CN112180726A - Spacecraft relative motion trajectory planning method based on meta-learning - Google Patents
Spacecraft relative motion trajectory planning method based on meta-learning Download PDFInfo
- Publication number
- CN112180726A CN112180726A CN202011055523.8A CN202011055523A CN112180726A CN 112180726 A CN112180726 A CN 112180726A CN 202011055523 A CN202011055523 A CN 202011055523A CN 112180726 A CN112180726 A CN 112180726A
- Authority
- CN
- China
- Prior art keywords
- meta
- relative motion
- training
- spacecraft
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/0265—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric the criterion being a learning criterion
Landscapes
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a spacecraft relative motion track planning method based on meta-learning, which is characterized in that training of a meta-learner is realized through a meta-training process and a fine-tuning process, an initial state, a terminal state, a constraint condition and a performance index of a relative motion track planning task are input into the meta-learner, and a predicted discrete optimal relative motion track curve is output. The method not only avoids the problem that a numerical method for planning the relative motion track of the spacecraft needs a large amount of calculation time, but also avoids the problem that a traditional machine learning method needs to rely on a large amount of track samples; when a new relative motion trajectory planning problem is encountered, the whole original model does not need to be retrained at a huge cost, but only needs to be subjected to one-step or several-step fine adjustment under a new working condition, so that the method is simple and quick, and the time consumption is greatly reduced.
Description
Technical Field
The invention belongs to the technical field of spacecraft trajectory planning, and particularly relates to a spacecraft relative motion trajectory planning method based on meta-learning.
Background
The existing spacecraft trajectory planning method comprises a plurality of specific methods, such as an analytical method based on a minimum value principle, a numerical method based on calculation, a method based on approximate simplification and the like. (1) The analysis method is suitable for the conditions that a system dynamic model is simple and the state variables and the control variables are few (for example, the lunar landing soft landing trajectory planning of a two-dimensional plane only comprises 3 state equations, 2 state quantities of x and y direction positions and 1 control quantity of a thrust direction), and the analysis method can provide an algebraic analytic solution of a problem. (2) The numerical method transforms the trajectory planning problem into some specific numerical problems (such as a two-point boundary value problem), and then calls a mature optimizer to solve the problems, so as to obtain a numerical solution of the original problem finally. (3) The original trajectory planning problem is simplified and approximated to some extent, so that the problem solving difficulty can be reduced, and further, some numerical methods which cannot be solved quickly can be solved quickly.
However, the above methods have technical problems:
(1) analytical methods require complex derivations and in the face of complex kinetic problems involving large amounts of constraints, analytical solutions are not necessarily derivable.
(2) The numerical method is suitable for the trajectory planning problem of various complexity degrees, but generally needs a large amount of calculation time consumption.
(3) The simplification method requires that the problem must have certain particularity to be simplified, and the obtained solution is only an approximate solution of the original problem and is difficult to be popularized to the problem of arbitrary trajectory planning.
Therefore, how to provide a spacecraft relative motion trajectory planning method based on meta-learning is a problem that needs to be solved urgently by those skilled in the art.
Disclosure of Invention
In view of the above, the invention provides a spacecraft relative motion trajectory planning method based on meta-learning, which not only avoids the problem that a numerical method for spacecraft relative motion trajectory planning needs a large amount of computation time consumption, but also avoids the problem that a traditional machine learning method needs to rely on a large amount of trajectory samples.
In order to achieve the purpose, the invention adopts the following technical scheme:
a spacecraft relative motion track planning method based on meta-learning solves an optimal control problem through a numerical method to obtain optimal relative motion tracks under various working conditions, then training of the meta-learner is achieved through a meta-training process and a fine-tuning process, initial states, terminal states, constraint conditions and performance indexes of a relative motion track planning task are input into the meta-learner, and a predicted discrete optimal relative motion track curve is output.
Preferably, the meta-learner comprises a planning of a relative motion trajectory between two spacecrafts, including a target spacecraft and a service spacecraft, wherein the trajectory is an approximate optimal approaching trajectory of the service spacecraft from an initial relative position to approach the target spacecraft.
Preferably, under the condition of setting the rotation angular velocity of the target spacecraft, for each initial relative state, planning the optimal approximate relative motion track by adopting a GPOPS method to obtain discrete optimal approximate relative motion track points and corresponding discrete optimal control quantity; and changing the rotation angular velocity of the target spacecraft to finish the planning of the optimal approximate relative motion trajectory in all initial relative states under all the rotation angular velocities of the target spacecraft.
Preferably, all the tracks obtained after planning are divided into a meta training set and a meta testing set according to a certain proportion; dividing the meta-training set into a support set and a query set according to a certain proportion; and dividing the meta-test set into a training set and a test set according to a certain proportion.
Preferably, in the meta-training process, updating the original model by adopting a dual gradient updating mode to obtain a training model; and in the fine tuning process, fine tuning is carried out on the parameters of the training model by adopting the training set to obtain a fine tuning model.
Preferably, the updating of the original model is performed by using a dual gradient updating method, and the method for obtaining the training model comprises the following steps:
1) in each track category, traversing all tracks in the category; then calculating a loss function of the original model on each track;
2) calculating the gradient of the loss function to the original model parameters;
3) updating the original model parameters for one time according to the calculated gradient;
4) after the updating of the original model parameters of each track in the category is finished, the updated model parameters are obtained, and the loss function of the updated model on the query set is calculated;
5) and calculating the gradient of the loss function to the original model parameters, and performing secondary updating on the original model parameters to obtain a training model.
Preferably, in the meta-training process, a query set is used to evaluate the training effect of the meta-learner.
Preferably, the method for obtaining the fine tuning model by fine tuning the parameters of the training model by using the training set comprises the following steps:
a) in the track category of each training set, traversing all tracks in the category; then calculating a loss function of the training model on each track;
b) calculating the gradient of the loss function to the parameters of the training model;
c) and updating the parameters of the training model for one time according to the calculated gradient to obtain a fine tuning model.
Preferably, in the fine tuning process, a test set is used for testing and evaluating the fine tuning model.
Preferably, the model parameters are updated once by adopting methods including SGD and Adam.
The invention has the beneficial effects that:
the training of the meta-learner is realized through the meta-training process and the fine-tuning process, so that the meta-learner outputs the predicted discrete optimal relative motion trajectory curve, the problem that a numerical method for spacecraft relative motion trajectory planning needs a large amount of calculation time is solved, and the problem that a traditional machine learning method needs to rely on a large amount of trajectory samples is solved; when a new relative motion trajectory planning problem is encountered, the whole original model does not need to be retrained at a huge cost, but only needs to be subjected to one-step or several-step fine adjustment under a new working condition, so that the method is simple and quick, and the time consumption is greatly reduced.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is a schematic diagram of the distribution of the N-way K-shot relative motion trajectory of the present invention.
FIG. 2 is a schematic diagram of the meta-training process and the fine-tuning process of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention provides a spacecraft relative motion track planning method based on meta-learning, which is characterized in that an optimal control problem is solved through a numerical method to obtain optimal relative motion tracks under various working conditions, then the training of a meta-learner is realized through a meta-training process and a fine-tuning process, the initial state, the terminal state, the constraint condition and the performance index of a relative motion track planning task are input into the meta-learner, and a predicted discrete optimal relative motion track curve is output.
The meta-learner comprises a relative motion trajectory planning between two spacecrafts including a target spacecraft and a service spacecraft, wherein the trajectory is an approximate optimal approaching trajectory of the service spacecraft from an initial relative position to approach the target spacecraft. Under the condition of setting the rotation angular velocity of the target spacecraft, for each initial relative state, the optimal approximate relative motion trajectory planning is completed by numerical methods including but not limited to a GPOPS method and the like, and discrete optimal approximate relative motion trajectory points and corresponding discrete optimal control quantities are obtained; and changing the rotation angular velocity of the target spacecraft to finish the planning of the optimal approximate relative motion trajectory in all initial relative states under all the rotation angular velocities of the target spacecraft. Dividing all the tracks obtained after planning into a meta training set and a meta testing set according to a certain proportion; dividing the meta-training set into a support set and a query set according to a certain proportion; and dividing the meta-test set into a training set and a test set according to a certain proportion.
In the meta-training process, updating the original model by adopting a double gradient updating mode to obtain a training model; and in the fine tuning process, fine tuning is carried out on the parameters of the training model by adopting the training set to obtain a fine tuning model. And in the meta-training process, a query set is adopted to evaluate the training effect of the meta-learner. And in the fine adjustment process, testing and evaluating the fine adjustment model by adopting a test set.
In the meta-training process and the fine-tuning process, one-step gradient updating or multi-step gradient updating can be adopted.
Updating the original model by adopting a double gradient updating mode, wherein the method for obtaining the training model comprises the following steps:
1) in each track category, traversing all tracks in the category; then calculating a loss function of the original model on each track;
2) calculating the gradient of the loss function to the original model parameters;
3) according to the calculated gradient, primary updating is carried out on the original model parameters by adopting methods including SGD (generalized mean) and Adam;
4) after the updating of the original model parameters of each track in the category is finished, the updated model parameters are obtained, and the loss function of the updated model on the query set is calculated;
5) and calculating the gradient of the loss function to the original model parameters, and performing secondary updating on the original model parameters to obtain a training model.
The method adopts an N-way K-shot small sample learning mode to train. In the learning mode of the N-way K-shot small samples, N-way represents N categories, and the method represents track planning tasks of different categories in N, such as approaching track planning tasks of target spacecrafts with different rotation angular velocities; k-shot represents K samples contained in each category, and K different approaching tracks are represented in the method, such as the approaching tracks which are approached from different initial positions for a target spacecraft with a specific rotation angular velocity.
In an N-way K-shot small sample learning mode in the meta-training process, N x K sample tracks are adopted to form a support set (support set), N x Q sample tracks are designed to form a query set (query set), and a non-intersection set between the support set and the query set is an empty set.
The method for fine tuning the parameters of the training model by adopting the training set comprises the following steps:
a) in the track category of each training set, traversing all tracks in the category; then calculating a loss function of the training model on each track;
b) calculating the gradient of the loss function to the parameters of the training model;
c) and updating the parameters of the training model once by adopting methods including SGD and Adam according to the calculated gradient to obtain a fine tuning model.
In the fine adjustment process, a 1-way K-shot small sample learning mode is adopted for fine adjustment. The type 1 trajectory planning task used in the fine tuning process is different from the type N trajectory planning task used in the meta-training process.
In fig. 1, 1 is a target spacecraft; 2 is a service spacecraft; 3 is a kind of optimal relative motion approaching track corresponding to the rotation angular velocity of a certain target spacecraft; 4 is an optimal relative motion approach trajectory corresponding to another target spacecraft rotation angular velocity.
As shown in FIG. 1, the invention designs the appropriate target spacecraft rotation angular velocity according to the actual mission requirements of trajectory planning.
Several suitable initial relative states and one terminal relative state are designed, and as shown in fig. 1, 6 points are equidistantly taken on a circular ring which surrounds the target spacecraft at equal distances, wherein K is equal to 6 points, and the initial relative states are taken as the initial relative states.
According to the actual task requirements of the trajectory planning, the performance index function of the task is designed, and the performance index function comprises but is not limited to fuel optimization (fuel is the most economical), time optimization (the time is the shortest), mixing optimization (the weighted sum of the time and the fuel is the minimum), and the like.
Under the condition of the rotation angular velocity of the specific target spacecraft, for each initial relative state, the optimal approximate relative motion trajectory planning is completed by methods including but not limited to GPOPS and the like, and the discrete optimal approximate relative motion trajectory point and the corresponding discrete optimal control quantity are obtained.
And changing the rotation angular velocity of the target spacecraft and repeating the previous step.
And finishing the planning of the optimal approximate relative motion trajectory in all initial relative states under the rotation angular velocity of all target spacecrafts.
All tracks obtained by planning are divided into a meta-training set (meta-train) and a meta-test set (meta-test) according to a certain proportion (such as 80% and 20%).
The meta-training set is divided into a support set (support set) and a query set according to a certain proportion.
The meta-test set is divided into a training set (train set) and a test set (test set) according to a certain proportion.
As shown in FIG. 2, the training of the meta-learner of the present invention is divided into a meta-training process and a fine-tuning process.
1. MSE (mean square error) is used as the loss function.
2. In the meta-training process, model parameters are initialized randomly.
3. In the meta-training process, a support set training meta-learner is used for inputting the initial state, the terminal state, the constraint condition and the performance index of a relative motion track planning task, and the discrete optimal relative motion track point obtained by planning is expected to be output. A loss function is computed for the original model on each trace in each category of the support set.
4. In the meta-training process, a dual gradient updating mode is adopted to train the meta-learner:
A) in each track category, traversing all tracks in the category; the loss function of the original model is then calculated on each trajectory.
B) The gradient of the loss function to the original model parameters is calculated.
C) According to the calculated gradient, the original model parameters are updated once by methods including but not limited to SGD (random gradient descent), Adam and the like. The number of updating steps may be one step or multiple steps.
D) And after the updating of the original model parameters of each track in the category is finished, K groups of updated model parameters are obtained, and the loss function of the updated model on the query set is calculated.
E) Calculating the gradient of the loss function to the original model parameters, and carrying out secondary updating on the original model parameters.
5. And in the meta-training process, a query set is adopted to evaluate the training effect of the meta-learner. Namely, the element learner inputs the initial state, the terminal state, the constraint condition and the performance index of each track in the query set, outputs the predicted discrete optimal relative motion track point, and then carries out MSE operation on the predicted output of the element learner and the expected discrete track point in the query set.
6. And after the meta-training process of one type of track category is finished, continuing to perform meta-training on the track of the next type. Until meta-training of the trajectories for all classes is completed.
7. And in the fine adjustment process, the model parameters are fine adjusted by adopting a training set. In particular, the model parameters are updated once with each trajectory in the training set, similar to steps 4A), 4B), 4C), but excluding 4D).
8. And in the fine adjustment process, testing and evaluating the model by adopting a test set. Specifically, similar to step 5.
The invention solves the problem that a numerical method for planning the relative motion trail of the spacecraft needs a large amount of calculation time consumption: when trajectory planning is carried out by adopting numerical methods such as GPOPS and the like, the average time consumption is from dozens of minutes to dozens of minutes or even hours; the fine adjustment process of the new track planning by the method can be completed in only a few minutes.
The invention avoids the problem that the traditional machine learning method needs to rely on a large number of track samples: generally, only 5 tracks by 6 tracks or 30 tracks are needed to complete training, and the traditional machine learning method generally needs more than 300 tracks to train convergence. When the method faces to a new type of track planning task, the new track planning task can be quickly adapted only by a small amount of fine adjustment work, and the traditional machine learning method needs to retrain the new track after the new track is brought into a sample set.
The invention is funded by the subject of the open fund in the laboratory of the space intelligent control technology, and the subject number is as follows: HTKJ2020KL 502006.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
1. A spacecraft relative motion track planning method based on meta-learning is characterized in that an optimal control problem is solved through a numerical method to obtain optimal relative motion tracks under various working conditions, then training of the meta-learner is achieved through a meta-training process and a fine-tuning process, initial states, terminal states, constraint conditions and performance indexes of relative motion track planning tasks are input into the meta-learner, and predicted discrete optimal relative motion track curves are output.
2. The method as claimed in claim 1, wherein the meta-learner comprises a relative motion trajectory planning between two spacecrafts including a target spacecraft and a service spacecraft, and the trajectory is an approximate optimal approach trajectory for the service spacecraft to approach the target spacecraft from an initial relative position.
3. The spacecraft relative motion trajectory planning method based on meta-learning according to claim 2, characterized in that under the condition of setting the rotation angular velocity of the target spacecraft, for each initial relative state, a GPOPS method is adopted to complete the optimal approximate relative motion trajectory planning, and discrete optimal approximate relative motion trajectory points and corresponding discrete optimal control quantities are obtained; and changing the rotation angular velocity of the target spacecraft to finish the planning of the optimal approximate relative motion trajectory in all initial relative states under all the rotation angular velocities of the target spacecraft.
4. The spacecraft relative motion trajectory planning method based on meta-learning according to claim 3, characterized in that all trajectories obtained after planning are divided into a meta-training set and a meta-testing set according to a certain proportion; dividing the meta-training set into a support set and a query set according to a certain proportion; and dividing the meta-test set into a training set and a test set according to a certain proportion.
5. The spacecraft relative motion trajectory planning method based on meta-learning according to claim 4, characterized in that, in the meta-training process, updating of an original model is performed in a dual gradient updating manner to obtain a training model; and in the fine tuning process, fine tuning is carried out on the parameters of the training model by adopting the training set to obtain a fine tuning model.
6. The spacecraft relative motion trajectory planning method based on meta-learning according to claim 5, wherein the updating of the original model is performed in a dual gradient updating manner, and the method for obtaining the training model comprises the following steps:
1) in each track category, traversing all tracks in the category; then calculating a loss function of the original model on each track;
2) calculating the gradient of the loss function to the original model parameters;
3) updating the original model parameters for one time according to the calculated gradient;
4) after the updating of the original model parameters of each track in the category is finished, the updated model parameters are obtained, and the loss function of the updated model on the query set is calculated;
5) and calculating the gradient of the loss function to the original model parameters, and performing secondary updating on the original model parameters to obtain a training model.
7. The spacecraft relative motion trajectory planning method based on meta-learning of claim 6, wherein in the meta-training process, a query set is adopted to evaluate the training effect of the meta-learner.
8. The spacecraft relative motion trajectory planning method based on meta-learning according to claim 5, wherein the parameters of the training model are finely tuned by adopting a training set, and the method for obtaining the fine tuning model comprises the following steps:
a) in the track category of each training set, traversing all tracks in the category; then calculating a loss function of the training model on each track;
b) calculating the gradient of the loss function to the parameters of the training model;
c) and updating the parameters of the training model for one time according to the calculated gradient to obtain a fine tuning model.
9. The spacecraft relative motion trajectory planning method based on meta-learning according to claim 8, wherein in the fine tuning process, a test set is adopted to perform test evaluation on the fine tuning model.
10. A spacecraft relative motion trajectory planning method based on meta-learning according to claim 6 or 8, characterized in that model parameters are updated once by adopting a method including SGD and Adam.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011055523.8A CN112180726A (en) | 2020-09-29 | 2020-09-29 | Spacecraft relative motion trajectory planning method based on meta-learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011055523.8A CN112180726A (en) | 2020-09-29 | 2020-09-29 | Spacecraft relative motion trajectory planning method based on meta-learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112180726A true CN112180726A (en) | 2021-01-05 |
Family
ID=73946146
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011055523.8A Pending CN112180726A (en) | 2020-09-29 | 2020-09-29 | Spacecraft relative motion trajectory planning method based on meta-learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112180726A (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106444738A (en) * | 2016-05-24 | 2017-02-22 | 武汉科技大学 | Mobile robot path planning method based on dynamic motion primitive learning model |
| US20180293723A1 (en) * | 2017-04-11 | 2018-10-11 | Electronics And Telecommunications Research Institute | Apparatus and method for detecting anomaly in plant pipe using multiple meta-learning |
| US20190147298A1 (en) * | 2017-11-14 | 2019-05-16 | Magic Leap, Inc. | Meta-learning for multi-task learning for neural networks |
| CN109919299A (en) * | 2019-02-19 | 2019-06-21 | 西安交通大学 | A kind of meta learning algorithm based on meta learning device gradually gradient calibration |
| CN110288878A (en) * | 2019-07-01 | 2019-09-27 | 科大讯飞股份有限公司 | Adaptive learning method and device |
| CN110727271A (en) * | 2019-10-30 | 2020-01-24 | 北京科技大学 | Robot motion primitive determining method and device |
| CN111047085A (en) * | 2019-12-06 | 2020-04-21 | 北京理工大学 | Hybrid vehicle working condition prediction method based on meta-learning |
| CN111199458A (en) * | 2019-12-30 | 2020-05-26 | 北京航空航天大学 | Recommendation system based on meta-learning and reinforcement learning |
| CN111353610A (en) * | 2020-02-28 | 2020-06-30 | 创新奇智(青岛)科技有限公司 | Model parameter determination method and device, storage medium and electronic equipment |
| CN111523669A (en) * | 2020-03-06 | 2020-08-11 | 国网浙江省电力有限公司电力科学研究院 | Knowledge graph meta-information-based least-order relation model training method |
-
2020
- 2020-09-29 CN CN202011055523.8A patent/CN112180726A/en active Pending
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106444738A (en) * | 2016-05-24 | 2017-02-22 | 武汉科技大学 | Mobile robot path planning method based on dynamic motion primitive learning model |
| US20180293723A1 (en) * | 2017-04-11 | 2018-10-11 | Electronics And Telecommunications Research Institute | Apparatus and method for detecting anomaly in plant pipe using multiple meta-learning |
| US20190147298A1 (en) * | 2017-11-14 | 2019-05-16 | Magic Leap, Inc. | Meta-learning for multi-task learning for neural networks |
| CN111328400A (en) * | 2017-11-14 | 2020-06-23 | 奇跃公司 | Meta-learning for multi-task learning of neural networks |
| CN109919299A (en) * | 2019-02-19 | 2019-06-21 | 西安交通大学 | A kind of meta learning algorithm based on meta learning device gradually gradient calibration |
| CN110288878A (en) * | 2019-07-01 | 2019-09-27 | 科大讯飞股份有限公司 | Adaptive learning method and device |
| CN110727271A (en) * | 2019-10-30 | 2020-01-24 | 北京科技大学 | Robot motion primitive determining method and device |
| CN111047085A (en) * | 2019-12-06 | 2020-04-21 | 北京理工大学 | Hybrid vehicle working condition prediction method based on meta-learning |
| CN111199458A (en) * | 2019-12-30 | 2020-05-26 | 北京航空航天大学 | Recommendation system based on meta-learning and reinforcement learning |
| CN111353610A (en) * | 2020-02-28 | 2020-06-30 | 创新奇智(青岛)科技有限公司 | Model parameter determination method and device, storage medium and electronic equipment |
| CN111523669A (en) * | 2020-03-06 | 2020-08-11 | 国网浙江省电力有限公司电力科学研究院 | Knowledge graph meta-information-based least-order relation model training method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106969778B (en) | Path planning method for cooperative pesticide application of multiple unmanned aerial vehicles | |
| CN103440368B (en) | A kind of multi-model dynamic soft measuring modeling method | |
| CN103745273B (en) | Semiconductor fabrication process multi-performance prediction method | |
| CN105302096B (en) | Intelligent factory scheduling method | |
| CN111047085B (en) | Hybrid vehicle working condition prediction method based on meta-learning | |
| CN106779372A (en) | Based on the agricultural machinery dispatching method for improving immune Tabu search algorithm | |
| CN103577676A (en) | Grey weighting method for sewage treatment process comprehensive evaluation | |
| WO2021082394A1 (en) | Layout-variable taxiing-out time prediction system based on big data deep learning | |
| CN106249717A (en) | A kind of control method for coordinating based on the modeling of executor's saturated multi-agent system | |
| CN112395777B (en) | Engine calibration parameter optimization method based on automobile exhaust emission simulation environment | |
| CN113156900B (en) | Machining deformation control method based on meta reinforcement learning | |
| CN112131206A (en) | Multi-model database OrientDB parameter configuration automatic tuning method | |
| CN108198268A (en) | A kind of production equipment data scaling method | |
| CN112180726A (en) | Spacecraft relative motion trajectory planning method based on meta-learning | |
| CN106094751A (en) | The dispatching method of a kind of raw material and device | |
| Berneti | Design of fuzzy subtractive clustering model using particle swarm optimization for the permeability prediction of the reservoir | |
| CN108681824A (en) | Public transit system monthly plan based on heuritic approach is arranged an order according to class and grade algorithm | |
| CN110705132B (en) | Guidance control system performance fusion evaluation method based on multi-source heterogeneous data | |
| CN107798353A (en) | A kind of batch process monitoring data processing method | |
| CN117114220A (en) | Vehicle scheduling algorithm based on graph neural network | |
| CN111341102A (en) | Motion primitive library construction method and device and motion primitive connection method and device | |
| CN106295872A (en) | The data processing method of air station flight ripple based on maximum transfer machine meeting and device | |
| CN115859521A (en) | Neural network-based milling error reconstruction method and system | |
| CN102662832A (en) | Evaluation method of production process data correction software | |
| CN115630561A (en) | Automatic optimization method and device for neural network thermal load prediction model |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210105 |
|
| RJ01 | Rejection of invention patent application after publication |