corresponding temperature labels in the data set includes:
[0017] Set the size and step size of the sliding window;
[0018] The sequence of multiple satellite temperature data selected by the sliding window in the satellite temperature data is a time series temperature data, and the first temperature data after the sliding window is the corresponding temperature label;
[0019] Each time the sliding window moves a step, a corresponding time series temperature data and temperature label are obtained.
[0020] In an embodiment of the present invention, the step size of the sliding window is 1.
[0021] In one embodiment of the present invention, the training process of the LSTM-MCQR model is as follows:
[0022] Constructing a data set, each set of data in the data set includes time series temperature data and corresponding temperature labels;
[0023] A random quantile level ranging from 0 to 1 is concatenated to the time series temperature data in each group of data in the data set to construct a training set;
[0024] Build an LSTM-MCQR initial model, which includes an LSTM layer, a flattening layer and a linear layer. The LSTM layer is used to learn and extract the long-term dependency features of the sequence from the input data, the flattening layer is used to flatten and reduce the multidimensional features output by the LSTM layer, and the linear layer is used to perform weighted combination of the flattened features to output the predicted temperature;
[0025] The time series temperature data at the spliced random quantile level is taken as input, and the corresponding temperature label is taken as output. The LSTM-MCQR initial model is trained to update the model parameters. After the training, the LSTM-MCQR model is obtained.
[0026] In an embodiment of the present invention, the number of linear layers is 3.
[0027] In one embodiment of the present invention, the activation function of the linear layer is a ReLU activation function.
[0028] In one embodiment of the present invention, the loss function of the LSTM-MCQR is defined as:
[0029]
[0030] in, is the loss function, Y is the temperature label in the training set, is the predicted temperature of the model, d is the number of training samples, α i is the random quantile level.
[0031] In one embodiment of the present invention, the Adam algorithm is used as the neural network gradient optimization algorithm when training the LSTM-MCQR model.
[0032] In one embodiment of the present invention, when training the LSTM-MCQR model, the training set data is divided into data of a specific batch size and trained in batches.
[0033] The main advantages of the technical solution of the present invention are as follows:
[0034] The satellite temperature prediction method based on Monte Carlo quantile regression of the present invention combines LSTM with MCQR, combines time series temperature data with random quantile levels as the input of the LSTM-MCQR model, and calculates the final predicted temperature value and uncertainty through multiple predicted temperatures output by the LSTM-MCQR model, which can not only give full play to the ability of LSTM to mine and extract time series information for temperature prediction, but also quantify uncertainty based on the MCQR principle. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0036] Figure 1 This is a flow chart of a satellite temperature prediction method based on Monte Carlo quantile regression according to an embodiment of the present invention;
[0037] Figure 2 A flowchart of a satellite temperature prediction method based on Monte Carlo quantile regression according to an embodiment of the present invention;
[0038] Figure 3 A schematic diagram of data preparation in a satellite temperature prediction method based on Monte Carlo quantile regression according to an embodiment of the present invention;
[0039] Figure 4 This is a diagram of the LSTM-MCQR model architecture in a satellite temperature prediction method based on Monte Carlo quantile regression according to an embodiment of the present invention;
[0040] Figure 5 This is a flow chart of the training process of the LSTM-MCQR model in the satellite temperature prediction method based on Monte Carlo quantile regression according to an embodiment of the present invention;
[0041] Figure 6 The present invention is a flowchart of using the LSTM-MCQR model to perform prediction in a satellite temperature prediction method based on Monte Carlo quantile regression according to an embodiment of the present invention. DETAILED DESCRIPTION
[0042] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0043] The technical solution provided by the embodiments of the present invention is described in detail below with reference to the accompanying drawings.
[0044] The embodiment of the present invention provides a satellite temperature prediction method based on Monte Carlo quantile regression, as shown in the attached Figure 1 As shown, including:
[0045] S1, obtaining satellite temperature data at multiple time points before the time point to be predicted.
[0046] Predicting temperature means giving a predicted value of future temperature based on the changing trend of historical temperature data. Therefore, historical temperature data is the prerequisite for predicting temperature. This step obtains satellite temperature data to provide a data basis for subsequent predictions.
[0047] S2, construct time series temperature data based on satellite temperature data.
[0048] The temperature data at different time points on the satellite is a typical time series data. Long Short Term Memory (LSTM) is a common method for processing time series data prediction. It uses a special gate structure to determine the characteristics of information transmitted under optimal conditions. In recent years, LSTM has been widely used in the prediction of remaining service life in the fields of rolling bearings, lithium batteries, circuit systems, etc., with good performance. The present invention intends to use the ability of LSTM to extract time series data for prediction to predict satellite temperature. This step constructs time series temperature data to provide a basis for the use of LSTM.
[0049] S3, horizontally concatenate the time series temperature data with a random quantile ranging from 0 to 1 for multiple times to obtain multiple input data.
[0050] In terms of data uncertainty quantification, a widely used method is Monte Carlo Quantile Regression (MCQR) based on quantile regression theory. During the training process, the MCQR method combines the quantile levels randomly sampled from the uniform distribution with the training data, which can effectively quantify the uncertainty of various data and has been applied in the reconstruction of the temperature field of the convolutional neural network. Based on the idea of the MCQR method, in this step, random quantile levels are spliced in the time series temperature data, and multiple input data are constructed so that the model can have multiple outputs, thereby completing the uncertainty quantification.
[0051] S4, input multiple input data into the pre-trained LSTM-MCQR model respectively to obtain multiple corresponding predicted temperatures, wherein the LSTM-MCQR model includes an LSTM layer, a flattening layer and a linear layer. The LSTM layer is used to learn and extract the long-term dependency features of the sequence from the input data, the flattening layer is used to flatten and reduce the multi-dimensional features output by the LSTM layer, and the linear layer is used to perform weighted combination of the flattened features to output the predicted temperature.
[0052] It is understandable that in the field of machine learning, before using the model for prediction, data preparation, model building or training is required in advance. Then in the application scenario, the data is input into the model to obtain the output of the model as the prediction result. The preparation of the training data set, the building or prediction of the model only needs to be done once, and the prediction of subsequent application scenarios can be performed repeatedly. Therefore, the satellite temperature prediction method of the present invention uses a pre-trained LSTM-MCQR model. The process of training the model is not performed every time a prediction is made, so it is not described in detail. The detailed training process can be found in the subsequent section.
[0053] While maintaining the basic architecture of the LSTM model, the LSTM-MCQR model combines the quantile level of random sampling in the uniform random distribution with the training data in the processing of input data. It can not only utilize the LSTM's ability to process time series data for prediction, but also quantify data uncertainty based on the principle of MCQR.
[0054] S5, uses the mean of multiple predicted temperatures as the final predicted temperature, and uses the standard deviation of multiple predicted temperatures as the uncertainty.
[0055] In summary, the satellite temperature prediction method based on Monte Carlo quantile regression provided by the embodiment of the present invention combines LSTM with MCQR. After the input data is combined with the random quantile level, it is used as the input of the LSTM-MCQR model. The final predicted temperature value and uncertainty are calculated through multiple predicted temperatures output by the LSTM-MCQR model. This can not only give full play to the ability of LSTM to mine and extract time series information for temperature prediction, but also quantify uncertainty based on the MCQR principle.
[0056] The following is a detailed description of the steps and principles of the satellite temperature prediction method based on Monte Carlo quantile regression provided by an embodiment of the present invention:
[0057] Please see attached Figure 2 The complete process of the satellite temperature prediction method based on Monte Carlo quantile regression includes three stages: data preparation, temperature prediction model construction and training, model prediction, and data uncertainty quantification. The data preparation stage is used to build a training data set. In the temperature prediction model construction and training stage, the LSTM-MCQR initial model is built and the model is trained with the data in the training data set. In the model prediction and data uncertainty quantification stage, the trained LSTM-MCQR model is used to predict temperature and quantify uncertainty.
[0058] 1. Data Preparation
[0059] The data preparation phase includes data acquisition and data collation. Data acquisition is to obtain the original satellite temperature data, which can be measured by sensors on the satellite in orbit and transmitted to the ground receiving equipment through wireless communication.
[0060] Data sorting is to convert the original satellite temperature data into time series data that LSTM can recognize and use. Generally speaking, satellite temperature data is a typical time series data. Figure 3 As shown, a specific time t corresponds to a temperature value x. Assume that the acquired data set samples are represented as {x1,x2,...,x n}. Use a sliding window (similar to a time window) to extract different sequences from the data set sample set. By setting different window sizes and step sizes, you can extract different sequences from the data set samples {x1, x2, ..., x n} to extract time series temperature data of different lengths and quantities. Figure 3 The process of reorganizing a data set using a window of size seq (1≤seq≤n) is shown. Assume that the step size is 1 and the first sequence X1 consists of {x1,x2,...,x seq}, the second sequence X2 consists of {x2,x3,...,x seq+1}. Similarly, the input data {X1,X2,...,X d Of course, those skilled in the art can set the step size of the sliding window to 2, 3, etc., which is not specifically limited in the embodiments of the present invention. For example, when the step size is 2, the first sequence X1 consists of {x1, x2, ..., x seq}, the second sequence X2 consists of {x3,x4,...,x seq+2}composition.
[0061] Figure 3 It also shows the formation of temperature labels. The first sequence X1 is used to predict the time point t seq+1 The temperature of the first label is Y1 (Y1 = x seq+1 ) indicates. X2 predicts t seq+2 The temperature at the time, so we get Y2 (Y2 = x seq+2 ). Therefore, label Y i (i=1,2,...,d) and X i (i=1,2,...,d) one-to-one correspondence. Finally, the constructed training data set is represented as {(X i ,Y i )|i=1,2,...,d}.
[0062] 2. Construction and training of temperature prediction model
[0063] According to the MCQR method, each X i will match a quantile level α randomly sampled from a uniform distribution U(0,1) i , and {(X i ,α i )|i=1,2,...,d} is used as the input data set of the model. The LSTM-MCQR model is as follows Figure 4As shown in the figure, it includes LSTM layer, flattening layer and linear layer. The LSTM layer is used to learn and extract the long-term dependency features of the sequence from the input data. The flattening layer is used to flatten and reduce the multi-dimensional features output by the LSTM layer. The linear layer is used to weightedly combine the flattened features to output the predicted temperature. During training, the training data set {((X i ,α i ),Y i )|i=1,2,...,d} is input into an LSTM layer with h hidden neurons to obtain the relationship between the sequences. Then, a flattening layer and three linear layers with ReLU activation function are applied to flatten and reduce the multi-dimensional features output by the LSTM layer. Finally, the linear layer output is used to predict the temperature.
[0064] For ease of description, the LSTM-MCQR model is denoted as M(X,α;θ), where θ is the model parameter. The predicted temperature is denoted as Based on the input training data set {((X i ,α i ),Y i )|i=1,2,...,d},Monte Carlo Pinball loss function of LSTM-MCQR model Defined as:
[0065]
[0066] Through the above definition and model construction, model training can be performed to learn and find the appropriate parameter θ. It is worth noting that in each cycle of the training process, the quantile level α will be randomly sampled from the uniform distribution U(0,1). Then the reorganized data {(X i ,α i )|i=1,2,...,d} as the input data of the model, and {Y i |i=1,2,...,d} as label data. In addition, the newly formed training data set {((X i ,α i ),Y i )|i=1,2,...,d} will be divided into data of a specific batch size, and the total batch size is N batch The Adam optimization method is used to update the model parameters θ, and the loss function is shown in formula (1). In summary, the flowchart of the training process can be found in the attached Figure 5 .
[0067] 3. Model prediction and uncertainty quantification
[0068] In the model prediction process, the total number of predictions is represented by D pre The temperature data is X pre, which contains the temperature data of seq moments, and forms the time series data that conforms to the model input according to the sliding window method. pre Combined with different quantile levels α sampled from a uniform distribution U(0,1). Thus, (X pre ,α λ ) is the input data for training the model. pre ,α λ ) is input into the model trained in Section 2 to obtain different predicted temperatures
[0069] In order to quantitatively describe the uncertainty of the data, all The average value is used to calculate the approximate predicted temperature As the final predicted temperature of this method, the uncertainty is the standard deviation of all predicted values. and The calculation formula is shown in formula (2).
[0070]
[0071] In summary, the process of satellite temperature prediction and uncertainty quantification using the LSTM-MCQR model can be Figure 6 .
[0072] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In addition, "front", "back", "left", "right", "upper" and "lower" in this article are all referenced to the placement state shown in the accompanying drawings.
[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A satellite temperature prediction method based on Monte Carlo quantile regression, characterized in that: include: Obtain satellite temperature data at multiple time points before the time point to be predicted; constructing time series temperature data according to the satellite temperature data; The time series temperature data are horizontally spliced with a random quantile ranging from 0 to 1 for multiple times to obtain multiple input data; The input data are respectively input into the pre-trained LSTM-MCQR model to obtain a plurality of corresponding predicted temperatures, wherein the LSTM-MCQR model comprises an LSTM layer, a flattening layer and a linear layer, the LSTM layer is used to learn and extract the long-term dependency features of the sequence from the input data, the flattening layer is used to flatten and reduce the multi-dimensional features output by the LSTM layer, and the linear layer is used to perform weighted combination of the flattened features to output the predicted temperature; The average of the multiple predicted temperatures is used as the final predicted temperature, and the standard deviation of the multiple predicted temperatures is used as the uncertainty.
2. The satellite temperature prediction method based on Monte Carlo quantile regression according to claim 1 is characterized in that: The training process of the LSTM-MCQR model is as follows: Constructing a data set, each set of data in the data set includes time series temperature data and corresponding temperature labels; A random quantile level ranging from 0 to 1 is concatenated to the time series temperature data in each group of data in the data set to construct a training set; Build an LSTM-MCQR initial model, which includes an LSTM layer, a flattening layer and a linear layer. The LSTM layer is used to learn and extract the long-term dependency features of the sequence from the input data, the flattening layer is used to flatten and reduce the multidimensional features output by the LSTM layer, and the linear layer is used to perform weighted combination of the flattened features to output the predicted temperature; The time series temperature data at the spliced random quantile level is taken as input, and the corresponding temperature label is taken as output. The LSTM-MCQR initial model is trained to update the model parameters. After the training, the LSTM-MCQR model is obtained.
3. The satellite temperature prediction method based on Monte Carlo quantile regression according to claim 2 is characterized in that: The method for constructing the time series temperature data and the corresponding temperature labels in the data set includes: Set the size and step size of the sliding window; The sequence of multiple satellite temperature data selected by the sliding window in the satellite temperature data is a time series temperature data, and the first temperature data after the sliding window is the corresponding temperature label; Each time the sliding window moves a step, a corresponding time series temperature data and temperature label are obtained.
4. The satellite temperature prediction method based on Monte Carlo quantile regression according to claim 3 is characterized in that: The step size of the sliding window is 1.
5. The satellite temperature prediction method based on Monte Carlo quantile regression according to claim 4 is characterized in that: The linear layer has three layers.
6. The satellite temperature prediction method based on Monte Carlo quantile regression according to claim 5 is characterized in that: The activation function of the linear layer is the ReLU activation function.
7. The satellite temperature prediction method based on Monte Carlo quantile regression according to claim 6 is characterized in that: The loss function of the LSTM-MCQR is defined as: in, is the loss function, Y is the temperature label in the training set, is the predicted temperature of the model, d is the number of training samples, α i is the random quantile level.
8. The satellite temperature prediction method based on Monte Carlo quantile regression according to claim 2 is characterized in that: The Adam algorithm is used as the neural network gradient optimization algorithm when training the LSTM-MCQR model.
9. The satellite temperature prediction method based on Monte Carlo quantile regression according to claim 2, characterized in that: When training the LSTM-MCQR model, the training set data is divided into data of a specific batch size and trained in batches.