CN116520873A - Spacecraft intelligent control method aiming at time optimal low thrust transfer - Google Patents
Spacecraft intelligent control method aiming at time optimal low thrust transfer Download PDFInfo
- Publication number
- CN116520873A CN116520873A CN202310283159.8A CN202310283159A CN116520873A CN 116520873 A CN116520873 A CN 116520873A CN 202310283159 A CN202310283159 A CN 202310283159A CN 116520873 A CN116520873 A CN 116520873A
- Authority
- CN
- China
- Prior art keywords
- neural network
- optimal
- deep neural
- spacecraft
- departure
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 87
- 238000012546 transfer Methods 0.000 title claims abstract description 76
- 238000013528 artificial neural network Methods 0.000 claims abstract description 111
- 238000012549 training Methods 0.000 claims abstract description 68
- 238000013507 mapping Methods 0.000 claims abstract description 5
- 238000012360 testing method Methods 0.000 claims description 13
- 230000007704 transition Effects 0.000 claims description 8
- 230000001172 regenerating effect Effects 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 description 14
- 230000008569 process Effects 0.000 description 10
- 230000009471 action Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000000446 fuel Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000008685 targeting Effects 0.000 description 1
- 238000004800 variational method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
- G05D1/10—Simultaneous control of position or course in three dimensions
- G05D1/101—Simultaneous control of position or course in three dimensions specially adapted for aircraft
- G05D1/106—Change initiated in response to external conditions, e.g. avoidance of elevated terrain or of no-fly zones
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Aviation & Aerospace Engineering (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
Abstract
The invention discloses an intelligent spacecraft control method aiming at time optimal low thrust transfer, which comprises the following steps: setting a departure orbit range and an arrival orbit range of a spacecraft with optimal time and low thrust transfer; selecting a departure track and an arrival track from a track range, setting a plurality of departure position points on the departure track, setting a plurality of arrival position points on the arrival track, generating a small thrust sample set by using a neighbor point iteration method based on the plurality of departure position points and the plurality of arrival position points, wherein the small thrust sample comprises a state variable, a corresponding cooperative variable and optimal transfer time; constructing a deep neural network; training a deep neural network by using the low thrust sample set to fit the mapping relation between the state variable and the cooperative variable and the optimal transfer time; and solving the time optimal low thrust transfer problem by using a deep neural network. The invention can realize the rapid generation of the sample with optimal control of the small thrust transfer, improve the sample generation efficiency and realize the rapid solution of the small thrust transfer problem.
Description
Technical Field
The invention relates to the technical field of spacecraft control, in particular to an intelligent spacecraft control method aiming at time optimal low thrust transfer.
Background
The low thrust propulsion technology can obviously save the propulsion fuel consumption due to smaller specific impact of the engine, and is commonly used for deep space exploration. However, because the thrust is very small, a long time is required for completing the acceleration process, which increases the difficulty for designing the small thrust transfer track, so that the transfer time is an index which is often required to be optimized in the small thrust transfer process, and the problem of time optimal small thrust transfer is derived. Aiming at the realization and solving of the optimal transfer track of the low thrust time, an optimal control solving method is adopted on the theoretical level at present. According to whether the index function is directly optimized or not, the solving method of the optimal control is divided into a direct method, an indirect method and a mixing method. The direct method uses pseudo-spectrum method to replace nonlinear problem with higher-order interpolation. The indirect method is based on the Pontrian Jin Jixiao value principle, the variational method and other theories, derives first-order necessary conditions for optimal control from the Hamiltonian function, and converts the low-thrust track optimization problem into a two-point boundary value problem comprising a state variable and a cooperative variable; the mixing method is a comprehensive application of the direct method and the indirect method. Among them, the direct method has a faster solving efficiency but a lower accuracy. The indirect method has the greatest advantages that the obtained solution has higher precision, better control quantity continuity and meets the first-order optimality condition; however, in the solving process, the cooperative variables have no definite physical meaning, the initial values are difficult to guess, and the boundary conditions and the constraint conditions are sensitive, so that the two-point boundary value problem is difficult to solve. At present, aiming at the difficulty of solving the two-point boundary problem of an indirect method, besides adopting a classical targeting method, the problem of solving the two-point boundary problem by taking a cooperative variable initial value as a parameter to be optimized and utilizing a non-linear programming, intelligent algorithm and other parameter optimization methods is a common solving thought, but still has the problems of larger calculated amount and lower solving success rate.
With the continuous development of machine learning technology, the method for solving the two-point edge value problem by using a machine learning method is proposed so as to solve the time optimal low thrust transfer problem. Specifically, the deep neural network can perform high-precision approximation on the strong nonlinear model, so that the nonlinear model of the two-point boundary value problem is fitted by using the deep neural network, the mapping relation from the state variable to the cooperative variable is learned, and the time optimal low thrust transfer problem is rapidly solved by using the trained neural network model. However, a large number of training samples are required for training the neural network, the generation of the existing training samples still depends on the indirect method for solving, so that a large amount of computing resources and computing time are consumed for generating the training sample set, the cost is high, and as each training sample is independently solved by the indirect method, the problem of low solving success rate still exists.
Disclosure of Invention
In order to solve part or all of the technical problems in the prior art, the invention provides an intelligent spacecraft control method aiming at time optimal low thrust transfer.
The technical scheme of the invention is as follows:
the intelligent spacecraft control method for the time optimal low thrust transfer is provided, and comprises the following steps:
setting a departure orbit range and an arrival orbit range of a spacecraft with optimal time and low thrust transfer;
selecting a departure track and an arrival track from a departure track range and an arrival track range, setting a plurality of departure position points on the departure track, setting a plurality of arrival position points on the arrival track, and generating a small thrust sample set by using a neighbor point iteration method based on the plurality of departure position points and the plurality of arrival position points, wherein the small thrust sample comprises a state variable and a corresponding cooperative variable and optimal transition time thereof, and the state variable comprises track numbers of the departure position and the arrival position;
constructing a deep neural network;
training a deep neural network by using the low thrust sample set to fit the mapping relation between the state variable and the cooperative variable and the optimal transfer time;
and solving the time optimal low thrust transfer problem by using the trained deep neural network, and acquiring a cooperative variable and optimal transfer time to control the spacecraft.
In some possible implementations, the generating the set of low thrust samples by using the neighbor point iteration method based on the plurality of departure location points and the plurality of arrival location points includes:
step 21, selecting a first departure position point on a departure track as an initial neighbor point, selecting a first arrival position point on an arrival track as an initial arrival position, and solving a solution of optimal control of the spacecraft from the first departure position point to the first arrival position point by using an indirect method;
step 22, using the solution of the optimal control of the spacecraft, which is transferred from the last departure position point to the current arrival position, as the initial value of the Newton iteration method, and solving the solution of the optimal control of the spacecraft, which is transferred from the next departure position point to the current arrival position, by using the Newton iteration method;
step 23, repeating the step 22 until obtaining solutions of optimal control of the spacecraft in which all departure position points are transferred to the first arrival position point;
step 24, taking the next arrival position point as an arrival position, transferring the first departure position point to a solution of optimal control of the spacecraft of the previous arrival position point as an initial value of a Newton iteration method, and solving the solution of optimal control of the spacecraft of the first departure position point transferred to the current arrival position by using the Newton iteration method;
step 25, repeating the step 22 until all departure position points are transferred to the solution of the optimal control of the spacecraft of the current arrival position;
step 26, repeating the steps 24 and 25 until a solution of optimal control of the spacecraft is obtained, wherein all departure position points on the departure track are transferred to all arrival position points on the arrival track;
step 27, taking the optimal control solution of each spacecraft and the number of the tracks of the matched departure position point and the arrival position point as a small thrust sample to obtain a small thrust sample set comprising a plurality of small thrust samples, wherein the optimal control solution of the spacecraft comprises: a covariate and an optimal transition time.
In some possible implementations, step 22 and step 24 are performed in parallel.
In some possible implementations, the selecting of the departure orbit and the arrival orbit multiple times is performed within a set departure orbit range and an arrival orbit range of the spacecraft, and a plurality of corresponding small thrust samples are generated for each selected departure orbit and arrival orbit.
In some possible implementations, if the generation success rate of the low thrust sample set is less than the preset success rate threshold, resetting the departure orbit range and the arrival orbit range of the spacecraft with optimal time for low thrust transfer, and regenerating the low thrust sample set based on the reset departure orbit range and the arrival orbit range of the spacecraft until the generation success rate of the low thrust sample set is above the preset success rate threshold.
In some possible implementations, the deep neural network is a fully connected neural network.
In some possible implementations, the training the deep neural network with the set of low thrust samples includes:
dividing the low thrust sample set into a training sample set, a verification sample set and a test sample set according to a preset dividing proportion;
and training the deep neural network by using the training sample set.
In some possible implementations, the training the deep neural network with the training sample set includes:
and taking the state variable in the training sample as the input of the deep neural network, taking the cooperative variable and the optimal transfer time in the training sample as the output of the deep neural network, and training the deep neural network.
In some possible implementations, the training the deep neural network with the state variable in the training sample as an input of the deep neural network and the cooperative variable and the optimal transition time in the training sample as an output of the deep neural network includes:
step 41, sequentially inputting state variables in a plurality of training samples into a deep neural network to obtain predicted cooperative variables and predicted optimal transfer time output by the deep neural network;
step 42, calculating a preset loss function according to the cooperative variables and the optimal transfer time in the training samples and the predicted cooperative variables and the predicted optimal transfer time output by the deep neural network corresponding to the training samples;
and 43, judging whether a preset training stopping condition is met, if so, taking the current deep neural network as a deep neural network for completing training, if not, updating network parameters of the deep neural network by using a preset loss function, and returning to the step 41.
In some possible implementations, the method further includes:
testing the prediction precision of the trained deep neural network by using a test sample set;
if the prediction precision of the deep neural network is smaller than a first preset precision value and larger than a second preset precision value, the deep neural network is improved, and the deep neural network is trained by reusing the low-thrust sample set based on the improved deep neural network;
and if the prediction precision of the deep neural network is below a second preset precision value, regenerating a small thrust sample set with more small thrust sample numbers, and retraining the deep neural network by using the small thrust sample set.
The technical scheme of the invention has the main advantages that:
the intelligent spacecraft control method aiming at the time optimal low thrust transfer can realize the rapid generation of the sample optimally controlled by the low thrust transfer, improve the sample generation efficiency, reduce the sample generation cost, realize the rapid solution of the time optimal low thrust transfer problem, and have high solution efficiency and solution success rate, thereby being applicable to the on-orbit on-line control of the spacecraft.
Drawings
The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention and without limitation to the invention. In the drawings:
FIG. 1 is a flow chart of a spacecraft intelligent control method for time optimal low thrust transfer in accordance with an embodiment of the present invention;
FIG. 2 is a schematic diagram of a relative position relationship between neighboring points according to an embodiment of the present invention;
FIG. 3 is a schematic diagram illustrating a parallel computing process of a neighbor point iteration method according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a deep neural network according to an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to specific embodiments of the present invention and corresponding drawings. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The following describes in detail the technical scheme provided by the embodiment of the invention with reference to the accompanying drawings.
Referring to fig. 1, an embodiment of the present invention provides a spacecraft intelligent control method for time-optimal low thrust transfer, which includes steps 1 to 5:
step 1, setting a departure orbit range and an arrival orbit range of a spacecraft with optimal time and low thrust transfer.
In an embodiment of the invention, the range of the departure orbit and the range of the arrival orbit of the spacecraft, which are used for optimally transferring the low thrust according to the departure orbit and the arrival orbit possibly related to the actual on-orbit control of the spacecraft, are set so as to ensure the credibility when the problem of the time optimal low thrust transfer is solved by using a deep neural network after the follow-up training is completed.
And 2, selecting a departure track and an arrival track from a departure track range and an arrival track range, setting a plurality of departure position points on the departure track, setting a plurality of arrival position points on the arrival track, and generating a low-thrust sample set by using a neighbor point iteration method based on the plurality of departure position points and the plurality of arrival position points, wherein the low-thrust sample comprises a state variable and a corresponding cooperative variable and optimal transition time thereof, and the state variable comprises track numbers of the departure position and the arrival position.
Specifically, in an embodiment of the present invention, a set of low thrust samples is generated by using a neighbor point iteration method based on a plurality of departure location points and a plurality of arrival location points, including the steps of:
step 21, selecting a first departure position point on a departure track as an initial neighbor point, selecting a first arrival position point on an arrival track as an initial arrival position, and solving a solution of optimal control of the spacecraft from the first departure position point to the first arrival position point by using an indirect method;
step 22, using the solution of the optimal control of the spacecraft, which is transferred from the last departure position point to the current arrival position, as the initial value of the Newton iteration method, and solving the solution of the optimal control of the spacecraft, which is transferred from the next departure position point to the current arrival position, by using the Newton iteration method;
step 23, repeating the step 22 until obtaining solutions of optimal control of the spacecraft in which all departure position points are transferred to the first arrival position point;
step 24, taking the next arrival position point as an arrival position, transferring the first departure position point to a solution of optimal control of the spacecraft of the previous arrival position point as an initial value of a Newton iteration method, and solving the solution of optimal control of the spacecraft of the first departure position point transferred to the current arrival position by using the Newton iteration method;
step 25, repeating the step 22 until all departure position points are transferred to the solution of the optimal control of the spacecraft of the current arrival position;
step 26, repeating the steps 24 and 25 until a solution of optimal control of the spacecraft is obtained, wherein all departure position points on the departure track are transferred to all arrival position points on the arrival track;
step 27, taking the optimal control solution of each spacecraft and the number of the tracks of the matched departure position point and the arrival position point as a small thrust sample to obtain a small thrust sample set comprising a plurality of small thrust samples, wherein the optimal control solution of the spacecraft comprises: a covariate and an optimal transition time.
In the process of generating the low thrust sample by using the neighbor point iteration method, if the solution of the optimal control from a certain departure position to a certain arrival position fails, the solution of the low thrust sample corresponding to the current departure track and the arrival track is directly ended.
Referring to fig. 2, for the optimal trajectory transfer problem, the close position points on the same track may be regarded as neighbor points, for example, the departure position-1 and the departure position-2 in fig. 2 may be regarded as two neighbor points, and the arrival position-1 and the arrival position-2 may be regarded as two neighbor points. In an embodiment of the invention, a solution of optimal control of a spacecraft corresponding to a last departure position is used as an initial value of a newton iteration method, a solution of optimal control of the spacecraft corresponding to a neighbor point of the solution is solved by using the newton iteration method, namely a solution of optimal control of the spacecraft corresponding to a next departure position, and a solution of optimal control of the spacecraft corresponding to a last arrival position is used as an initial value of the newton iteration method, and a solution of optimal control of the spacecraft corresponding to a neighbor point of the solution is solved by using the newton iteration method, so that rapid solution of optimal control of low thrust transfer can be realized, a large number of low thrust samples can be rapidly generated, and a higher solution success rate is achieved.
The specific process of solving the solution of the optimal control of the spacecraft from the departure position to the arrival position by using the indirect method and solving the solution of the optimal control of the spacecraft from the departure position to the arrival position by using the newton iteration method are disclosed in the art, and are not specifically described herein.
Referring to fig. 2, in order to further improve the success rate of solving and ensure the accuracy of the solution obtained by solving using the newton iteration method, in an embodiment of the present invention, the difference θ between the closest point angles between two adjacent position points on the track is smaller than the preset angle threshold epsilon.
The angle threshold epsilon is determined according to the sample set scale of actual requirements, and can be set to be 5-30 degrees, for example. The smaller epsilon, the greater the number of location points on each departure track and arrival track, and the greater the number of correspondingly generated low thrust samples.
Referring to fig. 3, in the specific step of generating the set of low thrust samples by using the neighbor point iteration method according to an embodiment of the present invention, the numerical calculations of step 22 and step 24 are independent of each other, and for this purpose, step 22 and step 24 may be performed in parallel. Specifically, after calculating the optimal control solution corresponding to one position point, the optimal control solutions corresponding to a plurality of neighbor points of the sample can be calculated simultaneously through parallel calculation, so that the generation efficiency of the low thrust sample is further improved.
Further, in an embodiment of the present invention, multiple selections of the departure track and the arrival track are performed within a set departure track range and arrival track range of the spacecraft, and a plurality of corresponding low thrust samples are generated for each selected departure track and arrival track.
Specifically, for the selected departure track and arrival track, a plurality of corresponding small thrust samples are generated by using the defined neighbor point iteration method, and all the generated small thrust samples form a small thrust sample set.
Through selecting the departure track and the arrival track for a plurality of times, and generating a plurality of corresponding low thrust samples aiming at the departure track and the arrival track selected each time, a larger number of low thrust samples can be obtained, so that the accuracy of the deep neural network obtained by subsequent training is ensured.
Further, in an embodiment of the present invention, if the generation success rate of the small thrust sample set is smaller than a preset success rate threshold, resetting a departure orbit range and an arrival orbit range of the spacecraft with optimal time for small thrust transfer, and regenerating the small thrust sample set based on the reset departure orbit range and arrival orbit range of the spacecraft until the generation success rate of the small thrust sample set is above the preset success rate threshold, wherein the generation success rate of the small thrust sample set=the actual generation number of the small thrust samples/the theoretical generation number of the small thrust samples×100%.
In an embodiment of the present invention, the success rate threshold is specifically set according to the number of samples required in practice and the accuracy of the deep neural network, for example, may be 80%, 85% or 90%.
In an embodiment of the present invention, the above-defined neighbor point iteration method is applicable to different two-body orbit dynamics models, including two-body dynamics models under rectangular coordinate system, spherical coordinate system and cylindrical coordinate system, and orbit dynamics models based on improved spring festival points.
And 3, constructing a deep neural network.
In one embodiment of the present invention, the deep neural network is a fully connected neural network.
Referring to fig. 4, in particular, the fully-connected neural network includes an input layer, a hidden layer, and an output layer connected in sequence, and the nonlinear activation function between layers adopts a ReLu function.
And step 4, training the deep neural network by using the low-thrust sample set to fit the mapping relation between the state variable and the cooperative variable and the optimal transfer time.
In one embodiment of the present invention, training a deep neural network using a set of low thrust samples includes:
dividing the low thrust sample set into a training sample set, a verification sample set and a test sample set according to a preset dividing proportion, wherein the training sample set is used for training the deep neural network, the verification sample set is used for testing the deep neural network in the training process, so that the super parameters of the deep neural network are adjusted according to the test result, and the test sample set is used for testing the prediction precision of the trained deep neural network.
The dividing ratio is specifically set according to actual requirements, for example, may be 8:1:1.
Further, in an embodiment of the present invention, training the deep neural network using the training sample set includes:
and taking the state variable in the training sample as the input of the deep neural network, taking the cooperative variable and the optimal transfer time in the training sample as the output of the deep neural network, and training the deep neural network.
The input of the deep neural network is a state variable, namely the number of tracks of a departure position and an arrival position, is 12-dimensional, the output of the deep neural network is a cooperative variable and optimal transfer time, and is 7-dimensional, and the cooperative variable comprises 6-dimensional and the optimal transfer time comprises 1-dimensional.
Specifically, in an embodiment of the present invention, a state variable in a training sample is used as an input of a deep neural network, a cooperative variable and an optimal transition time in the training sample are used as an output of the deep neural network, and the deep neural network is trained, including the following steps 41-43:
and step 41, sequentially inputting state variables in a plurality of training samples into the deep neural network to obtain predicted cooperative variables and predicted optimal transfer time output by the deep neural network.
In one embodiment of the invention, the state variables in the training samples are input from the input end of the deep neural network, sequentially processed by parameters of each layer in the deep neural network, and output from the output end of the deep neural network, wherein the information output by the output end is the predicted collaborative variable and the predicted optimal transfer time corresponding to the state variables.
In an embodiment of the present invention, the deep neural network may be an untrained or untrained network model, each layer of the network model is provided with an initialization parameter, and the parameters of each layer may be updated and adjusted continuously during the training process of the network model.
And 42, calculating a preset loss function according to the cooperative variables and the optimal transfer time in the training samples and the predicted cooperative variables and the predicted optimal transfer time output by the deep neural network corresponding to the training samples.
Specifically, a loss function used for deep neural network training is preset, and a preset loss function is calculated according to the cooperative variables and the optimal transfer time in a plurality of training samples and the predicted cooperative variables and the predicted optimal transfer time output by the deep neural network.
In one embodiment of the present invention, the loss function is an absolute error loss function or a relative error loss function.
And 43, judging whether a preset training stopping condition is met, if so, taking the current deep neural network as the deep neural network for completing training, if not, updating parameters of the deep neural network by using a preset loss function, and returning to the step 41.
In an embodiment of the present invention, the training stop condition is specifically set according to the actual situation, for example, the training iteration number reaches a set iteration algebra.
Further, in an embodiment of the present invention, the following formula may be used to update parameters of the deep neural network:
wherein Θ is t+1 Parameters representing the deep neural network at the t+1st iteration, Θ t Parameters representing the deep neural network at the t-th iteration, delta [ · ]]Representing an optimizer, η representing a learning rate, L representing a loss function, Θ representing parameters of the deep neural network. The optimizer is specifically configured according to the actual situation, for example Adam, SGD, etc., and the learning rate is preset and used for controlling the speed of parameter updating.
Further, in an embodiment of the present invention, the method further includes:
testing the prediction precision of the trained deep neural network by using a test sample set;
if the prediction precision of the deep neural network is smaller than a first preset precision value and larger than a second preset precision value, the deep neural network is improved, and the deep neural network is trained by reusing the low-thrust sample set based on the improved deep neural network;
and if the prediction precision of the deep neural network is below a second preset precision value, regenerating a small thrust sample set with more small thrust sample numbers, and retraining the deep neural network by using the small thrust sample set.
The preset precision value is specifically set according to practical situations, for example, the first preset precision value is 90%, and the second preset precision value is 70%.
In one embodiment of the present invention, the improvement of the deep neural network includes: and adjusting at least one of super parameters such as the network layer number, the neuron number of each layer, the learning rate, the iteration algebra and the like of the deep neural network.
And step 5, solving the time optimal low thrust transfer problem by using the trained deep neural network, and obtaining the cooperative variable and the optimal transfer time to control the spacecraft.
Specifically, when the time optimal low thrust transfer problem is needed to be solved, determining the orbit root numbers of the departure position and the arrival position of the spacecraft in the time optimal low thrust transfer problem, and inputting the orbit root numbers of the departure position and the arrival position of the spacecraft into the trained deep neural network to obtain the cooperative variable and the optimal transfer time output by the deep neural network.
The intelligent spacecraft control method for the time optimal low thrust transfer, provided by the embodiment of the invention, can realize the rapid generation of the sample for the low thrust transfer optimal control, improve the sample generation efficiency, reduce the sample generation cost, realize the rapid solution of the time optimal low thrust transfer problem, and have high solution efficiency and solution success rate, thus being applicable to the on-orbit on-line control of the spacecraft.
It should be noted that in this document, relational terms such as "first" and "second" and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting thereof; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. An intelligent spacecraft control method for time optimal low thrust transfer is characterized by comprising the following steps:
setting a departure orbit range and an arrival orbit range of a spacecraft with optimal time and low thrust transfer;
selecting a departure track and an arrival track from a departure track range and an arrival track range, setting a plurality of departure position points on the departure track, setting a plurality of arrival position points on the arrival track, and generating a small thrust sample set by using a neighbor point iteration method based on the plurality of departure position points and the plurality of arrival position points, wherein the small thrust sample comprises a state variable and a corresponding cooperative variable and optimal transition time thereof, and the state variable comprises track numbers of the departure position and the arrival position;
constructing a deep neural network;
training a deep neural network by using the low thrust sample set to fit the mapping relation between the state variable and the cooperative variable and the optimal transfer time;
and solving the time optimal low thrust transfer problem by using the trained deep neural network, and acquiring a cooperative variable and optimal transfer time to control the spacecraft.
2. The intelligent spacecraft control method for time optimal low thrust transfer of claim 1, wherein generating a low thrust sample set by a neighbor point iteration method based on a plurality of departure location points and a plurality of arrival location points comprises:
step 21, selecting a first departure position point on a departure track as an initial neighbor point, selecting a first arrival position point on an arrival track as an initial arrival position, and solving a solution of optimal control of the spacecraft from the first departure position point to the first arrival position point by using an indirect method;
step 22, using the solution of the optimal control of the spacecraft, which is transferred from the last departure position point to the current arrival position, as the initial value of the Newton iteration method, and solving the solution of the optimal control of the spacecraft, which is transferred from the next departure position point to the current arrival position, by using the Newton iteration method;
step 23, repeating the step 22 until obtaining solutions of optimal control of the spacecraft in which all departure position points are transferred to the first arrival position point;
step 24, taking the next arrival position point as an arrival position, transferring the first departure position point to a solution of optimal control of the spacecraft of the previous arrival position point as an initial value of a Newton iteration method, and solving the solution of optimal control of the spacecraft of the first departure position point transferred to the current arrival position by using the Newton iteration method;
step 25, repeating the step 22 until all departure position points are transferred to the solution of the optimal control of the spacecraft of the current arrival position;
step 26, repeating the steps 24 and 25 until a solution of optimal control of the spacecraft is obtained, wherein all departure position points on the departure track are transferred to all arrival position points on the arrival track;
step 27, taking the optimal control solution of each spacecraft and the number of the tracks of the matched departure position point and the arrival position point as a small thrust sample to obtain a small thrust sample set comprising a plurality of small thrust samples, wherein the optimal control solution of the spacecraft comprises: a covariate and an optimal transition time.
3. The intelligent control method for a space vehicle for time optimal low thrust transfer according to claim 2, wherein step 22 and step 24 are performed in parallel.
4. The spacecraft intelligent control method for time optimal low thrust transfer according to claim 1 or 2, wherein selection of a plurality of departure tracks and arrival tracks is performed within a set departure track range and arrival track range of a spacecraft, and a plurality of corresponding low thrust samples are generated for each selected departure track and arrival track.
5. The spacecraft intelligent control method for time-optimal low-thrust transfer according to claim 1 or 2, wherein if the generation success rate of the low-thrust sample set is smaller than a preset success rate threshold, resetting a departure orbit range and an arrival orbit range of the spacecraft for time-optimal low-thrust transfer, and regenerating the low-thrust sample set based on the reset departure orbit range and arrival orbit range of the spacecraft until the generation success rate of the low-thrust sample set is above the preset success rate threshold.
6. The spacecraft intelligent control method for time optimal low thrust transfer of claim 1, wherein the deep neural network is a fully connected neural network.
7. The spacecraft intelligent control method for time-optimal low thrust transfer of claim 1 or 6, wherein training the deep neural network with a set of low thrust samples comprises:
dividing the low thrust sample set into a training sample set, a verification sample set and a test sample set according to a preset dividing proportion;
and training the deep neural network by using the training sample set.
8. The method for intelligent control of a spacecraft for time-optimal low thrust transfer of claim 7, wherein training the deep neural network with a training sample set comprises:
and taking the state variable in the training sample as the input of the deep neural network, taking the cooperative variable and the optimal transfer time in the training sample as the output of the deep neural network, and training the deep neural network.
9. The intelligent spacecraft control method for time optimal low thrust transfer of claim 8, wherein the training the deep neural network with the state variable in the training sample as the input of the deep neural network and the cooperative variable and the optimal transfer time in the training sample as the output of the deep neural network comprises:
step 41, sequentially inputting state variables in a plurality of training samples into a deep neural network to obtain predicted cooperative variables and predicted optimal transfer time output by the deep neural network;
step 42, calculating a preset loss function according to the cooperative variables and the optimal transfer time in the training samples and the predicted cooperative variables and the predicted optimal transfer time output by the deep neural network corresponding to the training samples;
and 43, judging whether a preset training stopping condition is met, if so, taking the current deep neural network as a deep neural network for completing training, if not, updating network parameters of the deep neural network by using a preset loss function, and returning to the step 41.
10. The intelligent control method for a spacecraft for time optimal low thrust transfer of claim 7, further comprising:
testing the prediction precision of the trained deep neural network by using a test sample set;
if the prediction precision of the deep neural network is smaller than a first preset precision value and larger than a second preset precision value, the deep neural network is improved, and the deep neural network is trained by reusing the low-thrust sample set based on the improved deep neural network;
and if the prediction precision of the deep neural network is below a second preset precision value, regenerating a small thrust sample set with more small thrust sample numbers, and retraining the deep neural network by using the small thrust sample set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310283159.8A CN116520873A (en) | 2023-03-22 | 2023-03-22 | Spacecraft intelligent control method aiming at time optimal low thrust transfer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310283159.8A CN116520873A (en) | 2023-03-22 | 2023-03-22 | Spacecraft intelligent control method aiming at time optimal low thrust transfer |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116520873A true CN116520873A (en) | 2023-08-01 |
Family
ID=87400149
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310283159.8A Pending CN116520873A (en) | 2023-03-22 | 2023-03-22 | Spacecraft intelligent control method aiming at time optimal low thrust transfer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116520873A (en) |
-
2023
- 2023-03-22 CN CN202310283159.8A patent/CN116520873A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107193212B (en) | Aero-engine nonlinear predictive control method based on novel wolf optimization algorithm | |
| CN111047085A (en) | Hybrid vehicle working condition prediction method based on meta-learning | |
| CN114839880B (en) | Self-adaptive control method based on flexible joint mechanical arm | |
| CN110716575A (en) | UUV real-time collision avoidance planning method based on deep double-Q network reinforcement learning | |
| CN113093526B (en) | Overshoot-free PID controller parameter setting method based on reinforcement learning | |
| CN109062040B (en) | PID (proportion integration differentiation) predicting method based on system nesting optimization | |
| CN113138555A (en) | GRNN electric spindle thermal error modeling method based on genetic algorithm optimization | |
| CN111553118A (en) | Multidimensional continuous optimization variable global optimization method based on reinforcement learning | |
| CN115867920A (en) | Method for configuring a control agent for a technical system and control device | |
| CN113141012A (en) | Power grid power flow regulation and control decision reasoning method based on deep deterministic strategy gradient network | |
| Mu et al. | Mixed reinforcement learning for efficient policy optimization in stochastic environments | |
| CN110851911B (en) | Terminal state calculation model training method, control sequence searching method and device | |
| Simutis et al. | A comparative study on random search algorithms for biotechnical process optimization | |
| CN116880191A (en) | Intelligent control method of process industrial production system based on time sequence prediction | |
| CN116520873A (en) | Spacecraft intelligent control method aiming at time optimal low thrust transfer | |
| CN111624872B (en) | PID controller parameter setting method and system based on adaptive dynamic programming | |
| CN116819973A (en) | Track tracking control method | |
| CN114967472A (en) | Unmanned aerial vehicle trajectory tracking state compensation depth certainty strategy gradient control method | |
| CN114384931B (en) | Multi-target optimal control method and equipment for unmanned aerial vehicle based on strategy gradient | |
| CN116300755A (en) | Double-layer optimal scheduling method and device for heat storage-containing heating system based on MPC | |
| CN104537224A (en) | Multi-state system reliability analysis method and system based on self-adaptive learning algorithm | |
| Zhao et al. | RNN learning controller trained by ADRC | |
| CN105068423B (en) | Method for realizing intelligent parameter identification of steam turbine and speed regulating system thereof by one key | |
| Baker et al. | Connectionist learning systems for control | |
| CN112015081B (en) | Parameter self-tuning method of SISO (SISO) compact-format model-free controller based on PSO-LSTM (particle swarm optimization-least Square transform) cooperative algorithm |
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 |