CN109361447B - Telemetry elastic transport method and device based on machine learning - Google Patents
Telemetry elastic transport method and device based on machine learning Download PDFInfo
- Publication number
- CN109361447B CN109361447B CN201811052935.9A CN201811052935A CN109361447B CN 109361447 B CN109361447 B CN 109361447B CN 201811052935 A CN201811052935 A CN 201811052935A CN 109361447 B CN109361447 B CN 109361447B
- Authority
- CN
- China
- Prior art keywords
- telemetry
- network
- parameter
- data
- information
- 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.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/14—Relay systems
- H04B7/15—Active relay systems
- H04B7/185—Space-based or airborne stations; Stations for satellite systems
- H04B7/18502—Airborne stations
- H04B7/18506—Communications with or from aircraft, i.e. aeronautical mobile service
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Abstract
The telemetry elastic transport method and device based on machine learning that the invention discloses a kind of, method includes: that data acquisition learns with training, transmitting terminal acquires telemetry in real time and is sent to receiving end, transmitting terminal neural network is trained simultaneously, until the deviation of training result and telemetry is lower than preset threshold；Parameter transmitting and digital simulation, the neural network parameter for completing training objective is passed to receiving end by transmitting terminal, and receiving end constructs the neural network of isomorphism according to neural network parameter and allows to predict that instruction information generates telemetering analogue data using what transmitting terminal was sent；Prediction ratio sentences and unruly-value rejecting, and transmitting terminal acquires telemetry and simultaneously carries out with neuron network simulation output result than sentencing, and combines and fly to control prior information generation instruction information to be used for subsequent processing.The transmission capacity of telemetry, the reliability and flexibility of the end-to-end information transmission of room for promotion Radio Link can be greatly reduced with the time-varying characteristics of dynamically adapting telemetry parameter comentropy in this method.
Description
Technical field
The present invention relates to Space-based information transmission technical field, in particular to a kind of telemetry elasticity based on machine learning
Transmission method and device.
Background technique
Deep space ultra-distance communication is faced with the technologies such as transmission power is limited, transmission range is remote, reception weak output signal and chooses
War, reducing deep space channel transmission data amount is deep space communication important technology difficult point to be solved.Currently, generalling use both at home and abroad
Compression method reduces the data volume of scientific application data transmission, especially optical detection data；For telemetry, have
Document proposes the method for real-time online compression, compression ratio about 40%, but this method only can be suitably used for having the distant of regular length
Measured data frame and anti-channel error code performance is insufficient, flexibility is needed to be further improved with reliability.
Machine learning techniques as developing fast one research field in recent years, caused domestic and international space flight mechanism with
The extensive concern of researcher, and satellitosis monitoring, telemetry parameter prediction, satellite failure diagnosis, intelligent independent control, appoint
Business data processing etc. achieves certain research achievement:
1. thering is document to generate the real-time surveillance satellite telemetering of health status knowledge base by cluster in terms of satellitosis monitoring
State has document comprehensively to construct the rule of Orbital detection by the algorithm that excavation lies in telemetry behind
There is embedded on-line condition monitoring modular structure of the document based on system on chip in library, and there are also documents to pass through general mathematical expression side
Formula carries out satellitosis expression, the general in-orbit state analysis of design and early warning system, effectively obtains the work of satellite in orbit in real time
State.
2. having document by returning the search capability of particle swarm optimization algorithm and supporting vector in terms of telemetry parameter prediction
Return the Nonlinear Mapping performance of machine to combine to obtain the prediction of postorder telemetry parameter, has document by using in shallow-layer learning model
Support vector regression and deep learning model depth neural network in shot and long term memory network to measured data carry out trend
Prediction has document to improve spacecraft telemetry predetermined speed by modified probabilistic neural network, and there are also documents to pass through synthesis
Using the segmentation of long-term telemetry, time sequence model cluster and classification, the mode modeling based on multisequencing and based on random
The Evolution Modes process modeling approach of process preferably solves the long-term telemetry modeling problem of satellite.
3. thering is document to be based on BP neural network in terms of satellite failure diagnosis and carrying out offline autonomous learning and real-time online event
Barrier diagnosis carries out real-time online diagnosis to telemetry, has document to defend by a kind of neural network method realization of particle group optimizing
The failure predication of star has document to close using the mapping that particle filter method is established between characteristic parameter and residual error and fault mode
System, and fault diagnosis model is established using BP neural network, there are also documents to use particle swarm algorithm Optimization of Wavelet neural network
Method and ant group algorithm optimization radial base neural net method to satellite telemetering data carry out fault diagnosis modeling.
4. having document whether judging that telemetry is abnormal by neural network to improve data in terms of intelligent independent control
Speed and quick response processing capacity are analyzed, fuzzy neural network is based on there are also document and proposes a kind of intelligent dynamic task scheduling
Method.
5. having document by the deep neural network of one multitask of building and to network mould in terms of task data processing
Type is trained to rebuild high-resolution colour picture, and there are also documents by a kind of data intelligence blending algorithm, and to reduce precision low
Data source in influence of the error to fusion results.
From the analysis above, we can see that correlative study of the machine learning techniques in terms of space industry telemetering and application are not yet related to sky
Between technical field of information transmission.
Summary of the invention
The present invention is directed to solve at least some of the technical problems in related technologies.
For this purpose, an object of the present invention is to provide a kind of telemetry elastic transport method based on machine learning,
The transmission capacity of telemetry can be greatly reduced in this method, and what the further end-to-end information of room for promotion Radio Link was transmitted can
By property and flexibility.
It is another object of the present invention to propose a kind of telemetry flexible delivery device based on machine learning.
In order to achieve the above objectives, one aspect of the present invention embodiment proposes a kind of telemetry elasticity based on machine learning
Transmission method, comprising the following steps: data acquisition learns with training, and transmitting terminal acquires telemetry in real time and is sent to reception
End, while transmitting terminal neural network is trained, until the deviation of training result and the telemetry is lower than preset threshold；
The neural network parameter for completing training objective is passed to the receiving end by parameter transmitting and digital simulation, the transmitting terminal, and
And the permission that the receiving end is constructed the neural network of isomorphism and sent using the transmitting terminal according to the neural network parameter
Prediction instruction information generates telemetering analogue data；Prediction ratio sentences and unruly-value rejecting, and the transmitting terminal acquires the telemetry simultaneously
It carries out with neuron network simulation output result than sentencing, and combines and fly control prior information generation instruction information to be used for subsequent processing.
The telemetry elastic transport method based on machine learning of the embodiment of the present invention, can be by passing at transmitting-receiving both ends
It passs necessary characteristic parameter and establishes isomorphism neural network model to realize the Accurate Prediction of telemetry, there is higher data pressure
Contracting flexibility and lower data transfer bandwidth demand acquire by data and train study, parameter transmitting with digital simulation, in advance
Survey the flexibility that information transmission is improved than sentencing with processing links such as unruly-value rejectings the elastic transport for realizing telemetry and reliable
Property, there can be the pre- of certain discrepancy tolerance by learning training building with the time-varying characteristics of dynamically adapting telemetry parameter comentropy
Survey grid network, is greatly reduced the transmission capacity of telemetry, and the further end-to-end information of room for promotion Radio Link is transmitted reliable
Property and flexibility.
In addition, the telemetry elastic transport method according to the above embodiment of the present invention based on machine learning can also have
There is following additional technical characteristic:
Further, in one embodiment of the invention, the prediction further comprises: institute than sentencing and unruly-value rejecting
Transmitting terminal is stated to acquire the telemetry and carry out with neuron network simulation output result than sentencing；If the deviation is less than described
Preset threshold, then sending to the receiving end allows to predict to indicate, is otherwise sentenced by accumulation data with the winged control prior information
Whether current measured data of breaking belongs to outlier；If belonging to the outlier, the current measured data is rejected, and connect to described
Receiving end send it is described allow predict indicate, otherwise to the receiving end send network reset instruction, and back to data acquisition with
The training study stage.
Further, in one embodiment of the invention, the transmitting terminal specifically executes following steps: acquisition boat in real time
The telemetry that each subsystem sensor of its device obtains；Training is iterated to the telemetry using machine learning method,
And obtain output parameter；The output parameter is handled, is suitable for constructing telemetry elastic transport model to extract
Characteristic parameter；Network model is constructed using the sign parameter and carries out data simulation output, and is compared with the telemetry
Sentence, carries out unruly-value rejecting on demand and according to result formation allows to predict to instruct or the network reset indicates than sentencing；According to default
Logic rules send the telemetry, the characteristic parameter, the permission predictive information and/or the net to the receiving end
Network reset information.
Further, in one embodiment of the invention, the receiving end specifically executes following steps: receiving transmitting terminal
The telemetry, the characteristic parameter, the permission predictive information and/or the network reset information sent；To described
Telemetry is processed and displayed；Believed according to the characteristic parameter and the permission predictive information and/or the network reset
Breath, the neural network prediction model of construction/reconstruct and the transmitting terminal isomorphism, and generate network available information and prediction output knot
Fruit；According to the permission predictive information and the network available information, start telemetry local simulation output process, with the time
Driving method and the prediction export as a result, according to the network reset information, stop local simulation output process.
Further, in one embodiment of the invention, the format of the telemetry include transmitting terminal framing when
Between stab or frame number, with the receiving end carry out state synchronization operation.
Further, in one embodiment of the invention, wherein the artificial neural network with Time-delayed Feedback is used,
In, network characterization parameter includes input number of nodes, input weight, hidden layer number of nodes, hidden layer weight and bias matrix, output
Layer weight and bias matrix, to carry out network reconfiguration for the receiving end.
In order to achieve the above objectives, another aspect of the present invention embodiment proposes a kind of telemetry bullet based on machine learning
Property transmitting device, comprising: acquisition and training module acquire for data and learn with training, and transmitting terminal acquires telemetry in real time
And it is sent to receiving end, while being trained to transmitting terminal neural network, until the deviation of training result and the telemetry
Lower than preset threshold；Transmitting and analog module will complete training objective for parameter transmitting and digital simulation, the transmitting terminal
Neural network parameter passes to the receiving end, and the receiving end constructs the nerve of isomorphism according to the neural network parameter
Network simultaneously allows to predict that instruction information generates telemetering analogue data using what the transmitting terminal was sent；Than sentencing and rejecting module, use
Sentence in prediction ratio and unruly-value rejecting, the transmitting terminal acquire the telemetry and compared with neuron network simulation output result
Sentence, and combines and fly control prior information generation instruction information to be used for subsequent processing.
The telemetry flexible delivery device based on machine learning of the embodiment of the present invention, can be by passing at transmitting-receiving both ends
It passs necessary characteristic parameter and establishes isomorphism neural network model to realize the Accurate Prediction of telemetry, there is higher data pressure
Contracting flexibility and lower data transfer bandwidth demand acquire by data and train study, parameter transmitting with digital simulation, in advance
Survey the flexibility that information transmission is improved than sentencing with processing links such as unruly-value rejectings the elastic transport for realizing telemetry and reliable
Property, there can be the pre- of certain discrepancy tolerance by learning training building with the time-varying characteristics of dynamically adapting telemetry parameter comentropy
Survey grid network, is greatly reduced the transmission capacity of telemetry, and the further end-to-end information of room for promotion Radio Link is transmitted reliable
Property and flexibility.
In addition, the telemetry flexible delivery device according to the above embodiment of the present invention based on machine learning can also have
There is following additional technical characteristic:
Further, in one embodiment of the invention, described to be further used for than sentencing and rejecting module: the transmission
End acquires the telemetry and carries out with neuron network simulation output result than sentencing；If the deviation is less than the default threshold
Value, then sending to the receiving end allows to predict to indicate, otherwise current by accumulation data and the winged control prior information judgement
Whether measured data belongs to outlier；If belonging to the outlier, the current measured data is rejected, and send out to the receiving end
Send it is described allow predict indicate, otherwise to the receiving end send network reset instruction, and back to data acquire with training learn
The habit stage.
Further, in one embodiment of the invention, the transmitting terminal specifically includes: data acquisition module is used for
The telemetry that each subsystem sensor of acquisition spacecraft obtains in real time；Network training module, for using machine learning method
Training is iterated to the telemetry, and obtains output parameter；Characteristic extracting module, for being carried out to the output parameter
Processing, to extract the characteristic parameter for being suitable for constructing telemetry elastic transport model；Prediction ratio sentences module, described in utilizing
Sign parameter building network model simultaneously carries out data simulation output, and carries out with the telemetry than sentencing, and carries out outlier on demand and picks
It removes and according to than sentencing, result formation allows to predict to instruct or the network reset indicates；Data transmission module, for according to default
Logic rules send the telemetry, the characteristic parameter, the permission predictive information and/or the net to the receiving end
Network reset information.
Further, in one embodiment of the invention, the receiving end specifically includes: data reception module is used for
Receive the telemetry, the characteristic parameter, the permission predictive information and/or network reset letter that transmitting terminal is sent
Breath；Data processing module, for the telemetry to be processed and displayed；Network reconfiguration module, for according to the spy
Levy parameter and the permission predictive information and/or the network reset information, the nerve of construction/reconstruct and the transmitting terminal isomorphism
Network Prediction Model, and generate network available information and prediction output result；Analog output module, for being allowed in advance according to described
Measurement information and the network available information start telemetry local simulation output process, with time driven manner with it is described pre-
Output is surveyed as a result, stopping local simulation output process according to the network reset information.
The additional aspect of the present invention and advantage will be set forth in part in the description, and will partially become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
Above-mentioned and/or additional aspect and advantage of the invention will become from the following description of the accompanying drawings of embodiments
Obviously and it is readily appreciated that, in which:
Fig. 1 is the process according to the telemetry elastic transport method based on machine learning of one embodiment of the invention
Figure；
Fig. 2 is the stable state change type telemetry schematic diagram according to one embodiment of the invention；
Fig. 3 is the slowly varying type telemetry period slow varying parameter schematic diagram according to one embodiment of the invention；
Fig. 4 is the aperiodic slow varying parameter schematic diagram of slowly varying type telemetry according to one embodiment of the invention；
Fig. 5 is the random change type telemetry schematic diagram according to one embodiment of the invention；
Fig. 6 is the Dynamical Recurrent Neural Networks configuration diagram according to one embodiment of the invention；
Fig. 7 is the Dynamical Recurrent Neural Networks training process schematic diagram according to one embodiment of the invention；
Fig. 8 is the transmitting terminal implementation process diagram according to one embodiment of the invention；
Fig. 9 is the receiving end implementation process diagram according to one embodiment of the invention；
Figure 10 is to predict ratio according to the telemetry elastic transport method based on machine learning of one embodiment of the invention
Sentence effect diagram；
Figure 11 is the structure according to the telemetry flexible delivery device based on machine learning of one embodiment of the invention
Schematic diagram；
Figure 12 is the telemetry flexible delivery device based on machine learning according to a specific embodiment of the invention
Structural schematic diagram.
Specific embodiment
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, it is intended to is used to explain the present invention, and is not considered as limiting the invention.
The telemetry elastic transport based on machine learning proposed according to embodiments of the present invention is described with reference to the accompanying drawings
Method and device describes the telemetry elasticity based on machine learning proposed according to embodiments of the present invention with reference to the accompanying drawings first
Transmission method.
Fig. 1 is the flow chart of the telemetry elastic transport method based on machine learning of one embodiment of the invention.
As shown in Figure 1, should telemetry elastic transport method based on machine learning the following steps are included:
In step s101, data acquisition learns with training, and transmitting terminal acquires telemetry in real time and is sent to receiving end,
Transmitting terminal neural network is trained simultaneously, until the deviation of training result and telemetry is lower than preset threshold.
Learn it is understood that the embodiment of the present invention carries out data acquisition first with training, transmitting terminal acquires distant in real time
Measured data is simultaneously sent to receiving end, while being trained to transmitting terminal neural network, until training result and measured data is inclined
Difference is lower than engineering given threshold.Wherein, receiving end need not be to transmitting terminal feedback information, suitable for having long transmission time-delay characteristics
Unidirectional telemetry data transmission scene.
Further, in one embodiment of the invention, the format of telemetry includes the timestamp of transmitting terminal framing
Or frame number, to carry out state synchronization operation with receiving end.
It is understood that telemetry format may include the timestamp or frame number of transmitting terminal framing, for receiving
End carries out state synchronization operation.
Specifically, the classification about telemetry specifically includes:
Spacecraft telemetry is typical time series, due to the collected physical quantity of each subsystem sensor, number
It is not identical according to precision, thus telemetry can include different number, physical meaning and measuring accuracy telemetering parameter, one
As the telemetry parameter that is arranged have temperature, voltage, electric current, state etc., and also there is different telemetry parameters different time changes to become
Gesture, corresponding network characterization parameter are also just different.
From uncertain different, work of the telemetry towards spacecraft platform and loading device of the height of scientific application data
State, the timing variations of each parameter have certain rule, and vulnerable to outside control or such environmental effects.The embodiment of the present invention
In terms of telemetry timing variations trend by spacecraft key telemetry be divided into stable state change type, slowly varying type, with
3 class of machine change type, slowly varying type are further divided into period slow varying parameter and aperiodic slow varying parameter, in which:
(1) stable state change type: having relatively stable mean value, variance smaller, occasionally there is outlier, and (certain satellite is real as shown in Figure 2
Survey MEMS parameter).
(2) slowly varying type: period slow varying parameter: embodying stable period of change in timing, and data value is also in
Mechanical periodicity, as shown in Figure 3 (certain satellite surveys telemetering and recycles frame count)；Aperiodic slow varying parameter: it is embodied in timing point
Section variation tendency, data value have sequential function feature, such as linear function, SIN function, occasionally have outlier, as shown in Figure 4
(certain satellite battery group temperature).
(3) random change type: data variation rule is not obvious, although having mean value variance larger, as shown in Figure 5
(certain satellite GPS lock number of satellites).
Further, in one embodiment of the invention, wherein the artificial neural network with Time-delayed Feedback is used,
In, network characterization parameter includes input number of nodes, input weight, hidden layer number of nodes, hidden layer weight and bias matrix, output
Layer weight and bias matrix, to carry out network reconfiguration for receiving end.
It is understood that the embodiment of the present invention uses the artificial neural network with Time-delayed Feedback, network characterization parameter packet
Include input number of nodes (correlation length), input weight, hidden layer number of nodes, hidden layer weight and bias matrix, output layer weight
With bias matrix etc., network reconfiguration is carried out for receiving end.
Specifically, machine learning method used in the embodiment of the present invention specifically includes:
The embodiment of the present invention can be using the characteristic parameter optimal method based on Dynamical Recurrent Neural Networks as machine
Learning method.This method is using improved Elman network as Dynamical Recurrent Neural Networks framework, by adjusting network input layer section
The network characterizations parameters such as points, the number of plies for accepting layer and number of nodes are iterated preferably the prediction effect of telemetering time series,
Not only have adapt to time-varying characteristics ability, also enhance neural network forecast ability, can be used to a plurality of types of telemetries into
Row prediction modeling.
The neural network of the embodiment of the present invention can be divided into 4 layers, as shown in fig. 6, include input layer, hidden layer, accept layer and
Output layer, in which: input layer completes telemetry input；Output layer take remote measurement data weighting summation；The excitation letter of hidden layer
Number takes Signmoid nonlinear function.The present invention improves the undertaking layer of neural network, from the list of standard Elman network
Layer structure is improved to multilayered structure, and the output result of hidden layer passes through the delay and storage of accepting layer, internal from by way of connection
Tieback can be used to the output valve for remembering hidden layer previous moment or top n moment, be equivalent to single step to the input of hidden layer
Or the delay operator of multi-step delay, the ability that network itself handles multidate information is increased, to achieve the purpose that dynamic modeling.
The date expression of this method is as follows:
Wherein, y is that m ties up output layer knot vector, and m value is 1 in the invention；X is that n ties up hidden layer knot vector, n value
Adjustment can be iterated according to training effect；U is that r ties up input layer vector, and r value can be iterated tune according to training effect
It is whole；Feedback state vector is tieed up for n, j is sequence number, and value is 1~k, and k is the number of plies, and k value is adjustable；It is arrived for hidden layer
Output layer connection weight；For input layer to hidden layer connection weight；To accept layer to hidden layer connection weight；G () is
The transmission function of output neuron is the linear combination of hidden layer output, and the embodiment of the present invention uses improved ReLU function, is used
R () is indicated；F () is the transmission function of hidden layer neuron, and the embodiment of the present invention uses Signmoid function, with S ()
It indicates.
In addition, the embodiment of the present invention carries out network training using supervised learning method, optimization algorithm is selected and standard BP mind
Through the identical gradient descent algorithm of network, it is described as follows.
Error function uses MSE (Mean Squared Error), i.e. mean square error, and mean square error refers in mathematical statistics
The desired value of the difference square of estimates of parameters and parameter true value, specific formula are as follows:
Wherein, ytResult, that is, predicted value is exported for t moment (t=1,2 ..., N) network,For t moment telemetry, that is, mesh
Scale value.
Input layer value is normalized, i.e. ut∈[0,1]；Node layer value is exported to carry out at renormalization
Reason.
Output layer activation primitive is improved ReLU function PReLU, it may be assumed that
Wherein, a value is 1, then has:
R (x)=x,R ' (x)=1 (6),
Hidden layer activation primitive is Signmoid function, it may be assumed that
Then there is its derivative are as follows:
S ' (x)=S (x) × (1-S (x)) (8).
Network training target is so that MSE is minimized, therefore network training process is exactly each node layer of continuous iterated revision
The value of weight and amount of bias makes the smallest optimization process of MSE.Network model parameter mainly includes input layer number, hides
Node layer number accepts layer Delay Feedback network number of plies etc., and initiation parameter mainly includes that maximum number of iterations, minimal gradient take
Value, learning rate etc..
It is analyzed from algorithm, network training process is exactly that gradient decline combines Feedback error, is illustrated such as Fig. 7
It is shown.By taking output layer weight and biasing iterative calculation as an example, have:
Wherein, l indicates iteration order, and α indicates that weight change rate, β indicate that offset change rate, above-mentioned formula show the party
Method can according to the deviation statistics amount (MSE) of current round predicted value and target value, be calculated next one iteration weight with
The adjustment amount of biasing can further decrease MSE value, achieve the purpose that iteration optimization.
In step s 102, parameter transmitting and digital simulation, transmitting terminal pass the neural network parameter for completing training objective
Pass receiving end, and the permission that receiving end is constructed the neural network of isomorphism and sent using transmitting terminal according to neural network parameter
Prediction instruction information generates telemetering analogue data.
It is understood that parameter transmitting and digital simulation, transmitting terminal passes the neural network parameter for completing training objective
Receiving end is passed, receiving end constructs the neural network of isomorphism accordingly and allows to predict that instruction information generates using what transmitting terminal was sent
Telemetering analogue data.
In step s 103, prediction ratio is sentenced and unruly-value rejecting, transmitting terminal acquisition telemetry are simultaneously defeated with neuron network simulation
Result is carried out than sentencing out, and is combined and flown control prior information generation instruction information to be used for subsequent processing.
It is understood that prediction ratio sentences and unruly-value rejecting, it is simultaneously defeated with neuron network simulation that transmitting terminal acquires telemetry
Result is carried out than sentencing out, after allowing to predict, reject outlier, network reset etc. that instruction information is used in conjunction with the formation of winged control prior information
Continuous processing.
Further, in one embodiment of the invention, prediction further comprises: transmitting terminal than sentencing and unruly-value rejecting
Acquisition telemetry is simultaneously carried out with neuron network simulation output result than sentencing；If deviation is less than preset threshold, to receiving end
Transmission allows to predict to indicate, otherwise passes through accumulation data with control prior information is flown and judge whether current measured data belongs to outlier；
If belonging to outlier, current measured data is rejected, and sending to receiving end allows prediction to indicate, otherwise send net to receiving end
Network resetting instruction, and acquired and the training study stage back to data.
It should be noted that elastic transport is embodied on more telemetry parameter common transports, i.e., different types of telemetry parameter
Different time-varying characteristics are embodied, needs to sentence using different prediction ratios and transmits strategy with information, therefore bring each telemetry parameter
The otherness of transinformation, so that information transmission statistic amount of the overall telemetering serial data stream within the unit time shows bullet
Property feature, i.e. telemetry compression efficiency show elastic time-varying characteristics.
Specifically, being classified according to telemetry, the embodiment of the present invention proposes that corresponding prediction transmits plan with information than sentencing
Slightly, specific as follows:
Strategy 1: transmission mean value is regularly updated, receipt of subsequent end is predicted accordingly.The strategy is distant for stable state change type
The prediction ratio of measured data is sentenced to be transmitted with information.
Strategy 2: one section of crude sampling numerical value is first transmitted after trip point, characteristic value is then obtained according to the number of segment according to statistics
New value is transmitted again and slope, receipt of subsequent end are predicted accordingly.The strategy is used for the prediction ratio of slowly varying type telemetry
Sentence and is transmitted with information.
Strategy 3: original telemetry is transmitted.The strategy is passed than sentencing with information for the prediction of random change type telemetry
It is defeated.
The transmitting terminal of the embodiment of the present invention and receiving end are further elaborated below.
Telemetry transmitting terminal, for completing acquisition data, training network, prediction than sentencing, transmitting data and characteristic parameter
Deng.Further, in one embodiment of the invention, as shown in figure 8, transmitting terminal specifically executes following steps: acquisition in real time
The telemetry that each subsystem sensor of spacecraft obtains；Training is iterated to telemetry using machine learning method, and
Obtain output parameter；Output parameter is handled, to extract the feature for being suitable for constructing telemetry elastic transport model ginseng
Number；Using sign parameter building network model and data simulation output is carried out, and carries out carrying out outlier on demand than sentencing with telemetry
It rejects and according to than sentencing, result formation allows to predict to instruct or network reset indicates；It is sent according to logic of propositions rule to receiving end
Telemetry, allows predictive information and/or network reset information at characteristic parameter.
Telemetry receiving end, for completing data receiver, network reconfiguration, simulation output etc..Further, in the present invention
One embodiment in, as shown in figure 9, receiving end specifically executes following steps: receiving telemetry, the feature that transmitting terminal is sent
Parameter allows predictive information and/or network reset information；Telemetry is processed and displayed；According to characteristic parameter and permit
Perhaps predictive information and/or network reset information, the neural network prediction model of construction/reconstruct and transmitting terminal isomorphism, and generate net
Network available information and prediction output result；According to predictive information and network available information is allowed, start this simulation of telemetry
Process is exported, with time driven manner and prediction output as a result, according to network reset information, stops local simulation output process.
The telemetry elastic transport method based on machine learning is carried out into one below by the mode of specific embodiment
Step illustrates.
By taking the transmission of certain satellite MEMS_X telemetry-acquisition data as an example, the prediction ratio obtained using present invention method is sentenced
The working effect of aspect is as shown in Figure 10, and MSE statistical estimation is 7.3538e-05.
With certain satellite in June, 2018 whole real-time telemetry data instance, work of the invention in terms of information transmission is imitated
Fruit carries out as described below.
(1) telemetry is summarized
Number of parameters: it is 31 total, it removes frame count and amounts to 30 physical parameters；
Satellite Tracking segmental arc: 122；
Acquisition data count: 12619；
Each tracking section sampled data number:
51 137 98 139 136 81 16 143 11 122 117 132 104 84 136 112 9 104 138
47 67 141 133 103 128 79 107 129 123 121 43 141 92 19 104 137 85 142 25 121
118 133 105 83 135 111 132 139 50 61 142 131 104 80 106 128 122 120 35 139 90
136 135 86 141 32 118 121 130 106 81 136 107 119 141 51 53 141 130 108 42 79
80 104 130 119 122 26 139 84 122 136 89 139 40 118 123 128 107 78 137 106 131
141 62 52 140 129 107 120 84 102 133 118 11 141 81 118 130 94 138 46
It is assumed that: each telemetry parameter is indicated with 4 byte floating numbers.
(2) the working effect analysis of telemetry parameter is transmitted using strategy 1
Affiliated parameter: 8 total；
Each parameter raw data amount: 12619 4 byte of moment point *=50476 bytes；
Tactful 1 initial data total amount: 50476 parameters of byte * 8=403808 bytes；
Tactful 1 update times: 60, interval point updates once, then amounts to 12619/60 ≈, 211 moment points；
Tactful 1 each parameter more amount of new data: 211 4 byte of moment point *=844 bytes；
Strategy 1 updates total amount of data: 844 parameters of byte * 8=6752 bytes.
(3) the working effect analysis of telemetry parameter is transmitted using strategy 2
Affiliated parameter: 15 total；
Tactful 2 each parameter raw data amounts: 12619 4 byte of moment point *=50476 bytes；
Tactful 2 initial data total amounts: 50476 parameters of byte * 15=757140 bytes；
Tactful 2 update times: 60, interval point updates primary, each section of update times in 122 tracking segmental arcs are as follows:
1 3 2 3 3 2 1 3 1 3 2 3 2 2 3 2 1 2 3 1 2 3 3 2 3 2 2 3 3 3 1 3 2 1 2
3 2 3 1 3 2 3 2 2 3 2 3 3 1 2 3 3 2 2 2 3 3 2 1 3 2 3 3 2 3 1 2 3 3 2 2 3 2 2
3 1 1 3 3 2 1 2 2 2 3 2 3 1 3 2 3 3 2 3 1 2 3 3 2 2 3 2 3 3 2 1 3 3 2 2 2 2 3
213223231, amount to 278 moment points；
Tactful 2 each parameter more amount of new data: 278 moment point * (+4 byte slope of+4 byte coefficient of 4 byte initial data)
=3336 bytes；
Strategy 2 updates total amount of data: 3336 parameters of byte * 15=50040 bytes.
(4) the working effect analysis of telemetry parameter is transmitted using strategy 3
Affiliated parameter: 7 total；
Tactful 3 each parameter raw data amounts: 12619 4 byte of moment point *=50476 bytes；
Tactful 3 initial data total amounts: 50476 parameters of byte * 7=353332 bytes；
Tactful 3 each parameter more amount of new data: 12619 4 byte of moment point *=50476 bytes；
Strategy 3 updates total amount of data: 50476 parameters of byte * 7=353332 bytes.
(5) total information laser propagation effect calculates as follows:
Tactful 1 initial data total amount: 50476 parameters of byte * 8=403808 bytes；
Tactful 2 initial data total amounts: 50476 parameters of byte * 15=757140 bytes；
Tactful 3 initial data total amounts: 50476 parameters of byte * 7=353332 bytes；
Strategy 1 updates total amount of data: 844 parameters of byte * 8=6752 bytes；
Strategy 2 updates total amount of data: 3336 parameters of byte * 15=50040 bytes；
Strategy 3 updates total amount of data: 50476 parameters of byte * 7=353332 bytes；
Ratio=update total amount of data/initial data total amount
=(6752+50040+353332)/(403808+757140+353332) ≈ 0.2708
By above-mentioned analysis it is found that channel capacity needed for the embodiment carries certain satellite telemetering data using the present invention is about
The 27% of original telemetry amount.
Further analysis shows that the telemetry data transmission compression efficiency of the embodiment of the present invention and spacecraft three classes telemetry parameter
Proportion relationship especially stochastic pattern data proportion is closely related, if the ratio of the total telemetry parameter of stochastic pattern telemetry parameter Zhan
Example is p, then the compression ratio limit value of the embodiment of the present invention is p.By taking the embodiment as an example, stochastic pattern telemetry parameter accounting is about
23.33%, therefore telemetry data transmission compression ratio limit value is 23.33% in above-described embodiment.Under normal conditions, spacecraft
Telemetry parameter has certain changing rule mostly, and stochastic pattern telemetry parameter proportion is lower, therefore the embodiment of the present invention has
Having good data compression effects, (if stochastic pattern telemetry parameter accounting is only 1% in certain space flight model task, the present invention is implemented
1%) the telemetry data transmission compression ratio limit value of example is then approached.It follows that telemetering can be greatly reduced in the embodiment of the present invention
Data transmission channel bandwidth demand.
The telemetry elastic transport method based on machine learning proposed according to embodiments of the present invention, can be by receiving
Hair both ends transmit necessary characteristic parameter and establish isomorphism neural network model to realize the Accurate Prediction of telemetry, have higher
Data compression flexibility and lower data transfer bandwidth demand, pass through data acquisition with training study, parameter transmitting and number
The elastic transport for realizing telemetry than sentencing with processing links such as unruly-value rejectings according to simulation, prediction, improves the spirit of information transmission
Activity and reliability can have certain inclined with the time-varying characteristics of dynamically adapting telemetry parameter comentropy by learning training building
The transmission capacity of telemetry, the further end-to-end information of room for promotion Radio Link is greatly reduced in the prediction network of poor tolerance
The reliability and flexibility of transmission.
The telemetry elastic transport based on machine learning proposed according to embodiments of the present invention is described referring next to attached drawing
Device.
Figure 11 is the structural representation of the telemetry flexible delivery device based on machine learning of one embodiment of the invention
Figure.
As shown in figure 11, being somebody's turn to do the telemetry flexible delivery device 10 based on machine learning includes: acquisition and training module
100, transmitting is with analog module 200 and than sentencing and rejecting module 300.
Wherein, acquisition acquires for data with training module 100 and learns with training, and transmitting terminal acquires telemetry simultaneously in real time
It is sent to receiving end, while transmitting terminal neural network is trained, until the deviation of training result and telemetry is lower than pre-
If threshold value.Transmitting is transmitted with analog module 200 for parameter and digital simulation, transmitting terminal will complete the neural network of training objective
Parameter passes to receiving end, and receiving end is constructed the neural network of isomorphism according to neural network parameter and sent using transmitting terminal
Allow predict instruction information generate telemetering analogue data.Than sentencing and rejecting module 300 for predicting than sentencing and unruly-value rejecting, hair
Sending end acquisition telemetry is simultaneously carried out with neuron network simulation output result than sentencing, and winged control prior information is combined to generate instruction letter
Breath is to be used for subsequent processing.The device 10 of the embodiment of the present invention can be led to the time-varying characteristics of dynamically adapting telemetry parameter comentropy
Overfitting training building has the prediction network of certain discrepancy tolerance, and the transmission capacity of telemetry is greatly reduced, further mentions
The reliability and flexibility of the end-to-end information transmission of Radio Link between lift-off.
Further, in one embodiment of the invention, be further used for than sentencing and rejecting module 300: transmitting terminal is adopted
Collection telemetry is simultaneously carried out with neuron network simulation output result than sentencing；If deviation is less than preset threshold, sent out to receiving end
Sending allows to predict to indicate, otherwise judges whether current measured data belongs to outlier with control prior information is flown by accumulation data；Such as
Fruit belongs to outlier, then rejects current measured data, and sending to receiving end allows prediction to indicate, otherwise send network to receiving end
Resetting instruction, and acquired and the training study stage back to data.
Further, in one embodiment of the invention, transmitting terminal specifically includes: data acquisition module, network training
Module, characteristic extracting module, prediction ratio sentence module and data transmission module.
Wherein, the telemetry that data acquisition module is obtained for acquiring each subsystem sensor of spacecraft in real time.Network
Training module is used to be iterated training to telemetry using machine learning method, and obtains output parameter.Feature extraction mould
Block is for handling output parameter, to extract the characteristic parameter for being suitable for constructing telemetry elastic transport model.Prediction
Than sentencing module for constructing network model using sign parameter and carrying out data simulation output, and press than sentencing with telemetry
It need to carry out unruly-value rejecting and according to result formation allows to predict to instruct or network reset indicates than sentencing.Data transmission module is for pressing
Telemetry is sent to receiving end according to logic of propositions rule, characteristic parameter, allows predictive information and/or network reset information.
Specifically, transmitting terminal operational module includes:
Data acquisition module, the telemetry obtained for acquiring each subsystem sensor of spacecraft in real time, and be network
Training module, prediction are than sentencing module, data transmission module offer input data；
Network training module is iterated instruction using the telemetry that machine learning method provides data acquisition module
Practice, and training result is sent to characteristic extracting module, training objective can be designed and adjust according to requirement of engineering with network parameter
It is whole；
Characteristic extracting module, the output parameter provided network training module are handled, and are extracted and are suitable for building telemetering
The characteristic parameter of data elastic transport model is simultaneously sent to prediction than sentencing module, data outputting module, and different physical quantities are available not
It is characterized with characteristic parameter collection；
Prediction ratio sentences module, and it is defeated that the characteristic parameter building network model provided using characteristic extracting module carries out digital simulation
Out, and with data acquisition module the real-time telemetry data provided are carried out than sentencing, and carry out unruly-value rejecting on demand and according to than sentencing result
Formation allows to predict to be sent to data transmission module, or forms network reset information and be sent to data transmission module and restart network
Training module；
Data transmission module receives the original telemetry of data acquisition module, the network characterization ginseng of characteristic extracting module
Number, prediction than sentence module allow predict or network reset information, and according to certain logic rules to receiving end send.
Wherein, the information transmitted between each module of transmitting terminal is as shown in figure 12, specifically includes that
(1) S_In: the telemetry that each subsystem sensor of spacecraft measures；
(2) S1: original telemetry；
(3) S2: neural metwork training result；
(4) S3: network training module instruction of restarting；
(5) S4: allow predictive information/network reset information；
(6) S5: network characterization parameter；
(7) S_Out: original telemetry/network characterization parameter/permission predictive information/network reset information.
Further, in one embodiment of the invention, receiving end specifically includes: data reception module, data processing
Module, network reconfiguration module and analog output module.
Wherein, data reception module is used to receive the telemetry of transmitting terminal transmission, characteristic parameter, allows predictive information
And/or network reset information.Data processing module is for being processed and displayed telemetry.Network reconfiguration module is used for root
According to characteristic parameter and allow predictive information and/or network reset information, the neural network prediction of construction/reconstruct and transmitting terminal isomorphism
Model, and generate network available information and prediction output result.Analog output module is used for according to permission predictive information and network
Available information starts telemetry local simulation output process, with time driven manner and prediction output as a result, according to network weight
Confidence breath stops local simulation output process.
Specifically, receiving end operational module includes:
Data reception module receives original telemetry that transmitting terminal sends, network model characteristic parameter or allows to predict
It indicates information, original telemetry is sent to data processing module, by network model characteristic parameter or network reset direct information
Network reconfiguration module will allow prediction or network reset direct information analog output module；
Data processing module carries out processing to the telemetry that data reception module or analog output module are sent and shows；
Network reconfiguration module, the network model characteristic parameter provided according to data reception module and network reset information, structure
The neural network prediction model with transmitting terminal isomorphism is made/reconstructs, it is defeated with analog output module Internet available information and prediction
Result out；
Analog output module, the net for allowing predictive information and network reconfiguration module to send sent according to data reception module
Network available information starts telemetry local simulation output process, interacts telemetering with network reconfiguration module with time driven manner
Data prediction result, and it is sent to data processing module；According to the network reset information that data reception module is sent, stop local
Simulation output process.
Wherein, the information transmitted between each module in receiving end is as shown in figure 12, specifically includes that
(1) R_In: original telemetry/network characterization parameter/permission predictive information/network reset information；
(2) R1: original telemetry；
(3) R2: network characterization parameter/network reset information；
(4) R3: allow predictive information/network reset information；
(5) R4: network available information/PREDICTIVE CONTROL information/prediction result；
(6) R5: analog telemetering data.
12 pairs of telemetry flexible delivery devices based on machine learning are further elaborated with reference to the accompanying drawing.
Telemetry flexible delivery device system based on machine learning is as shown in figure 12, main to be including each point of spacecraft
4 parts such as system telemetering pickup, transmitting terminal, space communication link, receiving end, wherein transmitting terminal and receiving end are the devices
Nucleus module.
(1) each subsystem telemetering pickup of spacecraft
Each subsystem telemetering pickup of spacecraft is responsible for completing spacecraft platform and loading device each unit device work shape
State and spacecraft environment status information capture carry out data by spacecraft bus network and transmitting terminal to form telemetry
Interaction.
Telemetering pickup device type multiplicity, measurement physical quantity differ from one another, and data precision is by sensor accuracy class, number
Word scale is levied the factors such as bit wide, telemetry-acquisition period and is determined, each telemetry has dynamic time-varying characteristics.
(2) transmitting terminal
Transmitting terminal is sentenced for completing acquisition data, training network, prediction ratio, transmits the work such as data and characteristic parameter, such as
Shown in Fig. 1, specifically include that
Module SM-1: data acquisition module, the telemetry measured for acquiring each subsystem sensor of spacecraft in real time,
And input data is provided for module SM-2, module SM-4, module SM-5；
Module SM-2: network training module changes to the module SM-1 telemetry provided using machine learning method
Generation training, and training result is supplied to module SM-3, training objective and network parameter can be designed according to requirement of engineering with
Adjustment；
Module SM-3: characteristic extracting module handles the module SM-2 output parameter provided, extracts distant suitable for constructing
The characteristic parameter of measured data elastic transport model is simultaneously sent to module SM-4, module SM-5, and different physical quantities can be joined with different characteristic
Manifold characterization；
Module SM-4: prediction ratio sentences module, and the characteristic parameter building network model provided using module SM-3 carries out data
Simulation output, and carry out with the module SM-1 real-time telemetry data provided than sentencing, unruly-value rejecting is carried out on demand and according to than sentencing knot
Fruit shape is sent to module SM-5 at permission predictive information, or forms network reset information and be sent to module SM-5 and restart module
SM-2；
Module SM-5: data transmission module, original telemetry, the network characterization ginseng of module SM-3 of receiving module SM-1
It is several, module SM-4 to allow prediction or network reset information, and sent according to certain logic rules to receiving end.
(3) space communication link
Space communication link is responsible for completing transmitting terminal telemetry and indicates the channel decoding of information, modulation /demodulation, nothing
The physical layers such as line propagation and data link layer work, and are put by Channel coding, modem, Up/Down Conversion device, high power
The devices composition such as big device, low-noise amplifier, phaselocked loop, high-gain aerial.
The main indicator of space communication link includes channel capacity, propagation delay time, bit error rate etc., by communication unit device water
The factors such as flat, spacecraft flight track, space environment are determined that each index has dynamic time-varying characteristics.
(4) receiving end
Receiving end is for completing the work such as data receiver, data processing, network reconfiguration, simulation output, as shown in Figure 1, main
Include:
Module RM-1: data reception module receives original telemetry, net that transmitting terminal is sent by space communication link
Network aspect of model parameter allows to predict instruction information, and original telemetry is sent to module RM-2, joins network model feature
Several or network reset direct information module RM-3, prediction or network reset direct information module RM-4 will be allowed；
Module RM-2: data processing module handles the module RM-1 or module RM-4 telemetry sent；
Module RM-3: network reconfiguration module is believed according to the module RM-1 network model characteristic parameter provided and network reset
Breath, the neural network prediction model of construction/reconstruct and transmitting terminal isomorphism are defeated with module RM-4 Internet available information and prediction
Result out；
Module RM-4: analog output module allows predictive information and module RM-3 to send according to what module RM-1 was sent
Network available information starts telemetry local simulation output process, with time driven manner and module RM-3 interaction prediction control
Information processed obtains the prediction result that module RM-3 is sent and is transmitted to module RM-2；The network reset sent according to module RM-1
Information stops local simulation output process.
It should be noted that the aforementioned explanation to the telemetry elastic transport embodiment of the method based on machine learning
It is also applied for the telemetry flexible delivery device based on machine learning of the embodiment, details are not described herein again.
The telemetry flexible delivery device based on machine learning proposed according to embodiments of the present invention, can be by receiving
Hair both ends transmit necessary characteristic parameter and establish isomorphism neural network model to realize the Accurate Prediction of telemetry, have higher
Data compression flexibility and lower data transfer bandwidth demand, pass through data acquisition with training study, parameter transmitting and number
The elastic transport for realizing telemetry than sentencing with processing links such as unruly-value rejectings according to simulation, prediction, improves the spirit of information transmission
Activity and reliability can have certain inclined with the time-varying characteristics of dynamically adapting telemetry parameter comentropy by learning training building
The transmission capacity of telemetry, the further end-to-end information of room for promotion Radio Link is greatly reduced in the prediction network of poor tolerance
The reliability and flexibility of transmission.
In addition, term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance
Or implicitly indicate the quantity of indicated technical characteristic.Define " first " as a result, the feature of " second " can be expressed or
Implicitly include at least one this feature.In the description of the present invention, the meaning of " plurality " is at least two, such as two, three
It is a etc., unless otherwise specifically defined.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.In the present specification, schematic expression of the above terms are not
It must be directed to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described can be in office
It can be combined in any suitable manner in one or more embodiment or examples.In addition, without conflicting with each other, the skill of this field
Art personnel can tie the feature of different embodiments or examples described in this specification and different embodiments or examples
It closes and combines.
Although the embodiments of the present invention has been shown and described above, it is to be understood that above-described embodiment is example
Property, it is not considered as limiting the invention, those skilled in the art within the scope of the invention can be to above-mentioned
Embodiment is changed, modifies, replacement and variant.
Claims (10)
1. a kind of telemetry elastic transport method based on machine learning, which comprises the following steps:
Data acquisition learns with training, and transmitting terminal acquires telemetry in real time and is sent to receiving end, while to transmission terminal nerve
Network is trained, until the deviation of training result and the telemetry is lower than preset threshold；
The neural network parameter for completing training objective is passed to the reception by parameter transmitting and digital simulation, the transmitting terminal
End, and the receiving end constructs the neural network of isomorphism according to the neural network parameter and utilizes transmitting terminal transmission
Allow to predict that instruction information generates telemetering analogue data；And
Prediction ratio sentences and unruly-value rejecting, and the transmitting terminal acquires the telemetry and carries out with neuron network simulation output result
Than sentencing, and combines and fly control prior information generation instruction information to be used for subsequent processing, and according to different types of telemetry parameter
Time-varying characteristics are sentenced using different prediction ratios and transmit strategy with information, wherein stable state change type telemetry is regularly updated
Transmit mean value；For slowly varying type telemetry, one section of crude sampling numerical value is first transmitted after trip point, then according to the section
Data statistics obtains characteristic value and transmits new value and slope again；For random change type telemetry, original telemetry is transmitted.
2. the telemetry elastic transport method according to claim 1 based on machine learning, which is characterized in that described pre-
Survey ratio sentences and unruly-value rejecting, further comprises:
The transmitting terminal acquires the telemetry and carries out with neuron network simulation output result than sentencing；
If the deviation is less than the preset threshold, sending to the receiving end allows to predict to indicate, otherwise passes through accumulation
Data and the winged control prior information judge whether current measured data belongs to outlier；
If belonging to the outlier, the current measured data is rejected, and allow to predict to refer to receiving end transmission is described
Show, otherwise sends network reset instruction to the receiving end, and acquire and the training study stage back to data.
3. the telemetry elastic transport method according to claim 2 based on machine learning, which is characterized in that the hair
Sending end specifically executes following steps:
The telemetry that each subsystem sensor of acquisition spacecraft obtains in real time；
Training is iterated to the telemetry using machine learning method, and obtains output parameter；
The output parameter is handled, to extract the characteristic parameter for being suitable for constructing telemetry elastic transport model；
Network model is constructed using the characteristic parameter and carries out data simulation output, and is carried out with the telemetry than sentencing,
Unruly-value rejecting is carried out on demand and according to than sentencing, result formation allows to predict to instruct or the network reset indicates；
The telemetry is sent to the receiving end according to logic of propositions rule, the characteristic parameter, described allows to predict to believe
Breath and/or the network reset information.
4. the telemetry elastic transport method according to claim 3 based on machine learning, which is characterized in that described to connect
Receiving end specifically executes following steps:
Receive the telemetry, the characteristic parameter, the permission predictive information and/or the network weight that transmitting terminal is sent
Confidence breath；
The telemetry is processed and displayed；
According to the characteristic parameter and the permission predictive information and/or the network reset information, construction/reconstruct and the hair
The neural network prediction model of sending end isomorphism, and generate network available information and prediction output result；
According to the permission predictive information and the network available information, start telemetry local simulation output process, with when
Between driving method and the prediction export as a result, according to the network reset information, stop local simulation output process.
5. the telemetry elastic transport method according to claim 1 based on machine learning, which is characterized in that described distant
The format of measured data includes the timestamp or frame number of transmitting terminal framing, to carry out state synchronization operation with the receiving end.
6. the telemetry elastic transport method according to claim 1 based on machine learning, which is characterized in that wherein,
Using the artificial neural network with Time-delayed Feedback, wherein network characterization parameter includes input number of nodes, input weight, hidden layer
Number of nodes, hidden layer weight and bias matrix, output layer weight and bias matrix, to carry out network weight for the receiving end
Structure.
7. a kind of telemetry flexible delivery device based on machine learning characterized by comprising
Acquisition and training module, acquire for data and learn with training, and transmitting terminal acquires telemetry in real time and is sent to reception
End, while transmitting terminal neural network is trained, until the deviation of training result and the telemetry is lower than preset threshold；
Transmitting and analog module will complete the neural network of training objective for parameter transmitting and digital simulation, the transmitting terminal
Parameter passes to the receiving end, and the receiving end constructs the neural network and benefit of isomorphism according to the neural network parameter
Allow to predict that instruction information generates telemetering analogue data with what the transmitting terminal was sent；And
Than sentencing and rejecting module, for predicting than sentencing and unruly-value rejecting, the transmitting terminal acquire the telemetry and with nerve
Network analog output result carry out than sentencing, and combine fly control prior information generate instruction information be used for subsequent processing, and according to
The time-varying characteristics of different types of telemetry parameter are sentenced using different prediction ratios and transmit strategy with information, wherein stable state is become
Change type telemetry regularly updates transmission mean value；For slowly varying type telemetry, first one section of transmission is original after trip point
Then sample magnitude obtains characteristic value according to the number of segment according to statistics and transmits new value and slope again；For random change type telemetering number
According to transmitting original telemetry.
8. the telemetry flexible delivery device according to claim 7 based on machine learning, which is characterized in that the ratio
Sentence and be further used for rejecting module:
The transmitting terminal acquires the telemetry and carries out with neuron network simulation output result than sentencing；
If the deviation is less than the preset threshold, sending to the receiving end allows to predict to indicate, otherwise passes through accumulation
Data and the winged control prior information judge whether current measured data belongs to outlier；
If belonging to the outlier, the current measured data is rejected, and allow to predict to refer to receiving end transmission is described
Show, otherwise sends network reset instruction to the receiving end, and acquire and the training study stage back to data.
9. the telemetry flexible delivery device according to claim 8 based on machine learning, which is characterized in that the hair
Sending end specifically includes:
Data acquisition module, the telemetry obtained for acquiring each subsystem sensor of spacecraft in real time；
Network training module for being iterated training to the telemetry using machine learning method, and obtains output ginseng
Number；
Characteristic extracting module is suitable for building telemetry elastic transport for handling the output parameter to extract
The characteristic parameter of model；
Prediction ratio sentences module, for using characteristic parameter building network model and carrying out data simulation output, and with it is described
Telemetry is carried out than sentencing, and is carried out unruly-value rejecting on demand and is allowed to predict instruction or the network reset according to than sentencing result formation
Instruction；
Data transmission module, for sending the telemetry, feature ginseng to the receiving end according to logic of propositions rule
Several, the described permission predictive information and/or the network reset information.
10. the telemetry flexible delivery device according to claim 9 based on machine learning, which is characterized in that described
Receiving end specifically includes:
Data reception module, for receive transmitting terminal transmission the telemetry, the characteristic parameter, it is described allow predict believe
Breath and/or the network reset information；
Data processing module, for the telemetry to be processed and displayed；
Network reconfiguration module is used for according to the characteristic parameter and the permission predictive information and/or the network reset information,
The neural network prediction model of construction/reconstruct and the transmitting terminal isomorphism, and generate network available information and prediction output result；
Analog output module, for it is local to start telemetry according to the permission predictive information and the network available information
Simulation output process is exported with time driven manner and the prediction as a result, stopping local mould according to the network reset information
Quasi- output process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811052935.9A CN109361447B (en) | 2018-09-10 | 2018-09-10 | Telemetry elastic transport method and device based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811052935.9A CN109361447B (en) | 2018-09-10 | 2018-09-10 | Telemetry elastic transport method and device based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109361447A CN109361447A (en) | 2019-02-19 |
| CN109361447B true CN109361447B (en) | 2019-09-17 |
Family
ID=65350757
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811052935.9A Active CN109361447B (en) | 2018-09-10 | 2018-09-10 | Telemetry elastic transport method and device based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109361447B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111490799B (en) * | 2019-01-29 | 2021-08-06 | 中国移动通信有限公司研究院 | Information transmission method and device |
| CN110321940B (en) * | 2019-06-24 | 2021-07-09 | 清华大学 | Aircraft telemetry data feature extraction and classification method and device |
| CN112149696B (en) * | 2019-06-26 | 2024-03-29 | 北京百度网讯科技有限公司 | Training method and device for graph embedding model |
| CN111382514B (en) * | 2020-03-12 | 2023-12-29 | 上海航天控制技术研究所 | Mars detection flight orbit accurate calculation method and system based on supervised learning |
| CN113543186A (en) * | 2020-04-13 | 2021-10-22 | 中国移动通信有限公司研究院 | Transmission method and device |
| CN112001066B (en) * | 2020-07-30 | 2022-11-04 | 四川大学 | Deep learning-based method for calculating limit transmission capacity |
| CN112147978A (en) * | 2020-08-25 | 2020-12-29 | 中国运载火箭技术研究院 | Telemetering data processing method adopting neural network |
| CN111970519B (en) * | 2020-08-28 | 2021-06-15 | 中国人民解放军国防科技大学 | Airborne video return method |
| CN112702106B (en) * | 2020-12-14 | 2022-02-08 | 西安电子科技大学 | Autonomous timing method, system, medium, device, terminal and application |
| CN114679382A (en) * | 2020-12-24 | 2022-06-28 | 华为技术有限公司 | Method and communication device for updating intelligent assembly |
| CN113139714A (en) * | 2021-03-26 | 2021-07-20 | 上海卫星工程研究所 | Method and system for evaluating future state of satellite functional component |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103442225A (en) * | 2013-07-26 | 2013-12-11 | 清华大学 | Remote sensing image transmission system based on database online learning update limited rate |
| CN105320987A (en) * | 2015-09-21 | 2016-02-10 | 航天东方红卫星有限公司 | Satellite telemetry data intelligent interpretation method based on BP neural network |
| CN106446506A (en) * | 2016-08-26 | 2017-02-22 | 航天东方红卫星有限公司 | Multi-method-combined automatic and intelligent prediction method of moonlet state |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102487293B (en) * | 2010-12-06 | 2014-09-03 | 中国人民解放军理工大学 | Satellite communication network abnormity detection method based on network control |
| US9413779B2 (en) * | 2014-01-06 | 2016-08-09 | Cisco Technology, Inc. | Learning model selection in a distributed network |
| CN105825271B (en) * | 2016-03-21 | 2018-10-19 | 南京邮电大学 | Satellite failure diagnosis and prediction method based on evidential reasoning |
| CN108460382B (en) * | 2018-03-26 | 2021-04-06 | 西安电子科技大学 | Optical remote sensing image ship detection method based on deep learning single-step detector |
-
2018
- 2018-09-10 CN CN201811052935.9A patent/CN109361447B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103442225A (en) * | 2013-07-26 | 2013-12-11 | 清华大学 | Remote sensing image transmission system based on database online learning update limited rate |
| CN105320987A (en) * | 2015-09-21 | 2016-02-10 | 航天东方红卫星有限公司 | Satellite telemetry data intelligent interpretation method based on BP neural network |
| CN106446506A (en) * | 2016-08-26 | 2017-02-22 | 航天东方红卫星有限公司 | Multi-method-combined automatic and intelligent prediction method of moonlet state |
Non-Patent Citations (1)
| Title |
|---|
| 卫星遥测时序数据中预测算法研究;钟足华;《南京航空航天大学硕士学位论文》;20150601;全篇 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109361447A (en) | 2019-02-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109361447B (en) | Telemetry elastic transport method and device based on machine learning | |
| Somu et al. | A hybrid model for building energy consumption forecasting using long short term memory networks | |
| Gopalakrishnan et al. | A comparative analysis of models for predicting delays in air traffic networks | |
| Widiasari et al. | Deep learning multilayer perceptron (MLP) for flood prediction model using wireless sensor network based hydrology time series data mining | |
| Chen et al. | Fengwu: Pushing the skillful global medium-range weather forecast beyond 10 days lead | |
| CN109063569B (en) | Semantic level change detection method based on remote sensing image | |
| CN109285346A (en) | A kind of city road net traffic state prediction technique based on key road segment | |
| Ozerdem et al. | Short term load forecasting using particle swarm optimization neural network | |
| CN111142049B (en) | Intelligent transformer fault diagnosis method based on edge cloud cooperation mechanism | |
| Samikwa et al. | Flood prediction using IoT and artificial neural networks with edge computing | |
| CN110991776A (en) | Method and system for realizing water level prediction based on GRU network | |
| McCarthy et al. | Amsterdam to dublin eventually delayed? lstm and transfer learning for predicting delays of low cost airlines | |
| CN113610329B (en) | Short-time rainfall approaching forecasting method of double-current convolution long-short term memory network | |
| CN114580696A (en) | PM (particulate matter)2.5Concentration prediction method | |
| CN113469425A (en) | Deep traffic jam prediction method | |
| Ismail et al. | Adaptive neural network prediction model for energy consumption | |
| Rochester | Characterizing insect migration systems in inland Australia with novel and traditional methodologies | |
| CN116781139A (en) | Flow prediction satellite path selection method and system based on reinforcement learning | |
| Ganchenko et al. | Neural network software diagnosis system of telemetry data | |
| Dridi et al. | Deep learning semantic compression: Iot support over lora use case | |
| Mandal et al. | Prediction of Wind Speed using Machine Learning | |
| CN112561133A (en) | Water flow prediction system and method based on neural network | |
| CN112836865A (en) | Airport airplane flow asynchronous prediction method based on federal learning | |
| Youwei et al. | Predictions to the Water Level of Changjiang Waterway Employing Deep Learning Algorithm LSTM | |
| CN111857015A (en) | Power transmission and transformation cloud intelligent controller |
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 |