CN117311390B - Intelligent combined guidance method for closed-loop tracking of aerospace shuttle aircraft - Google Patents
Intelligent combined guidance method for closed-loop tracking of aerospace shuttle aircraft Download PDFInfo
- Publication number
- CN117311390B CN117311390B CN202311463606.4A CN202311463606A CN117311390B CN 117311390 B CN117311390 B CN 117311390B CN 202311463606 A CN202311463606 A CN 202311463606A CN 117311390 B CN117311390 B CN 117311390B
- Authority
- CN
- China
- Prior art keywords
- information
- state
- aircraft
- prediction
- track
- 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
- 238000000034 method Methods 0.000 title claims abstract description 82
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 23
- 238000012549 training Methods 0.000 claims description 30
- 230000008569 process Effects 0.000 claims description 23
- 238000013528 artificial neural network Methods 0.000 claims description 22
- 238000011217 control strategy Methods 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 9
- 230000002265 prevention Effects 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 8
- 238000012360 testing method Methods 0.000 claims description 7
- 230000004888 barrier function Effects 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 6
- 230000008878 coupling Effects 0.000 claims description 4
- 238000010168 coupling process Methods 0.000 claims description 4
- 238000005859 coupling reaction Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 4
- 206010048669 Terminal state Diseases 0.000 claims description 3
- 230000015572 biosynthetic process Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000000926 separation method Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 230000003044 adaptive effect Effects 0.000 abstract description 11
- 238000012937 correction Methods 0.000 abstract description 6
- 230000006870 function Effects 0.000 description 45
- 230000008859 change Effects 0.000 description 9
- 239000010410 layer Substances 0.000 description 7
- 238000004088 simulation Methods 0.000 description 5
- 239000013598 vector Substances 0.000 description 5
- 230000004913 activation Effects 0.000 description 4
- 238000013459 approach Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000012614 Monte-Carlo sampling Methods 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000002203 pretreatment Methods 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000009966 trimming Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- 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
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the technical field of closed-loop tracking intelligent combined guidance of aerospace shuttle vehicles, and particularly relates to an aerospace shuttle vehicle closed-loop tracking intelligent combined guidance method, which comprises the following steps: s1: designing an LSTM aircraft track prediction algorithm; s2: establishing an RLV reentry segment error model and converting constraint control problems; s3: an aircraft tracking controller based on adaptive dynamic programming is designed. According to the invention, the prediction correction framework is adopted, the controller is designed by combining prediction and self-adaptive dynamic programming based on the LSTM method, so that the selected performance index function can be ensured to reach the optimum in a limited time domain, the optimum feedback guidance law is obtained, and the autonomy of the controller is improved.
Description
Technical Field
The invention belongs to the technical field of closed-loop tracking intelligent combined guidance of aerospace shuttle vehicles, and particularly relates to a closed-loop tracking intelligent combined guidance method of aerospace shuttle vehicles.
Background
The reusable carrier (RLV) refers to an aircraft which can freely go back and forth between the earth surface and a space orbit and has multiple purposes and repeated use, the necessary trend of fast, reliable and low-cost space access is realized in the future, the research hot spot in the current aerospace field is also realized, because the reentry section of the RLV has the factors of quick speed change, strong coupling, uncertainty of a model, external environment and the like, the design of a reentry section control system faces greater challenges, the optimization of the reentry section track of the RLV and the design of a guidance law are particularly critical for ensuring the controlled aircraft to safely and stably perform reentry flight, the goal of the optimization of the reentry track of the RLV is to realize the control of the flight track reaching a certain optimal target under the condition of meeting the condition constraint, the constraint of a control quantity and the like, the prediction correction guidance method is different from the standard track guidance, the prediction correction guidance method is not dependent on a reference track, the controller is firstly predicted according to the deviation of a predicted landing point and a desired landing point in the flight process, the controller is designed according to the deviation of the predicted landing point, the predicted landing point precision is not dependent on the initial reentry state, the initial state is more flexible and the development resistance of the initial reentry disturbance is increasingly strong, and the research direction of various countries are increasingly developed;
the track prediction of the aircraft is one of indispensable functional components in an intelligent flight system, in a complex game environment, the track of the aircraft is predicted in advance to provide a reference direction for a subsequent maneuver decision, and the track prediction is to estimate the track at the future moment according to a certain rule or method on the basis of the existing information;
from current research, the existing technologies can be divided into two categories: the former is widely applied, for example, in 2017, welsh festival of the Harbin industrial university and the like aiming at the periodic jump motion problem of hypersonic aircrafts, an extended Kalman filter combined with a Singer model is provided for carrying out state estimation, a target motion track is further recursively provided, in 2018, students such as Zhang Kai of an air force early warning college iteratively derive an aircraft model by constructing target motion and flight intention characteristics through Bayesian theory, and then the students realize track prediction through a Monte Carlo sampling method, wherein the method has good interpretability, but limited prediction precision and long prediction time only can be applied to certain specific scenes.
In summary, the rider in the prior art cannot ensure that the selected performance index function reaches the optimum in a limited time domain, and cannot effectively obtain the optimum feedback guidance law, so that the autonomy of the controller is low.
Disclosure of Invention
The invention aims to provide an aerospace shuttle aircraft closed-loop tracking intelligent combined guidance method, which adopts a prediction correction framework, performs prediction and self-adaptive dynamic programming based on an LSTM method to combine with a design controller, can ensure that a selected performance index function reaches the optimum in a limited time domain, obtains an optimal feedback guidance law and improves the autonomy of the controller.
The technical scheme adopted by the invention is as follows:
an aerospace shuttle vehicle closed loop tracking intelligent combined guidance method comprises the following steps:
s1: designing an LSTM aircraft track prediction algorithm;
s2: establishing an RLV reentry segment error model and converting constraint control problems;
s3: an aircraft tracking controller based on adaptive dynamic programming is designed.
Further, the step S1 includes the steps of:
s101: carrying out data preprocessing operation by combining system state data and error data of a distance end point at each time step to build an information data set and a database of the RLV track prediction problem;
s102: acquiring coupling information between the environment and the system through input data, constructing an LSTM network for state prediction, updating a predicted network weight by using a back propagation algorithm to obtain a predicted state model of the system, and realizing real-time prediction of a system state track of each time step;
s103: in the flight process, the obtained prediction model is utilized to continuously predict the flight end point, and the deviation between the predicted falling point and the expected end point is used as a control error and is input to a controller to adjust the control quantity.
Further, the data preprocessing in S101 includes information fusion and feature extraction.
Further, the method for establishing the information data set in S101 includes the following steps:
s10101: acquiring information, and acquiring various state information of the aircraft changing along with time by means of sensors, radars and the like;
s10102: preprocessing data by adopting a zero-mean standardized preprocessing method;
s10103: and constructing a training sample, and decomposing the track data into the training sample and the label.
Further, the constructing training samples includes the steps of:
the method comprises the steps of starting from a first track point in a data set, selecting the state information of the aircraft corresponding to the time of the first 20 track points downwards in time sequence to predict the state information of the next track point, wherein the state information of each time step is used as the input of corresponding cells of a neural network, selecting a separation interval as 1, starting from a second track point, and selecting a training sample by the same method.
Further, the step S102 includes the steps of:
s10201: taking state error information of the aircraft as input, and mapping the input data to a new space through an embedded function;
s10202: taking historical state information of the aircraft as input, and fusing error information with the historical state information of the aircraft;
s10203: predicting a future track of the aircraft according to the observed historical state information and the fusion information by utilizing the LSTM network;
s10204: and constructing a prediction model through sample data, carrying out normalization processing on the training samples and the prediction samples, inputting the training samples into a training network in the network model, adjusting the network structure according to the loss, and testing the network performance by using the test samples to obtain a prediction result.
Further, the information data set is 14-dimensional, and the state difference information comprising the current state and the terminal state of the aircraft comprises a ground center distance information difference, a longitude information difference, a latitude information difference, a speed information difference, a track angle information difference, a course angle information difference and a roll angle information.
Further, the step S2 includes the steps of:
and (3) establishing an RLV reentry section error model, designing a performance index function simultaneously reflecting formation errors, control quantity and collision prevention effect, converting a collision prevention problem in a scene into a constraint problem through a safety barrier function, and converting a collision prevention control problem into a stable control problem of an error system.
Further, the step S3 includes the steps of:
designing a control algorithm based on self-adaptive dynamic programming, constructing a judging network to approximate an optimal performance index function and solving an optimal control strategy, updating norms of all weights of the neural network by adopting a strategy gradient method, and finally obtaining the optimal control strategy by utilizing network output iteration
Further, the state of the RLV needs to strictly meet the following constraints:
1) Defining the state quantity x of the aircraft in the control algorithm, and meeting the starting point state condition x 0 And endpoint state condition x f ；
2) Affected by the performance of the aircraft, during reentry, a control quantity u is defined that satisfies a constraint u min ≤u≤u max The state quantity x satisfies the constraint x min ≤x≤x max Wherein u is min Represents the upper limit of the control quantity, u max Representing a control amount lower bound; wherein x is min Represent the upper bound of state quantity, x max Representing the state quantity lower bound.
The invention has the technical effects that:
according to the intelligent combined guidance method for the closed-loop tracking of the aerospace shuttle aircraft, a prediction correction framework is adopted, the controller is designed by combining prediction and self-adaptive dynamic programming based on an LSTM method, the problem of dimension disaster of a traditional dynamic programming control algorithm is solved, the guidance law is continuously and iteratively updated through learning, finally, the selected performance index function is ensured to be optimal in a limited domain, the optimal feedback guidance law is obtained, and the autonomy of the controller is improved.
Drawings
FIG. 1 is a block diagram of an aerospace shuttle closed-loop tracking intelligent combination guidance based on LSTM and adaptive dynamic programming of the present invention;
FIG. 2 is a diagram of an aircraft predictive model of an LSTM of the present invention;
FIG. 3 is a schematic of the LSTM of the present invention;
FIG. 4 is a graph of the predicted outcome of the state of the aircraft of the present invention;
FIG. 5 is a graph of the predicted absolute percent error change in aircraft state according to the present invention;
FIG. 6 is a graph of the state change of the tracking control based on adaptive dynamic programming of the present invention;
FIG. 7 is a graph of error variation for tracking control based on adaptive dynamic programming in accordance with the present invention;
FIG. 8 is a graph of the weight transformation of the aircraft evaluation neural network of the present invention;
FIG. 9 is a LSTM model parametric diagram of the present invention;
FIG. 10 is a diagram of initial conditions and end point constraints of an aircraft of the present invention.
Detailed Description
The present invention will be specifically described with reference to examples below in order to make the objects and advantages of the present invention more apparent. It should be understood that the following text is intended to describe only one or more specific embodiments of the invention and does not limit the scope of the invention strictly as claimed.
Example 1:
as shown in fig. 1, an aerospace shuttle vehicle closed-loop tracking intelligent combined guidance method comprises the following steps:
s1: designing an LSTM aircraft track prediction algorithm;
as shown in fig. 2 and 3, S1 includes the steps of:
s101: information data sets and databases of RLV track prediction problems are built through data preprocessing operations such as information fusion, feature extraction and the like by combining error data of system state data and distance end points of each time step;
in order to provide control directions for the controllers of the aircraft, it is first necessary to predict the state of the aircraft at a future time step, and when developing state predictions of the aircraft, future states and historical state information of the aircraft and state differences of the current state and the end state of the aircraft are related, so that predictions are to be made by analyzing the correlation of these features with future behavior based on a large amount of empirical data.
Defining the time step of observation as T obs The predicted time step is T pred Defining a historical state trajectory of an aircraft as o= { p t |t＝1,2,…,T obs P, where t Information representative of the status of the aircraft thereof at time t, comprising: (1) centroid distance information r, (2) longitude information θ, (3) latitude information(4) Speed information θ, (5) track angle information γ, (6) heading angle information χ, (7) roll angle information σ; defining the set of state differences of the current state and the end state of the aircraft as +.>The real future state information of the aircraft is represented by y= { p t |t＝T obs +1,T obs +2,…,T obs +T pred And } represents. Thus, the trajectory prediction problem can be described as:
Y＝f o (O,O e ) (1)
the goal of trajectory prediction is to find the slave set O, O e Mapping function f to set Y o So that the function can predict its future trajectory based on the historical trajectory of the aircraft.
Because the prediction method based on deep learning needs to train a network according to a large amount of data, firstly, an information data set of the state characteristics of the aircraft needs to be established, and the method can be divided into the following 3 steps:
s10101: acquisition information section: by means of sensors, radars and the like, various state information of the aircraft changing along with time is acquired, and a data set obtained by simulation of the method is 14-dimensional and comprises state characteristic information (ground pitch information, longitude information, latitude information, speed information, track angle information, course angle information and roll angle information) of the aircraft and state difference information (ground pitch information difference, longitude information difference, latitude information difference, speed information difference, track angle information difference, course angle information difference and roll angle information difference) of the current state and the end state of the aircraft.
S10102: data preprocessing: because of the magnitude difference between the data samples of the different state characteristic information of the aircraft, the data samples with larger magnitude are dominant in training the network, so that the convergence speed is slower and the accuracy is lower, and therefore, the data preprocessing is required. The invention adopts a zero-mean standardized pretreatment method, and the calculation formula is as follows:
wherein d i Is the original sample data of each characteristic information, d' i Is sample data after each feature information processing,is the average value of all sample data of each characteristic information, τ i Is the standard deviation of the entire sample data for each feature information, where i= {1, 2..14 }, represents each of the 14 dimensions of the dataset. I.e. the raw data of all data samples in one dimension is transformed into normalized data according to the mean and standard deviation of the state information of each dimension in the data set 14 dimensions. The data is unified to a specific interval by processing, so that the slow learning speed and low network accuracy caused by too large information difference of different data samples in the learning process are avoided.
S10103: the method for constructing the training sample comprises the following steps: trajectory prediction is a supervised learning problem that requires the decomposition of trajectory data into training samples and labels. Starting from the first track point in the data set, the state information of the next track point is predicted by selecting the state information of the aircraft corresponding to the time of the first 20 track points downwards in time sequence, wherein the state information of each time step is used as the input of the corresponding cells of the neural network. Then, in order to ensure the continuity of the samples over time, a separation interval of 1 is chosen, i.e. starting from the second trace point, training samples are chosen in the same way.
S102: and obtaining coupling information between the environment and the system through input data, constructing an LSTM network for state prediction, updating a predicted network weight by using a back propagation algorithm to obtain a predicted state model of the system, and realizing real-time prediction of a system state track of each time step.
After a data set is obtained, the mapping relation is learned by utilizing a deep learning mechanism according to an LSTM algorithm, the time sequence of the aircraft state information is considered, a time-based back propagation algorithm is designed, and the super parameters such as the sample iteration times, the learning rate and the like in the network training process are determined according to experience and offline experiments, so that the network weight is updated; finally, in the actual online prediction process, a trained aircraft state prediction network is adopted to realize real-time online prediction of the aircraft state, and the prediction process is shown in fig. 2-3.
First, data including aircraft state and error information is input into an embedded layer to extract the error information, thereby obtainingThen the historical state information data of the aircraft is extracted by a data processing layer and then is subjected to characteristic extraction and error information +.>And merging, wherein the merged data is used as the input of the LSTM layer, and the output of the merged data is the predicted track of the aircraft.
Here, S102 includes the steps of:
s10201: an embedding layer: status error information of aircraftAs input, the input data is mapped to the new space by embedding a function phi (). The mathematical expression is as follows:
in the method, in the process of the invention,is the predicted state error information of the aircraft at time step t. />Is a spatial representation of the state error information of the aircraft at time step t, W 1 Is the weight of the embedding function. Wherein, the definition of the embedding function phi (·) is as follows:
s10202: data processing layer: historical state information p of aircraft t As input, error information is usedAnd fusing the historical state information of the aircraft, wherein the following formula is as follows:
wherein, as follows, the ". Aldrich represents the Hadamard product of the matrix, W 2 Is the weight of the embedding function.Is the predicted status information of the aircraft at time step t. Thus, the information q is fused t Depending on both historical state information of the aircraft and error information of the aircraft state from the endpoint.
S10203: LSTM layer: based on the observed historical state information and fusion information q by using LSTM network t Predicting a future trajectory of the aircraft is shown by:
in dp t A derivative representing the state of the aircraft generated at time t; w (W) 3 Is the network weight of LSTM; w (W) 4 Is the weight of the embedding function. o (o) t And h t-1 The output and hidden states of LSTM, respectively, with the LSTM input being [ q ] t ,dp t-1 ]. Here dp t-1 The differential quantity of the data information is reflected, and the network considers the data change dynamics in the prediction process. The LSTM network structure is shown in fig. 3:
it can be seen that the inputs of the LSTM network at time t contain not only the input data x t Also contains the hidden state h from the previous moment t-1 ，x t And h t-1 Is left behindForgetting gate obtains the state C of discarding the previous layer of hidden cells by activating function t-1 And C t-1 The product of (2) becomes the cell state C at the time t t Is a part of (a); x is x t And h t-1 Determination of cell State C at time t by activation function at input Gate t Information to be updated, which becomes composition C t Is a part of the other part of the first part. Thus, cell state C at time t t Including useful information at the past time and new information at the present time. Finally, based on x t And h t-1 And determining an output result at an output gate of the cell state. Therefore, the output of the LSTM at the time t considers the useful information at the past time and the state at the current time, and the network structure is favorable for learning the relation between time series data, so that the processing capacity of the time series prediction problem is improved.
S10204: training network: the prediction model firstly performs normalization processing on the training samples and the prediction samples after the sample data construction is completed. And inputting the training samples into a network model to train the network, and adjusting the network structure according to the loss. And finally, testing the network performance by using a test sample to obtain a prediction result.
The network training process adopts a mean square error function as a loss function, namely:
wherein n represents the number of samples in batches during each training process,is the future state of the aircraft predicted by the LSTM model, Y k Is its true future state. The optimization goal of the deep neural network is to have the MSE approach 0. The process of deep neural network training is the process of updating the network weight. And updating the weight of the network by using a back propagation algorithm according to the loss function, and finally obtaining a prediction result through weight updating.
The final prediction result information of each state is expressed as%1) Information r of earth center distance ref (2) longitude information θ ref (3) latitude information(4) Speed information θ ref (5) track angle information γ ref (6) heading angle information χ ref (7) roll angle information sigma ref 。
S103: in the flight process, the obtained prediction model is utilized to continuously predict the flight end point, and the deviation between the predicted falling point and the expected end point is used as a control error and is input to a controller to adjust the control quantity.
S2: establishing an RLV reentry segment error model and converting constraint control problems;
the method for establishing the RLV reentry segment error model and converting the constraint control problem in S2 comprises the following steps:
and (3) establishing an RLV reentry section error model, designing a performance index function simultaneously reflecting formation errors, control quantity and collision prevention effect, and converting a collision prevention problem in a scene into a constraint problem through a safety barrier function, so that the collision prevention control problem is converted into an optimal stable control problem of an error system, and the safety of the system is ensured.
In the RLV reentry section, the influence of side force and earth rotation in the reentry process is ignored when the particle of the unpowered flight of the aircraft is considered and the earth is a rotational ellipsoid, and the sideslip angle is zero. The RLV reentry dynamics system is:
wherein, r, θ,v, γ, χ represent the earth's center distance, longitude, latitude, flight speed, track angle, and heading angle, respectively; sigma is the roll angle; m is the aircraft mass; g is gravity acceleration, g=μ g /r 2 Wherein mu g Is an attractive force parameter; l is lift, with L=q d SC L The method comprises the steps of carrying out a first treatment on the surface of the D is resistance, d=q d SC D Wherein S is the RLV aerodynamic reference area, q d Is dynamic pressure, q d ＝0.5ρv 2 The method comprises the steps of carrying out a first treatment on the surface of the ρ is the atmospheric density, there is +.>Wherein ρ is 0 R is the atmospheric density at sea level e The earth radius is the earth radius, and beta is a constant coefficient; lift coefficient C L And coefficient of resistance C D The function expressed as angle of attack α will be given in the subsequent simulations.
If the existing control amount is processed only, high-frequency buffeting is introduced and the convergence of the problem is affected, so that a new control variable needs to be introduced. Introducing new auxiliary control variable to realize decoupling of the control quantity from the state quantity, and enabling the new control quantity to be:
the RLV reentry segment dynamics model can be rewritten as:
in the method, in the process of the invention,is a new state quantity. The state matrix f (x') and the control matrix B are respectively:
B＝[0,0,0,0,0,0,1] T
defining the predicted state quantity of the next track given by the S1 final LSTM prediction model asError system for aircraft state and reference inputsThe following are provided:
where e represents the error in the current state of the aircraft and the target.
And (3) synthesizing the formula, the formula and the formula to obtain an RLV reentry segment error system:
in order to ensure safe and stable flight of the aircraft, the state and control quantity of the aircraft are required to meet constraint conditions, and the safety of the system is ensured. Based on the constraint, a constraint item based on a safety barrier function is designed, and a basis is provided for constraint requirements in subsequent performance index functions.
In order to ensure safe and stable flight of an aircraft, the state of an RLV needs to strictly meet some constraints:
1) Defining the state quantity x of the aircraft in the control algorithm, and meeting the starting point state condition x 0 And endpoint state condition x f 。
2) Affected by the performance of the aircraft, during reentry, a control quantity u is defined that satisfies a constraint u min ≤u≤u max The state quantity x satisfies the constraint x min ≤x≤x max Wherein u is min Represents the upper limit of the control quantity, u max Representing a control amount lower bound; wherein x is min Represent the upper bound of state quantity, x max Representing the state quantity lower bound.
For the state constraints of an aircraft, the aircraft state safety domain and state barrier function may be designed to:
security domain D x ：
D x ＝{x∈R n ∣x min ≤x≤x max } (14)
Obstacle function mu x ：
In the formula, when the state quantity of the aircraft at a certain moment is in the safety domain D x Otherwise, judging that the constraint condition is not satisfied, wherein eta is a positive real constant and satisfies eta > 1.
As can be seen, μ x Is a safety barrier function, passing function mu x Designing a related performance index function, when the state quantity of the aircraft approaches the safety domain D at a certain moment x When the performance index function is approaching infinity, the controller will change toward minimizing the performance index function at the next time, mu x The safety of the system state is ensured.
Similarly, for the control volume constraint condition of the aircraft, the safety domain of the control volume of the aircraft and the control volume obstacle function can be designed as follows:
security domain D u ：
D u ＝{u∈R n ∣u min ≤u≤u max } (16)
Obstacle function mu u ：
Next, define the performance index function J of the system as:
J＝∫ t ∞ U(e,u)dt (18)
wherein, the instantaneous performance index U (e, U) is defined as:
U(e,u)＝e T Qe+u T Ru+μ x e T e+μ u u T u (19)
wherein Q and R are positive oblique symmetry matrix.
It can be seen that the performance index function consists of four parts: first item e T Qe represents the aircraft state error cost, second term u T Ru represents the cost of the control quantity of the aircraft, and the third term mu x e T e represents the state constraint of the aircraft, the third term μ u u T u represents the control quantity constraint of the aircraft.
Then the optimum performance index function J * Can be expressed as:
in order to achieve smooth flight and constraints of the RLV reentry error system, the goal of the control is to find a set of systems that minimize the performance index function and limit the system state to the safe domain D x The control amount is limited to the security domain D u The control problem above can be written as:
s3: designing an aircraft tracking controller based on self-adaptive dynamic programming;
the aircraft tracking controller based on the adaptive dynamic programming is designed in the S3 and comprises the following steps:
designing a control algorithm based on self-adaptive dynamic programming, constructing a judging network to approximate an optimal performance index function and solving an optimal control strategy, updating norms of all weights of the neural network by adopting a strategy gradient method, and finally obtaining the optimal control strategy by utilizing network output iteration.
Because the designed performance index function is continuously available, the following Lyapunov equation can be obtained:
in the middle ofRepresents the bias of the performance index function J with respect to the error system e, and has J (0) =0. The Hamiltonian equation that can define a problem is：
When the performance index function is optimal, the Hamiltonian equation becomes the Hamiltonian-Jacobian-Belman equation, i.e
Optimum performance index function J * Obtained by solving the Hamiltonian-Jacobian-Belman equation above, and which has a unique solution, when solving J * When the method exists and is continuously available, the optimal control strategy can be obtained by solvingObtaining an aircraft control strategy that minimizes a performance index function
In order to solve the problem that the Hamiltonian-Jacobian-Bellman equation is difficult to solve in the practical application process, the invention approximates the optimal performance index function J by utilizing the principle of single-layer judging neural network approximation * Is the value of (1):
J * ＝W c T σ c (e)+ε c (26)
in which W is c ∈R n Is an ideal judgment neural network weight vector, sigma c (e)∈R n Represent the activation function of the judgment neural network epsilon c E R represents the approximation error of the judgment neural network. The bias derivative can be obtained by:
in the method, in the process of the invention,and->Indicating the bias of the activation function and approximation error, respectively. Substituting the above equation into the Hamiltonian equation can be obtained as:
in the formula e cH Is to judge the residual error generated by the neural network approximation, and in the framework of the adaptive dynamic programming, the fact that the ideal weight is unknown is considered, and the method is generally based on the estimated weight vectorEstablishing a judging neural network to approach an optimal performance index function, namely:
thus, the approximate hamiltonian amount may be as follows:
definition of the definitionWeight estimation error->The method comprises the following steps:
to adjust the critical evaluation neural network weight vectorDefining an error function->The error function is minimized by using a strategy gradient method, so that the judgment neural network weight adjustment rule is set as follows:
wherein alpha is c And > 0 is the learning rate of the judgment neural network. Thus, an ideal control strategy can be described as:
its approximations are shown as:
based on the three steps, the whole process of intelligent combined guidance based on LSTM and closed-loop tracking of the aerospace shuttle aircraft with adaptive dynamic programming is completed.
According to the technical scheme, a prediction correction framework is adopted, prediction and self-adaptive dynamic programming are carried out based on an LSTM method, a controller is designed in a combined mode, the problem of dimension disaster of a traditional dynamic programming control algorithm is solved, the guidance law is continuously and iteratively updated through learning, finally, the selected performance index function is ensured to be optimal in a limited domain, the optimal feedback guidance law is obtained, and the autonomy of the controller is improved.
Example 2:
in this embodiment, in order to verify the effectiveness of the algorithm provided by the present invention, the algorithm is integrated and designed in MATLAB/Simulink, and a simulation experiment is performed, and the main simulation process is as follows:
LSTM prediction model and network training parameter settings:
the configuration of the trajectory prediction model is shown in fig. 9. Wherein the input of the embedded network is the ground center distance information r, the longitude information theta and the latitude information of the aircraftThe output of the prediction network is the future position information of the aircraft, and the speed information theta, the track angle information gamma, the heading angle information χ, the roll angle information sigma and the 14-dimensional information formed by the corresponding errors of the final states.
The experiment is firstly carried out offline by using a large amount of track data to train the prediction network, and then the track is predicted by using a trained model in the actual online prediction process, so that the proposed prediction model needs shorter prediction time, and the real-time property of prediction is ensured. Experiments use historical 20 time step information to predict future information, i.e., T obs =20, T of future state information of the aircraft using 20 x 14 vector of overall state information of the aircraft as input each time pred *2*3 dimensional vectors are outputs. In this way, 42970 x 20 x 14 training sample amounts n are formed using the aircraft trajectory data. The LSTM model randomly selects 80% of the data set for training, retains the remaining 20% of the data set for testing, and then uses the 15% training set as the validation set, validating and trimming the network during training. The algorithm adopts single-step track prediction to carry out simulation experiment, namely T pred =1, the learning rate in the back propagation process is uniformly set to 0.01, and smooth interpolation processing is performed in order to prevent control fluctuation.
Adaptive dynamic programming tracking controller parameter settings:
the RLV base parameter settings are as follows: aircraft mass m= 104.305kg; gravitational parameter mu g ＝3.986×10 14 m 3 /s 2 The method comprises the steps of carrying out a first treatment on the surface of the Aircraft area s= 391.22m 2 The method comprises the steps of carrying out a first treatment on the surface of the Radius R of earth e ＝6.37×10 3 m, haipingSurface air density ρ 0 ＝0.00238kg/m 3 The method comprises the steps of carrying out a first treatment on the surface of the Lift coefficient C L -0.207+2.04 α, coefficient of resistance C D =0.0785-0.3529 α. System constant coefficient β= 1.3875 ×10 -4 ，k Q ＝9.44×10 -5 . Aircraft, restraint of productionAircraft initial conditions and endpoint constraints as shown in fig. 10, aircraft state quantity constraints: />
The neural network activation function isJudging the initial weight value W of the neural network c ＝[90,80,30,20,40,75,60,15] T Q=r=i, η=1.5, the learning rate of the nerve, the complex is α c ＝0.5。
In fig. 4, the predicted result of each state of the aircraft is shown to be a curve of time step change, and as can be seen from fig. 4, the prediction model is used for realizing accurate state prediction for the altitude, the speed, the latitude, the longitude, the track angle, the heading angle and the roll angle of the aircraft, so that the designed state prediction model based on the LSTM can realize accurate observation of the state of the aircraft.
In which, fig. 5 shows the absolute percentage error change curve between the state prediction and the real state of the aircraft, and as can be seen from fig. 5, the model errors predicted in different states are all below 6%, which verifies the accuracy of the prediction model.
As shown in FIG. 6, the state change curves of the aircraft in altitude, speed, latitude, longitude, track angle, course angle and roll angle are displayed, the state tracking is basically formed at about 500s, then stable tracking flight is kept, and the effectiveness and stability of the designed tracking control algorithm based on the adaptive dynamic programming are verified.
In the graph, fig. 7 shows a change curve of state errors of the aircraft in altitude, speed, latitude, longitude, track angle, course angle and roll angle, and after about 500s, the state errors of the aircraft gradually converge to 0, so that the validity of a designed tracking control algorithm based on adaptive dynamic programming is further verified.
Fig. 8 shows a process of changing the weight parameters of the aircraft evaluation neural network, and as can be seen from fig. 8, the weight parameters of the evaluation neural network are stably converged and approach to the corresponding optimal values in a limited time as time goes by.
The foregoing is merely a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention, which are intended to be comprehended within the scope of the present invention. Structures, devices and methods of operation not specifically described and illustrated herein, unless otherwise indicated and limited, are implemented according to conventional means in the art.
Claims (7)
1. An aerospace shuttle vehicle closed loop tracking intelligent combination guidance method is characterized in that: the method comprises the following steps: s1: designing an LSTM aircraft track prediction algorithm;
the step S1 comprises the following steps:
s101: carrying out data information fusion and feature extraction through the combination of system state data and error data of a distance end point at each time step, and constructing an information data set and a database of the RLV track prediction problem;
s102: acquiring coupling information between the environment and the system through input data, constructing an LSTM network for state prediction, updating a predicted network weight by using a back propagation algorithm to obtain a predicted state model of the system, and realizing real-time prediction of a system state track of each time step;
s103: in the flight process, the obtained prediction model is utilized to continuously predict the flight end point, and the deviation between the predicted falling point and the expected end point is used as a control error and is input to a controller to adjust the control quantity;
s2: establishing an RLV reentry segment error model and converting constraint control problems;
s3: designing an aircraft tracking controller based on self-adaptive dynamic programming;
the step S3 comprises the following steps:
designing a control algorithm based on self-adaptive dynamic programming, constructing a judging network to approximate an optimal performance index function and solving an optimal control strategy, updating norms of all weights of the neural network by adopting a strategy gradient method, and finally obtaining the optimal control strategy by utilizing network output iteration.
2. The aerospace shuttle closed-loop tracking intelligent combination guidance method according to claim 1, wherein the method comprises the following steps: the method for establishing the information data set in S101 includes the following steps:
s10101: acquiring information, and acquiring various state information of the aircraft changing along with time by means of a sensor and a radar;
s10102: preprocessing data by adopting a zero-mean standardized preprocessing method;
s10103: and constructing a training sample, and decomposing the track data into the training sample and the label.
3. An aerospace shuttle closed loop tracking intelligent combination guidance method according to claim 2, wherein: the construction training sample comprises the following steps:
the method comprises the steps of starting from a first track point in a data set, selecting the state information of the aircraft corresponding to the time of the first 20 track points downwards in time sequence to predict the state information of the next track point, wherein the state information of each time step is used as the input of corresponding cells of a neural network, selecting a separation interval as 1, starting from a second track point, and selecting a training sample by the same method.
4. The aerospace shuttle closed-loop tracking intelligent combination guidance method according to claim 1, wherein the method comprises the following steps: the step S102 includes the steps of:
s10201: taking state error information of the aircraft as input, and mapping the input data to a new space through an embedded function;
s10202: taking historical state information of the aircraft as input, and fusing error information with the historical state information of the aircraft;
s10203: predicting a future track of the aircraft according to the observed historical state information and the fusion information by utilizing the LSTM network;
s10204: and constructing a prediction model through sample data, carrying out normalization processing on the training samples and the prediction samples, inputting the training samples into a training network in the network model, adjusting the network structure according to the loss, and testing the network performance by using the test samples to obtain a prediction result.
5. The aerospace shuttle closed-loop tracking intelligent combination guidance method according to claim 1, wherein the method comprises the following steps: the information data set is 14 dimensions, the information data set comprises state characteristic information of an aircraft and state difference information of the current state and the terminal state of the aircraft, the state characteristic information of the aircraft comprises ground center distance information, longitude information, latitude information, speed information, track angle information, course angle information and roll angle information, and the state difference information of the current state and the terminal state of the aircraft comprises ground center distance information difference, longitude information difference, latitude information difference, speed information difference, track angle information difference, course angle information difference and roll angle information difference.
6. The aerospace shuttle closed-loop tracking intelligent combination guidance method according to claim 1, wherein the method comprises the following steps: the step S2 comprises the following steps:
and (3) establishing an RLV reentry section error model, designing a performance index function simultaneously reflecting formation errors, control quantity and collision prevention effect, converting a collision prevention problem in a scene into a constraint problem through a safety barrier function, and converting a collision prevention control problem into a stable control problem of an error system.
7. The aerospace shuttle closed-loop tracking intelligent combination guidance method according to claim 1, wherein the method comprises the following steps: the state of the RLV needs to strictly meet the following constraints: 1) Defining aircraft state quantities in control algorithmsSatisfying the starting state condition->And endpoint status condition->2) By the influence of the aircraft performance, during reentry, a control quantity is defined +.>Satisfy constraint->State quantity->Satisfy constraint->Wherein->Represents the upper limit of the control quantity->Representing a control amount lower bound; wherein->Represent the upper bound of state quantity->Representing the state quantity lower bound.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311463606.4A CN117311390B (en) | 2023-11-06 | 2023-11-06 | Intelligent combined guidance method for closed-loop tracking of aerospace shuttle aircraft |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311463606.4A CN117311390B (en) | 2023-11-06 | 2023-11-06 | Intelligent combined guidance method for closed-loop tracking of aerospace shuttle aircraft |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117311390A CN117311390A (en) | 2023-12-29 |
| CN117311390B true CN117311390B (en) | 2024-03-19 |
Family
ID=89297413
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311463606.4A Active CN117311390B (en) | 2023-11-06 | 2023-11-06 | Intelligent combined guidance method for closed-loop tracking of aerospace shuttle aircraft |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117311390B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110908396A (en) * | 2019-11-28 | 2020-03-24 | 天津大学 | Full-stage reentry return guidance method for reusable vehicle |
| CN111351488A (en) * | 2020-03-03 | 2020-06-30 | 南京航空航天大学 | Intelligent trajectory reconstruction reentry guidance method for aircraft |
| CN112114521A (en) * | 2020-07-30 | 2020-12-22 | 南京航空航天大学 | Intelligent prediction control entry guidance method for spacecraft |
| CN114510067A (en) * | 2022-01-28 | 2022-05-17 | 北京航空航天大学 | Approximate optimal guidance method for reusable aircraft |
| EP4261492A1 (en) * | 2022-04-13 | 2023-10-18 | Diehl Defence GmbH & Co. KG | Method for mid-ourse guidance of a missile which can be driven in thrust |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190184561A1 (en) * | 2017-12-15 | 2019-06-20 | The Regents Of The University Of California | Machine Learning based Fixed-Time Optimal Path Generation |
-
2023
- 2023-11-06 CN CN202311463606.4A patent/CN117311390B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110908396A (en) * | 2019-11-28 | 2020-03-24 | 天津大学 | Full-stage reentry return guidance method for reusable vehicle |
| CN111351488A (en) * | 2020-03-03 | 2020-06-30 | 南京航空航天大学 | Intelligent trajectory reconstruction reentry guidance method for aircraft |
| CN112114521A (en) * | 2020-07-30 | 2020-12-22 | 南京航空航天大学 | Intelligent prediction control entry guidance method for spacecraft |
| CN114510067A (en) * | 2022-01-28 | 2022-05-17 | 北京航空航天大学 | Approximate optimal guidance method for reusable aircraft |
| EP4261492A1 (en) * | 2022-04-13 | 2023-10-18 | Diehl Defence GmbH & Co. KG | Method for mid-ourse guidance of a missile which can be driven in thrust |
Non-Patent Citations (1)
| Title |
|---|
| 空天飞行器再入制导的预测控制;方炜等;宇航学报;20061130;第27卷(第6期);第1216-1222页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117311390A (en) | 2023-12-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Williams et al. | Information-theoretic model predictive control: Theory and applications to autonomous driving | |
| Wang et al. | Multivariate regression-based fault detection and recovery of UAV flight data | |
| Tang et al. | An improved kinematic model predictive control for high-speed path tracking of autonomous vehicles | |
| CN107563044B (en) | Four-rotor unmanned aerial vehicle path tracking control method based on online safety learning | |
| Zhang et al. | Trajectory generation by chance-constrained nonlinear MPC with probabilistic prediction | |
| CN109885883A (en) | A kind of control method of the unmanned vehicle transverse movement based on GK clustering algorithm model prediction | |
| Horn et al. | Neural network-based trajectory optimization for unmanned aerial vehicles | |
| Lin et al. | A gated recurrent unit-based particle filter for unmanned underwater vehicle state estimation | |
| Williams | Three-dimensional aircraft terrain-following via real-time optimal control | |
| Jin et al. | Development and validation of linear covariance analysis tool for atmospheric entry | |
| Jiang et al. | Implementation of human-like driver model based on recurrent neural networks | |
| Wu et al. | Learning-based fixed-wing UAV reactive maneuver control for obstacle avoidance | |
| Amirkhani et al. | Interval type-2 fuzzy cognitive map-based flight control system for quadcopters | |
| Hou et al. | A small UAV tracking algorithm based on AIMM-UKF | |
| Jardine et al. | Adaptive MPC using a dual fast orthogonal Kalman filter: Application to quadcopter altitude control | |
| CN117311390B (en) | Intelligent combined guidance method for closed-loop tracking of aerospace shuttle aircraft | |
| CN112161626B (en) | High-flyability route planning method based on route tracking mapping network | |
| Guan et al. | A new integrated navigation system for the indoor unmanned aerial vehicles (UAVs) based on the neural network predictive compensation | |
| Yuqi et al. | Time-varying parameters estimation with adaptive neural network EKF for missile-dual control system | |
| Hu et al. | Neural network-based robust control for hypersonic flight vehicle with uncertainty modelling | |
| Shi et al. | A-infinity control for path tracking with fuzzy hyperbolic tangent model | |
| Li et al. | Re-entry guidance method based on decoupling control variables and waypoint | |
| Gao et al. | Adaptive air-data estimation in wind disturbance based on flight data | |
| Wei et al. | Multi-sensor environmental perception and adaptive cruise control of intelligent vehicles using kalman filter | |
| Ghosh Roy et al. | Longitudinal aircraft parameter estimation using neuro-fuzzy and genetic algorithm based method |
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 |