CN116304884A - Spacecraft telemetry data health prediction method, system, equipment and storage medium - Google Patents
Spacecraft telemetry data health prediction method, system, equipment and storage medium Download PDFInfo
- Publication number
- CN116304884A CN116304884A CN202310524696.7A CN202310524696A CN116304884A CN 116304884 A CN116304884 A CN 116304884A CN 202310524696 A CN202310524696 A CN 202310524696A CN 116304884 A CN116304884 A CN 116304884A
- Authority
- CN
- China
- Prior art keywords
- data
- telemetry data
- telemetry
- predicted
- prediction model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000036541 health Effects 0.000 title claims abstract description 92
- 238000000034 method Methods 0.000 title claims abstract description 54
- 230000002159 abnormal effect Effects 0.000 claims abstract description 45
- 230000000007 visual effect Effects 0.000 claims abstract description 31
- 239000013598 vector Substances 0.000 claims description 63
- 238000012549 training Methods 0.000 claims description 62
- 230000006870 function Effects 0.000 claims description 37
- 230000015654 memory Effects 0.000 claims description 27
- 238000004422 calculation algorithm Methods 0.000 claims description 24
- 239000011159 matrix material Substances 0.000 claims description 22
- 238000005457 optimization Methods 0.000 claims description 17
- 230000004913 activation Effects 0.000 claims description 12
- 238000007781 pre-processing Methods 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 7
- 238000011176 pooling Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 5
- 230000003044 adaptive effect Effects 0.000 claims 2
- 230000002547 anomalous effect Effects 0.000 claims 1
- 230000008569 process Effects 0.000 description 9
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 6
- 238000001514 detection method Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000011049 filling Methods 0.000 description 4
- 238000009499 grossing Methods 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 206010000117 Abnormal behaviour Diseases 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 101100517648 Saccharomyces cerevisiae (strain ATCC 204508 / S288c) NUM1 gene Proteins 0.000 description 1
- 101100129590 Schizosaccharomyces pombe (strain 972 / ATCC 24843) mcp5 gene Proteins 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 230000006386 memory function Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/22—Indexing; Data structures therefor; Storage structures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Navigation (AREA)
Abstract
The present disclosure provides a spacecraft telemetry data health prediction method, system, device and storage medium, comprising reading historical telemetry data matched with telemetry data prediction requirements from a telemetry database based on the acquired telemetry data prediction requirements, inputting a pre-trained data health prediction model, outputting predicted telemetry data of a future time period corresponding to the telemetry data prediction requirements, and matching the predicted telemetry data with the telemetry data prediction requirements; if the predicted telemetry data and the historical telemetry data are matched, the predicted telemetry data and the historical telemetry data after abnormal data are removed are displayed in a visual mode at the same time; if the data deviation is not matched with the actual telemetry data, determining the data deviation of the predicted telemetry data and the actual telemetry data, and optimizing a data health prediction model based on the data deviation and the actual telemetry data added with Gaussian white noise. The method disclosed by the invention can achieve the aim of improving the trend prediction accuracy.
Description
Technical Field
The disclosure relates to the field of telemetry technology, and in particular relates to a spacecraft telemetry data health prediction method, system, equipment and storage medium.
Background
With the increase of the number of the on-orbit spacecrafts, the method is also particularly important for the comprehensive management of the spacecrafts.
The current health prediction of the spacecraft at home and abroad has the following defects:
1. the algorithm is single, only aiming at the data characteristics, no time sequence characteristics are added, the predicted output only has data, no corresponding time matching exists, only the data is analyzed, and the whole prediction service cannot be perfected;
2. the accuracy of data output is not high enough, the original data of the spacecraft has some fluctuation characteristics, and a machine learning algorithm cannot accurately learn some fine changes during trend learning;
3. the machine learning training time is long, the model parameters need to be manually adjusted when the model is trained, and a great amount of time is spent;
4. the traditional algorithm cannot verify the prediction result, so that the prediction accuracy of the whole algorithm is not particularly considered.
The existing known algorithm can not ensure the high efficiency of algorithm calculation and has higher accuracy, so that the problems of high time consumption and low accuracy of spacecraft prediction service are caused. Because of the large number of on-orbit spacecraft data, obvious data parameter change and long on-orbit time, in order to ensure accurate control of ground manpower on the health state of the spacecraft, an algorithm is needed to be introduced to complete the service of spacecraft data prediction.
The information disclosed in the background section of this application is only for enhancement of understanding of the general background of this application and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person skilled in the art.
Disclosure of Invention
The embodiment of the disclosure provides a spacecraft telemetry data health prediction method, a system, equipment and a storage medium, which can solve the technical problems in the prior art.
In a first aspect of embodiments of the present disclosure,
the utility model provides a spacecraft telemetry data health prediction method, which comprises the following steps:
reading historical telemetry data matching the telemetry data predicted demand from a telemetry database based on the acquired telemetry data predicted demand;
after preprocessing the historical telemetry data, inputting a pre-trained data health prediction model, outputting predicted telemetry data of a future time period corresponding to the telemetry data prediction requirement by the pre-trained data health prediction model, and matching the predicted telemetry data with the telemetry data prediction requirement;
if the predicted telemetry data are matched with the historical telemetry data, detecting whether the predicted telemetry data are abnormal data or not through the data health prediction model, and if the predicted telemetry data are not abnormal data, displaying the predicted telemetry data and the historical telemetry data in a visual mode at the same time; if the abnormal data exists, eliminating the abnormal data, and displaying the predicted telemetry data and the historical telemetry data after eliminating the abnormal data in a visual mode at the same time;
If the data deviation is not matched, determining the data deviation of the predicted telemetry data and the actual telemetry data, adding Gaussian white noise into the actual telemetry data, and optimizing a data health prediction model based on the data deviation and the actual telemetry data added with the Gaussian white noise until the predicted telemetry data output by the optimized data health prediction model is matched with the telemetry data prediction requirement.
In an alternative embodiment of the present invention,
the method for optimizing the data health prediction model based on the data deviation and the actual telemetry data added with Gaussian white noise comprises the following steps:
dividing actual telemetry data added with Gaussian white noise into a first data sequence and a second data sequence according to a preset proportion, and respectively extracting a first characteristic vector set corresponding to the first data sequence and a second characteristic vector set corresponding to the second data sequence according to the weight of a convolution kernel of the convolution layer, an index value of a filter and a first activation function corresponding to the convolution layer through the convolution layer of a data health prediction model to be optimized;
respectively distributing a first weight matrix for the first feature vector set and a second weight matrix for the second feature vector set, and integrating the first feature vector set and the second feature vector set into a comprehensive feature vector set;
Updating information of the comprehensive feature vector set through a hidden layer of the data health prediction model to be optimized, inputting the updated comprehensive feature vector into an output layer of the data health prediction model to be optimized, and determining a prediction output result;
based on the data deviation of the prediction output result and the actual telemetry data added with Gaussian white noise, a loss function is established, minimized training of the data deviation is carried out through a self-adaptive learning rate optimization algorithm, a target weight parameter is determined, and the weight parameter of the data health prediction model to be optimized is updated according to the target weight parameter.
In an alternative embodiment of the present invention,
the hidden layer comprises a memory unit and a cell unit, wherein the memory unit is used for selectively memorizing the content of each time step, and the cell unit is used for deciding whether to memorize the content of the previous moment;
and updating information of the comprehensive feature vector set through a hidden layer of the data health prediction model to be optimized, inputting the updated comprehensive feature vector into an output layer of the data health prediction model to be optimized, and determining a prediction output result comprises the following steps:
Determining a first output of the memory unit according to the comprehensive feature vector set, a first weight vector corresponding to the memory unit, a first bias parameter and a second activation function corresponding to the memory unit;
determining a second output of the cell unit according to the comprehensive feature vector set, a second weight vector corresponding to the cell unit, a second bias parameter and a cell state at a moment on the cell unit;
and inputting the first output and the second output into the output layer, and determining a predicted output result according to a third weight vector and a third bias parameter corresponding to the output layer.
In an alternative embodiment of the present invention,
the loss function is established, the data deviation minimization training is carried out through the self-adaptive learning rate optimization algorithm, and the target weight parameter determination comprises the following steps:
the loss function is shown in the following formula:
；
wherein,,LOSSthe loss value is indicated as such,Nthe number of iterations is indicated and,y t (i)represent the firstiThe index of the next iteration predicts the output value,a t (i)represent the firstiThe input of the iterative memory unit;
the minimizing training of the data deviation by the self-adaptive learning rate optimizing algorithm is shown in the following formula:
wherein,,P t-1 the data deviation is indicated as such, tThe time of day is indicated as such,Tthe total duration of time is indicated and,u o,t 、u f,t respectively representtUpdating the error term of the gate at the moment and outputting the error term of the gate,W o,t 、W f,t respectively representtThe weight of the gate is updated at the moment and the weight of the gate is output.
In an alternative embodiment of the present invention,
the detecting, by the data health prediction model, whether abnormal data exists in the predicted telemetry data includes:
based on a convolution pooling layer of the data health prediction model, obtaining local sequence visual field characteristics of the predicted telemetry data;
and outputting positive sample characteristics and negative sample characteristics in the local sequence visual field characteristics through a trained positive and negative sample classifier of the data health prediction model according to the local sequence visual field characteristics, wherein the positive sample characteristics are used for indicating characteristics corresponding to normal telemetry data in the local sequence visual field characteristics, and the negative sample characteristics are used for indicating characteristics corresponding to abnormal telemetry data in the local sequence visual field characteristics.
In an alternative embodiment of the present invention,
before the positive sample feature and the negative sample feature in the local sequence visual field feature are output by the trained positive and negative sample classifier of the data health prediction model, the method further comprises training the positive and negative sample classifier:
Mapping a first training subset and a second training subset in the training data set to the same distribution space through a mapping function based on a pre-acquired training data set, and training positive and negative sample classifiers to be trained in the distribution space, wherein the first training subset is used for indicating data with labels in the training data set, and the second training subset is used for indicating data without labels in the training data set;
and setting a classification loss function based on the first training subset and the second training subset, and dynamically updating the classification cost weight and the classification optimization target according to the classification loss function until the classification accuracy of the positive and negative sample classifier reaches a preset threshold.
In an alternative embodiment of the present invention,
and dynamically updating the classification cost weight and the classification optimization target according to the classification loss function by the following formula:
wherein,,Q(r)the weight of the classification cost is represented as,Ethe cost matrix is represented by a representation of the cost matrix,t n represent the firstnThe desired output of the plurality of outputs,p n represent the firstnA plurality of prediction outputs;
；/>
；
wherein,,H(r)the classification optimization objective is represented as such,M、Kthe number of positive samples and the number of negative samples are represented respectively,l pos 、l neg respectively represent the cross entropy loss value corresponding to the positive sample and the negative sample The corresponding cross-entropy loss value is used,
、/>
respectively represent the firstiMisclassification cost weight corresponding to positive samplejMisclassification cost weight corresponding to each negative sample, < ->
Representation oft+1The misclassification cost weight corresponding to the current sample at the moment,U over represents the ratio of the number of positive samples to the number of negative samples, +.>
Geometric mean representing the output result of the current sample, +.>
Indicating the accuracy of the output result of the current sample.
In a second aspect of the embodiments of the present disclosure,
there is provided a spacecraft telemetry data health prediction system comprising:
a first unit for reading historical telemetry data matching the telemetry data prediction requirements from a telemetry database based on the acquired telemetry data prediction requirements;
the second unit is used for preprocessing the historical telemetry data, inputting a pre-trained data health prediction model, outputting predicted telemetry data of a future time period corresponding to the telemetry data prediction requirement by the pre-trained data health prediction model, and matching the predicted telemetry data with the telemetry data prediction requirement;
the third unit is used for detecting whether abnormal data exist in the predicted telemetry data through the data health prediction model if the predicted telemetry data are matched with the historical telemetry data, and displaying the predicted telemetry data and the historical telemetry data in a visual mode at the same time if the abnormal data do not exist; if the abnormal data exists, eliminating the abnormal data, and displaying the predicted telemetry data and the historical telemetry data after eliminating the abnormal data in a visual mode at the same time;
And a fourth unit, configured to determine a data deviation between the predicted telemetry data and the actual telemetry data if the predicted telemetry data and the actual telemetry data are not matched, and add white gaussian noise to the actual telemetry data, and optimize a data health prediction model based on the data deviation and the actual telemetry data added with white gaussian noise until the predicted telemetry data output by the optimized data health prediction model matches the telemetry data prediction requirement.
In a third aspect of the embodiments of the present disclosure,
there is provided an electronic device including:
a processor;
a memory for storing processor-executable instructions;
wherein the processor is configured to invoke the instructions stored in the memory to perform the method described previously.
In a fourth aspect of embodiments of the present disclosure,
there is provided a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the method as described above.
The validity of the algorithm is judged by a loss function method, and the stored historical data of the database can be used as another effective measure to further check the rationality of the algorithm; after Gaussian white noise based on the original data is added, the original trend change is met, the predicted data is optimized, and the aim of improving the trend prediction accuracy is fulfilled.
According to the method, the unbalanced ratio is used as penalty of cost-sensitive misclassification, the problem that a model caused by data unbalance on classification tasks is biased to normal sample fitting can be integrally solved, in addition, a small-batch training mode is adopted in the neural network training process, a fixed cost matrix cannot be well adapted to unbalance of local area distribution, and the dynamic-change misclassification cost weight is utilized for self-adaptive updating, so that not only can the integral sample unbalance be considered, but also the unbalance on the local small-batch training can be considered.
Drawings
FIG. 1 is a flow chart of a method for spacecraft telemetry data health prediction in an embodiment of the disclosure;
FIG. 2 is a logic diagram of a spacecraft telemetry data health prediction method in accordance with an embodiment of the present disclosure;
fig. 3 is a schematic structural diagram of a spacecraft telemetry data health prediction system according to an embodiment of the disclosure.
Detailed Description
For the purposes of making the objects, technical solutions and advantages of the embodiments of the present disclosure more apparent, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure, and it is apparent that the described embodiments are only some embodiments of the present disclosure, not all embodiments. Based on the embodiments in this disclosure, all other embodiments that a person of ordinary skill in the art would obtain without making any inventive effort are within the scope of protection of this disclosure.
The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above-described figures, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the disclosure described herein may be capable of operation in sequences other than those illustrated or described herein.
It should be understood that, in various embodiments of the present disclosure, the size of the sequence number of each process does not mean that the execution sequence of each process should be determined by its functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present disclosure.
It should be understood that in this disclosure, "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements that are expressly listed or inherent to such process, method, article, or apparatus.
It should be understood that in this disclosure, "plurality" means two or more. "and/or" is merely an association relationship describing an association object, and means that three relationships may exist, for example, and/or B may mean: a exists alone, A and B exist together, and B exists alone. The character "/" generally indicates that the context-dependent object is an "or" relationship. "comprising A, B and C", "comprising A, B, C" means that all three of A, B, C comprise, "comprising A, B or C" means that one of the three comprises A, B, C, and "comprising A, B and/or C" means that any 1 or any 2 or 3 of the three comprises A, B, C.
It should be understood that in this disclosure, "B corresponding to a", "a corresponding to B", or "B corresponding to a" means that B is associated with a from which B may be determined. Determining B from a does not mean determining B from a alone, but may also determine B from a and/or other information. The matching of A and B is that the similarity of A and B is larger than or equal to a preset threshold value.
As used herein, "if" may be interpreted as "at … …" or "at … …" or "in response to a determination" or "in response to detection" depending on the context.
The technical scheme of the present disclosure is described in detail below with specific examples. The following embodiments may be combined with each other, and some embodiments may not be repeated for the same or similar concepts or processes.
Fig. 1 is a flow chart of a spacecraft telemetry data health prediction method according to an embodiment of the disclosure, as shown in fig. 1, the method includes:
s101, based on the acquired telemetry data prediction requirement, reading historical telemetry data matched with the telemetry data prediction requirement from a telemetry database;
illustratively, telemetry data prediction requirements include a number limit for the predicted outcome, a time format specification, an output form. The number limit includes the number of prediction points, such as: predicting 1000 points; the time format includes parameter intervals such as: a bit every 2 seconds; the output includes visual output and data output, such as: an array of data only and corresponding time or a graph containing input data and predicted data trends.
Telemetry parameters: the telemetry data input by the algorithm comprises spacecraft names, part names, spacecraft codes and time sequence data corresponding to time.
Specifically, the message can be received through the kafka message middle key, and the message comprises information such as a spacecraft name, a spacecraft code, a part code, a starting time, an ending time, a predicted parameter number, a predicted time interval and the like. After the algorithm receives the message, the corresponding information is matched, the corresponding telemetry parameters are read from the database, and after telemetry data are obtained, the values and time are converted into a data frame format array required by the algorithm.
S102, after preprocessing the historical telemetry data, inputting a pre-trained data health prediction model, outputting predicted telemetry data of a future time period corresponding to the telemetry data prediction requirement by the pre-trained data health prediction model, and matching the predicted telemetry data with the telemetry data prediction requirement;
illustratively, preprocessing the historical telemetry data may include screening, smoothing, and filling the data, which is still algorithmically implemented. Because of the complex data volume, the data needs to be preprocessed after the most original spacecraft telemetry parameters are obtained. The field removing comprises removing some very obvious abnormal points in the data, wherein the data can influence the learning of data trend in the process of training a model; smoothing comprises the step that after data is received, some obvious noise points exist, so that fluctuation and change of the data are overlarge, and the data can influence learning of data trend; filling includes that some points have missing values after data is received, and filling is needed through an algorithm, otherwise, a prediction result is affected.
And (3) performing preprocessing of outlier rejection, noise smoothing and missing value filling on the data by utilizing a data preprocessing algorithm flow, and storing the processed data as a data frame array corresponding to time.
The data health prediction model in the embodiment of the application can be constructed based on a combination of an LSTM (Long Short-Term Memory) model and a CNN (convolutional neural network, convolutional Neural Networks) model, and is correspondingly improved on the basis of the LSTM and the CNN, so that the predicted telemetry data of a future time period corresponding to the telemetry data prediction requirement is output.
Optionally, in the data health prediction model of the present application, based on consideration of computational complexity, only one convolutional layer is set in the CNN layer, the filter, kernel size and activation function are defined by a user, default are 128, 3 and 'relu', the pooling layer is selected as the largest pooling layer, the last layer is the LSTM layer, the unit is 32, and the activation function is sigmoid. During training, the data are converted into matrix data, the CNN is utilized to extract matrix data characteristics, data trend is learned, and a model is built according to the characteristics.
S103, if the predicted telemetry data are matched, detecting whether abnormal data exist in the predicted telemetry data through the data health prediction model, and if the abnormal data do not exist, displaying the predicted telemetry data and the historical telemetry data in a visual mode at the same time; if the abnormal data exists, eliminating the abnormal data, and displaying the predicted telemetry data and the historical telemetry data after eliminating the abnormal data in a visual mode at the same time;
Illustratively, if the predicted telemetry data of the future time period output by the data health prediction model matches the predicted demand of telemetry data, i.e., in terms of data format, data accuracy, etc., for example, the raw data is 2 seconds, then the predicted result is also 2 seconds; the user can also customize the predicted time format, for example, the original data is 2 seconds, the user needs 4 seconds as the predicted result, and the time format required by the user can be matched through a formula.
Further, satellites are an important device for human detection of the universe and are also the main carrier of communication tools. The satellite operates in a severe space environment, so that once serious faults occur, the satellite is difficult to repair, and safe and reliable operation of the satellite can be ensured through timely and effective anomaly detection and fault location.
The data imbalance is a common problem in anomaly detection tasks, and the satellite telemetry data received by the ground station is mostly normal data, and the cost of misjudging one abnormal data as normal by the system is different from misjudging one normal data as abnormal. The abnormal detection rather misjudges the normal data as abnormal, and can not miss an abnormal data.
In an alternative embodiment of the present invention,
the detecting, by the data health prediction model, whether abnormal data exists in the predicted telemetry data includes:
based on a convolution pooling layer of the data health prediction model, obtaining local sequence visual field characteristics of the predicted telemetry data;
and outputting positive sample characteristics and negative sample characteristics in the local sequence visual field characteristics through a trained positive and negative sample classifier of the data health prediction model according to the local sequence visual field characteristics, wherein the positive sample characteristics are used for indicating characteristics corresponding to normal telemetry data in the local sequence visual field characteristics, and the negative sample characteristics are used for indicating characteristics corresponding to abnormal telemetry data in the local sequence visual field characteristics.
Optionally, the local sequence view feature of the predicted telemetry data obtained based on the convolution pooling layer of the data health prediction model may be represented by the following formula:
；/>
；
wherein,,C(w)a one-dimensional convolution result is represented,wthe width of the time series sequence is indicated,Kthe matrix of kernel functions is represented,srepresenting the coordinates of the convolution kernel on the width axis,Irepresenting an array of time series of numbers,krepresenting the size of the array of convolution kernels, GThe characteristic sequences obtained after pooling are represented,prepresenting the field of view size of the pooled region,Fan array of feature sequences representing the input is presented,hrepresenting the size of the array of feature sequences.
The size of the convolution kernel should be determined by the length of the time series and the time scale of the abnormal behavior. Generally, when the ratio of the time series length to the abnormal behavior time is high, a convolution kernel with a wider field of view should be selected, so that extraction of a large number of useless sparse features can be avoided. On the contrary, when the time series is short, the convolution kernel with a narrow field of view is selected as much as possible so as not to extract the edge features.
It should be noted that, in the embodiment of the present application, the negative sample represents telemetry data during normal operation of the satellite, and the positive sample represents telemetry data when an anomaly occurs in the satellite. In response, the positive sample feature is used to indicate a feature corresponding to normal telemetry data in the local sequence field of view feature, and the negative sample feature is used to indicate a feature corresponding to abnormal telemetry data in the local sequence field of view feature. The positive and negative sample classifier can identify whether the data input into the data health prediction model is abnormal or not.
In an alternative embodiment of the present invention,
Before the positive sample feature and the negative sample feature in the local sequence visual field feature are output by the trained positive and negative sample classifier of the data health prediction model, the method further comprises training the positive and negative sample classifier:
mapping a first training subset and a second training subset in a training data set to the same distribution space through a mapping function based on a pre-acquired training data set, and training positive and negative sample classifiers to be trained in the distribution space;
and setting a classification loss function based on the first training subset and the second training subset, and dynamically updating the classification cost weight and the classification optimization target according to the classification loss function until the classification accuracy of the positive and negative sample classifier reaches a preset threshold.
Wherein the first training subset is used for indicating the data of the training data set with labels, the second training subset is used for indicating the data of the training data set without labels,
the first training subset and the second training subset are mapped to the same distribution space through the mapping function, in this way, the feature extraction and selection capacity of the positive and negative sample classifier is trained on the first training subset by utilizing satellite A data, then part of the network layer of the data health prediction model is migrated to the second training subset satellite B for training, and the abnormal detection effect of the second training subset is ensured by fine tuning of the unlabeled data of the second training subset.
In an alternative embodiment of the present invention,
and dynamically updating the classification cost weight and the classification optimization target according to the classification loss function by the following formula:
wherein,,Q(r)the weight of the classification cost is represented as,Ethe cost matrix is represented by a representation of the cost matrix,t n represent the firstnThe desired output of the plurality of outputs,p n represent the firstnA plurality of prediction outputs;
wherein,,H(r)the classification optimization objective is represented as such,M、Kthe number of positive samples and the number of negative samples are represented respectively,l pos 、l neg respectively represent the cross entropy loss value corresponding to the positive sample and the cross entropy loss value corresponding to the negative sample,
、/>
respectively represent the firstiMisclassification cost weight corresponding to positive samplejMisclassification cost weight corresponding to each negative sample, < ->
Representation oft+1The misclassification cost weight corresponding to the current sample at the moment,U over represents the ratio of the number of positive samples to the number of negative samples, +.>
Geometric mean representing the output result of the current sample, +.>
Indicating the accuracy of the output result of the current sample.
According to the method, the unbalanced ratio is used as penalty of cost-sensitive misclassification, the problem that a model caused by data unbalance on classification tasks is biased to normal sample fitting can be integrally solved, in addition, a small-batch training mode is adopted in the neural network training process, a fixed cost matrix cannot be well adapted to unbalance of local area distribution, and the dynamic-change misclassification cost weight is utilized for self-adaptive updating, so that not only can the integral sample unbalance be considered, but also the unbalance on the local small-batch training can be considered.
And S104, if the data deviation of the predicted telemetry data and the actual telemetry data is not matched, adding Gaussian white noise into the actual telemetry data, optimizing a data health prediction model based on the data deviation and the actual telemetry data added with the Gaussian white noise until the predicted telemetry data output by the optimized data health prediction model is matched with the telemetry data prediction requirement.
If the data are not matched, the current data health prediction model is not up to the precision requirement, and the data can be further optimized. The spacecraft telemetry parameters are accurate to 2 bits after the decimal point and have no large obvious change, and the Gaussian white noise is added to optimize the data without obvious fluctuation after the decimal point; and calculating the mean square error of the original telemetry data, wherein the mean value and the standard deviation of the Gaussian white noise are related to the mean square error of the original telemetry data, and the prediction result can be optimized by adding the proper Gaussian white noise into the mean square error of the original telemetry data.
In an alternative embodiment of the present invention,
the method for optimizing the data health prediction model based on the data deviation and the actual telemetry data added with Gaussian white noise comprises the following steps:
dividing actual telemetry data added with Gaussian white noise into a first data sequence and a second data sequence according to a preset proportion, and respectively extracting a first characteristic vector set corresponding to the first data sequence and a second characteristic vector set corresponding to the second data sequence according to the weight of a convolution kernel of the convolution layer, an index value of a filter and a first activation function corresponding to the convolution layer through the convolution layer of a data health prediction model to be optimized;
Illustratively, the preset ratio may include 5:5, or 2:8, the specific setting of the preset ratio in the embodiment of the present application is not limited. Wherein the first data sequence may be expressed as y1= [ sn ] 1 ,sn 2 ,…,sn n ]The second data sequence may be represented as y2= [ sm 1 ,sm 2 ,…,sm n ]，
According to the weight of the convolution kernel of the convolution layer, the index value of the filter and the first activation function corresponding to the convolution layer, the first feature vector set corresponding to the first data sequence and the second feature vector set corresponding to the second data sequence are extracted respectively, where the first feature vector set and the second feature vector set corresponding to the second data sequence are represented by the following formulas:
wherein,,
、/>
representing a first set of feature vectors and a second set of feature vectors, respectively, < >>
、/>
Representing a first bias parameter and a second bias parameter, respectively,/->
The activation function is represented as a function of the activation,NUM1、NUM2respectively representing the data quantity of the first data sequence and the data quantity of the second data sequence, +.>
、/>
Respectively representing a first convolution kernel weight corresponding to a first data sequence and a second convolution kernel weight corresponding to a second data sequence,/for>
、/>
Respectively representing the index value of the first filter corresponding to the first data sequence and the index value of the second filter corresponding to the second data sequence.
Respectively distributing a first weight matrix for the first feature vector set and a second weight matrix for the second feature vector set, and integrating the first feature vector set and the second feature vector set into a comprehensive feature vector set;
The first feature vector set and the second feature vector set may be assigned a first weight matrix and a second weight matrix by a full-connection layer of the data health prediction model, specifically, the first feature vector set and the second feature vector set may be flattened into one-dimensional vectors, then one weight matrix is used for linear transformation, finally one activation function is used for nonlinear transformation of the result, and vector stitching is performed on the first feature vector set and the second feature vector set, and the first weight matrix and the second weight matrix are combined to form a comprehensive feature vector set.
Updating information of the comprehensive feature vector set through a hidden layer of the data health prediction model to be optimized, inputting the updated comprehensive feature vector into an output layer of the data health prediction model to be optimized, and determining a prediction output result;
based on the data deviation of the prediction output result and the actual telemetry data added with Gaussian white noise, a loss function is established, minimized training of the data deviation is carried out through a self-adaptive learning rate optimization algorithm, a target weight parameter is determined, and the weight parameter of the data health prediction model to be optimized is updated according to the target weight parameter.
The LSTM cyclic neural network comprises a module with a memory unit, can learn the characteristics of the data in the time domain, has good processing performance on time series data, and is widely applied in a plurality of fields. The memory module of the LSTM cyclic network comprises three units: the input gate, the forgetting gate and the output gate respectively control the input, the update and the output of the information, so that the network has a certain memory function, but simultaneously, the network also has more learning parameters, the LSTM is simplified, a variant LSTM circulating neural network is provided, the forgetting gate and the input gate in the traditional LSTM circulating neural network are combined into an update gate, the update gate uses a Sigmoid layer to update the information, the memory unit keeps the information of how much proportion at the time t-1, and the information of how much proportion is supplemented at the time t, thereby ensuringt+1The output of time is fairly affected by the t-1 and t-time states so that the memory cell remains in an active state all the time.
In an alternative embodiment of the present invention,
the hidden layer comprises a memory unit and a cell unit, wherein the memory unit is used for selectively memorizing the content of each time step, and the cell unit is used for deciding whether to memorize the content of the previous moment;
And updating information of the comprehensive feature vector set through a hidden layer of the data health prediction model to be optimized, inputting the updated comprehensive feature vector into an output layer of the data health prediction model to be optimized, and determining a prediction output result comprises the following steps:
determining a first output of the memory unit according to the comprehensive feature vector set, a first weight vector corresponding to the memory unit, a first bias parameter and a second activation function corresponding to the memory unit;
determining a second output of the cell unit according to the comprehensive feature vector set, a second weight vector corresponding to the cell unit, a second bias parameter and a cell state at a moment on the cell unit;
and inputting the first output and the second output into the output layer, and determining a predicted output result according to a third weight vector and a third bias parameter corresponding to the output layer.
For example, the method for determining the predicted output result in the present application may refer to a method for performing predicted output on input data by using an existing LSTM recurrent neural network, which is not described herein in detail.
In an alternative embodiment of the present invention,
The loss function is established, the data deviation minimization training is carried out through the self-adaptive learning rate optimization algorithm, and the target weight parameter determination comprises the following steps:
the loss function is shown in the following formula:
；
wherein,,LOSSthe loss value is indicated as such,Nthe number of iterations is indicated and,y t (i)represent the firstiThe index of the next iteration predicts the output value,a t (i)represent the firstiThe input of the iterative memory unit;
the minimizing training of the data deviation by the self-adaptive learning rate optimizing algorithm is shown in the following formula:
wherein,,P t-1 the data deviation is indicated as such,tthe time of day is indicated as such,Tthe total duration of time is indicated and,u o,t 、u f,t respectively representtUpdating the error term of the gate at the moment and outputting the error term of the gate,W o,t 、W f,t respectively representtThe weight of the gate is updated at the moment and the weight of the gate is output.
Fig. 2 is a logic schematic diagram of a spacecraft telemetry data health prediction method according to an embodiment of the disclosure, and as shown in fig. 2, the spacecraft telemetry data health prediction method of the disclosure may include:
s01, obtaining a user telemetry requirement;
s02, reading telemetry parameters of the stored spacecraft of the database;
s03, preprocessing the read data;
s04, training the processed data;
s05, establishing an algorithm model;
s06, verifying whether the model achieves the expected accuracy
S07, if not, adjusting model parameters and jumping to S05;
S08, saving the trained model;
s09, reading the latest spacecraft telemetry parameters;
s10, calling a stored model to predict;
s11, adding a Gaussian white noise optimization result;
s12, matching the data with the user demand time;
s13, outputting the required array and drawing.
It should be noted that, the beneficial effects corresponding to the scheme of fig. 2 in the embodiment of the present disclosure may refer to the beneficial effects corresponding to the scheme of fig. 1, and the embodiment of the present disclosure is not described herein again.
In a second aspect of the embodiments of the present disclosure,
provided is a spacecraft telemetry data health prediction system, fig. 3 is a schematic structural diagram of the spacecraft telemetry data health prediction system according to an embodiment of the disclosure, and as shown in fig. 3, the system includes:
a first unit for reading historical telemetry data matching the telemetry data prediction requirements from a telemetry database based on the acquired telemetry data prediction requirements;
the second unit is used for preprocessing the historical telemetry data, inputting a pre-trained data health prediction model, outputting predicted telemetry data of a future time period corresponding to the telemetry data prediction requirement by the pre-trained data health prediction model, and matching the predicted telemetry data with the telemetry data prediction requirement;
The third unit is used for detecting whether abnormal data exist in the predicted telemetry data through the data health prediction model if the predicted telemetry data are matched with the historical telemetry data, and displaying the predicted telemetry data and the historical telemetry data in a visual mode at the same time if the abnormal data do not exist; if the abnormal data exists, eliminating the abnormal data, and displaying the predicted telemetry data and the historical telemetry data after eliminating the abnormal data in a visual mode at the same time;
and a fourth unit, configured to determine a data deviation between the predicted telemetry data and the actual telemetry data if the predicted telemetry data and the actual telemetry data are not matched, and add white gaussian noise to the actual telemetry data, and optimize a data health prediction model based on the data deviation and the actual telemetry data added with white gaussian noise until the predicted telemetry data output by the optimized data health prediction model matches the telemetry data prediction requirement.
In a third aspect of the embodiments of the present disclosure,
there is provided an electronic device including:
a processor;
a memory for storing processor-executable instructions;
wherein the processor is configured to invoke the instructions stored in the memory to perform the method described previously.
In a fourth aspect of embodiments of the present disclosure,
There is provided a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the method as described above.
The present invention may be a method, apparatus, system, and/or computer program product. The computer program product may include a computer readable storage medium having computer readable program instructions embodied thereon for performing various aspects of the present invention.
It will be appreciated by persons skilled in the art that the embodiments of the invention described above and shown in the drawings are by way of example only and are not limiting. The objects of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been shown and described in the examples and embodiments of the invention may be modified or practiced without departing from the principles described.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present disclosure, and not for limiting the same; although the present disclosure has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present disclosure.
Claims (10)
1. A method for health prediction of telemetry data of a spacecraft, comprising:
reading historical telemetry data matching the telemetry data predicted demand from a telemetry database based on the acquired telemetry data predicted demand;
after preprocessing the historical telemetry data, inputting a pre-trained data health prediction model, outputting predicted telemetry data of a future time period corresponding to the telemetry data prediction requirement by the pre-trained data health prediction model, and matching the predicted telemetry data with the telemetry data prediction requirement;
if the predicted telemetry data are matched with the historical telemetry data, detecting whether the predicted telemetry data are abnormal data or not through the data health prediction model, and if the predicted telemetry data are not abnormal data, displaying the predicted telemetry data and the historical telemetry data in a visual mode at the same time; if the abnormal data exists, eliminating the abnormal data, and displaying the predicted telemetry data and the historical telemetry data after eliminating the abnormal data in a visual mode at the same time;
if the data deviation is not matched, determining the data deviation of the predicted telemetry data and the actual telemetry data, adding Gaussian white noise into the actual telemetry data, and optimizing a data health prediction model based on the data deviation and the actual telemetry data added with the Gaussian white noise until the predicted telemetry data output by the optimized data health prediction model is matched with the telemetry data prediction requirement.
2. The method of claim 1, wherein the method of optimizing a data health prediction model based on the data bias and actual telemetry data incorporating gaussian white noise comprises:
dividing actual telemetry data added with Gaussian white noise into a first data sequence and a second data sequence according to a preset proportion, and respectively extracting a first characteristic vector set corresponding to the first data sequence and a second characteristic vector set corresponding to the second data sequence according to the weight of a convolution kernel of the convolution layer, an index value of a filter and a first activation function corresponding to the convolution layer through the convolution layer of a data health prediction model to be optimized;
respectively distributing a first weight matrix for the first feature vector set and a second weight matrix for the second feature vector set, and integrating the first feature vector set and the second feature vector set into a comprehensive feature vector set;
updating information of the comprehensive feature vector set through a hidden layer of the data health prediction model to be optimized, inputting the updated comprehensive feature vector into an output layer of the data health prediction model to be optimized, and determining a prediction output result;
Based on the data deviation of the prediction output result and the actual telemetry data added with Gaussian white noise, a loss function is established, minimized training of the data deviation is carried out through a self-adaptive learning rate optimization algorithm, a target weight parameter is determined, and the weight parameter of the data health prediction model to be optimized is updated according to the target weight parameter.
3. The method of claim 2, wherein the hidden layer comprises a memory unit for selectively memorizing the contents of each time step and a cell unit for deciding whether to memorize the contents of the previous time;
and updating information of the comprehensive feature vector set through a hidden layer of the data health prediction model to be optimized, inputting the updated comprehensive feature vector into an output layer of the data health prediction model to be optimized, and determining a prediction output result comprises the following steps:
determining a first output of the memory unit according to the comprehensive feature vector set, a first weight vector corresponding to the memory unit, a first bias parameter and a second activation function corresponding to the memory unit;
determining a second output of the cell unit according to the comprehensive feature vector set, a second weight vector corresponding to the cell unit, a second bias parameter and a cell state at a moment on the cell unit;
And inputting the first output and the second output into the output layer, and determining a predicted output result according to a third weight vector and a third bias parameter corresponding to the output layer.
4. The method of claim 2, wherein the establishing a loss function, performing data bias minimization training by an adaptive learning rate optimization algorithm, and determining the target weight parameter comprises:
the loss function is shown in the following formula:
the method comprises the steps of carrying out a first treatment on the surface of the Wherein,,LOSSthe loss value is indicated as such,Nthe number of iterations is indicated and,y t (i)represent the firstiThe index of the next iteration predicts the output value,a t (i)represent the firstiThe input of the iterative memory unit;
minimizing training of data bias by adaptive learning rate optimization algorithm is disclosed as followsThe formula is shown as follows:
the method comprises the steps of carrying out a first treatment on the surface of the Wherein,,P t-1 the data deviation is indicated as such,tthe time of day is indicated as such,Tthe total duration of time is indicated and,u o,t 、u f,t respectively representtUpdating the error term of the gate at the moment and outputting the error term of the gate,W o,t 、W f,t respectively representtThe weight of the gate is updated at the moment and the weight of the gate is output.
5. The method of claim 1, wherein the detecting, by the data health prediction model, whether anomalous data is present in the predicted telemetry data comprises:
based on a convolution pooling layer of the data health prediction model, obtaining local sequence visual field characteristics of the predicted telemetry data;
And outputting positive sample characteristics and negative sample characteristics in the local sequence visual field characteristics according to the local sequence visual field characteristics and positive and negative sample classifiers trained by the data health prediction model, wherein the positive sample characteristics are used for indicating characteristics corresponding to normal telemetry data in the local sequence visual field characteristics, and the negative sample characteristics are used for indicating characteristics corresponding to abnormal telemetry data in the local sequence visual field characteristics.
6. The method of claim 5, wherein before outputting positive and negative sample features in the local sequence view features by the trained positive and negative sample classifiers of the data health prediction model, the method further comprises training the positive and negative sample classifiers:
mapping a first training subset and a second training subset in the training data set to the same distribution space through a mapping function based on a pre-acquired training data set, and training positive and negative sample classifiers to be trained in the distribution space, wherein the first training subset is used for indicating data with labels in the training data set, and the second training subset is used for indicating data without labels in the training data set;
And setting a classification loss function based on the first training subset and the second training subset, and dynamically updating the classification cost weight and the classification optimization target according to the classification loss function until the classification accuracy of the positive and negative sample classifier reaches a preset threshold.
7. The method of claim 6, wherein the classification cost weights and classification optimization objectives are dynamically updated according to the classification loss function according to the following formula:
the method comprises the steps of carrying out a first treatment on the surface of the Wherein,,Q (r)the weight of the classification cost is represented as,Ethe cost matrix is represented by a representation of the cost matrix,t n represent the firstnThe desired output of the plurality of outputs,p n represent the firstnA plurality of prediction outputs;
；/>
the method comprises the steps of carrying out a first treatment on the surface of the Wherein,,H(r)the classification optimization objective is represented as such,M、Kthe number of positive samples and the number of negative samples are represented respectively,l pos 、l neg respectively representing the cross entropy loss value corresponding to the positive sample and the cross entropy loss value corresponding to the negative sample,/->
、/>
Respectively represent the firstiMisclassification cost weight corresponding to positive samplejMisclassification cost weight corresponding to each negative sample, < ->
Representation oft+1The misclassification cost weight corresponding to the current sample at the moment,U over represents the ratio of the number of positive samples to the number of negative samples, +.>
Geometric mean representing the output result of the current sample, +.>
Indicating the accuracy of the output result of the current sample.
8. A spacecraft telemetry data health prediction system, comprising:
a first unit for reading historical telemetry data matching the telemetry data prediction requirements from a telemetry database based on the acquired telemetry data prediction requirements;
the second unit is used for preprocessing the historical telemetry data, inputting a pre-trained data health prediction model, outputting predicted telemetry data of a future time period corresponding to the telemetry data prediction requirement by the pre-trained data health prediction model, and matching the predicted telemetry data with the telemetry data prediction requirement;
the third unit is used for detecting whether abnormal data exist in the predicted telemetry data through the data health prediction model if the predicted telemetry data are matched with the historical telemetry data, and displaying the predicted telemetry data and the historical telemetry data in a visual mode at the same time if the abnormal data do not exist; if the abnormal data exists, eliminating the abnormal data, and displaying the predicted telemetry data and the historical telemetry data after eliminating the abnormal data in a visual mode at the same time;
and a fourth unit, configured to determine a data deviation between the predicted telemetry data and the actual telemetry data if the predicted telemetry data and the actual telemetry data are not matched, and add white gaussian noise to the actual telemetry data, and optimize a data health prediction model based on the data deviation and the actual telemetry data added with white gaussian noise until the predicted telemetry data output by the optimized data health prediction model matches the telemetry data prediction requirement.
9. An electronic device, comprising:
a processor;
a memory for storing processor-executable instructions;
wherein the processor is configured to invoke the instructions stored in the memory to perform the method of any of claims 1 to 7.
10. A computer readable storage medium having stored thereon computer program instructions, which when executed by a processor, implement the method of any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310524696.7A CN116304884B (en) | 2023-05-11 | 2023-05-11 | Spacecraft telemetry data health prediction method, system, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310524696.7A CN116304884B (en) | 2023-05-11 | 2023-05-11 | Spacecraft telemetry data health prediction method, system, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116304884A true CN116304884A (en) | 2023-06-23 |
| CN116304884B CN116304884B (en) | 2023-07-28 |
Family
ID=86826115
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310524696.7A Active CN116304884B (en) | 2023-05-11 | 2023-05-11 | Spacecraft telemetry data health prediction method, system, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116304884B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116702083A (en) * | 2023-08-10 | 2023-09-05 | 武汉能钠智能装备技术股份有限公司四川省成都市分公司 | Satellite telemetry data anomaly detection method and system |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190191230A1 (en) * | 2012-06-26 | 2019-06-20 | BTS Software Solutions, LLC | Realtime telemetry data compression system |
| CN112101636A (en) * | 2020-08-26 | 2020-12-18 | 北京航空航天大学 | GRU and GARCH-based satellite long-period variance-variance degradation prediction evaluation method |
| CN112631240A (en) * | 2020-12-07 | 2021-04-09 | 中国人民解放军63920部队 | Spacecraft fault active detection method and device |
| CN115185937A (en) * | 2022-07-12 | 2022-10-14 | 中国人民解放军海军工程大学 | SA-GAN architecture-based time sequence anomaly detection method |
| CN115510950A (en) * | 2022-08-26 | 2022-12-23 | 北京机电工程研究所 | Aircraft telemetry data anomaly detection method and system based on time convolution network |
| CN115841149A (en) * | 2022-10-11 | 2023-03-24 | 中国电子科技集团公司第五十四研究所 | Confidence estimation-based satellite fault prediction model training and fault prediction method |
| WO2023056614A1 (en) * | 2021-10-09 | 2023-04-13 | 大连理工大学 | Method for predicting rotating stall of axial flow compressor on the basis of stacked long short-term memory network |
-
2023
- 2023-05-11 CN CN202310524696.7A patent/CN116304884B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190191230A1 (en) * | 2012-06-26 | 2019-06-20 | BTS Software Solutions, LLC | Realtime telemetry data compression system |
| CN112101636A (en) * | 2020-08-26 | 2020-12-18 | 北京航空航天大学 | GRU and GARCH-based satellite long-period variance-variance degradation prediction evaluation method |
| CN112631240A (en) * | 2020-12-07 | 2021-04-09 | 中国人民解放军63920部队 | Spacecraft fault active detection method and device |
| WO2023056614A1 (en) * | 2021-10-09 | 2023-04-13 | 大连理工大学 | Method for predicting rotating stall of axial flow compressor on the basis of stacked long short-term memory network |
| CN115185937A (en) * | 2022-07-12 | 2022-10-14 | 中国人民解放军海军工程大学 | SA-GAN architecture-based time sequence anomaly detection method |
| CN115510950A (en) * | 2022-08-26 | 2022-12-23 | 北京机电工程研究所 | Aircraft telemetry data anomaly detection method and system based on time convolution network |
| CN115841149A (en) * | 2022-10-11 | 2023-03-24 | 中国电子科技集团公司第五十四研究所 | Confidence estimation-based satellite fault prediction model training and fault prediction method |
Non-Patent Citations (3)
| Title |
|---|
| JINGYUE PANG 等: "Anomaly Detection for Satellite Telemetry Series with Prediction Interval Optimization", 《 2018 INTERNATIONAL CONFERENCE ON SENSING, DIAGNOSTICS, PROGNOSTICS, AND CONTROL (SDPC)》, pages 408 - 414 * |
| 宁顺成: "基于遥测时序数据的航天器异常检测技术应用研究", 《中国优秀硕士学位论文全文数据库》, pages 031 - 1830 * |
| 朱昶文;党建成;周军;: "基于健康度分析的卫星故障预测模型", 计算机应用, no. 2, pages 39 - 45 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116702083A (en) * | 2023-08-10 | 2023-09-05 | 武汉能钠智能装备技术股份有限公司四川省成都市分公司 | Satellite telemetry data anomaly detection method and system |
| CN116702083B (en) * | 2023-08-10 | 2023-12-26 | 武汉能钠智能装备技术股份有限公司四川省成都市分公司 | Satellite telemetry data anomaly detection method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116304884B (en) | 2023-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109978228B (en) | PM2.5 concentration prediction method, device and medium | |
| CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
| CN112733997B (en) | Hydrological time series prediction optimization method based on WOA-LSTM-MC | |
| CN116304884B (en) | Spacecraft telemetry data health prediction method, system, equipment and storage medium | |
| CN116596095B (en) | Training method and device of carbon emission prediction model based on machine learning | |
| CN111027686A (en) | Landslide displacement prediction method, device and equipment | |
| CN115510963A (en) | Incremental equipment fault diagnosis method | |
| CN115587666A (en) | Load prediction method and system based on seasonal trend decomposition and hybrid neural network | |
| CN116303786B (en) | Block chain financial big data management system based on multidimensional data fusion algorithm | |
| CN115098789B (en) | Multidimensional interest fusion recommendation method and device based on neural network and related equipment | |
| CN112819256A (en) | Convolution time sequence room price prediction method based on attention mechanism | |
| CN113435900A (en) | Transaction risk determination method and device and server | |
| CN115423594A (en) | Enterprise financial risk assessment method, device, equipment and storage medium | |
| CN115934490A (en) | Server performance prediction model training method, device, equipment and storage medium | |
| CN116542701A (en) | Carbon price prediction method and system based on CNN-LSTM combination model | |
| CN113449919B (en) | Power consumption prediction method and system based on feature and trend perception | |
| CN111461329B (en) | Model training method, device, equipment and readable storage medium | |
| CN111402028A (en) | Information processing method, device and equipment | |
| CN115794805B (en) | Method for supplementing measurement data of medium-low voltage distribution network | |
| CN111507461A (en) | Interpretability information determining method and device | |
| CN108764489B (en) | Model training method and device based on virtual sample | |
| CN110717577A (en) | Time series prediction model construction method for noting regional information similarity | |
| CN115829123A (en) | Natural gas demand prediction method and device based on grey model and neural network | |
| CN111179070A (en) | Loan risk timeliness prediction system and method based on LSTM | |
| CN112732777A (en) | Position prediction method, apparatus, device and medium based on time series |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |