CN115963855A - Unpowered reentry aircraft landing area prediction method based on deep learning - Google Patents
Unpowered reentry aircraft landing area prediction method based on deep learning Download PDFInfo
- Publication number
- CN115963855A CN115963855A CN202211731391.5A CN202211731391A CN115963855A CN 115963855 A CN115963855 A CN 115963855A CN 202211731391 A CN202211731391 A CN 202211731391A CN 115963855 A CN115963855 A CN 115963855A
- Authority
- CN
- China
- Prior art keywords
- aircraft
- deep learning
- lift
- angle
- boundary
- 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
- 238000013135 deep learning Methods 0.000 title claims abstract description 52
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000013461 design Methods 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 6
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 230000005251 gamma ray Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000004088 simulation Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000009827 uniform distribution Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
Images
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
- Feedback Control In General (AREA)
Abstract
The invention discloses a deep learning-based unpowered reentry aircraft landing area prediction method, and belongs to the technical field of aircraft navigation, guidance and control. Firstly, establishing a six-degree-of-freedom model for a certain aircraft, and estimating the aerodynamic coefficient of the aircraft based on an expansion state machine; and then, considering the influence of uncertainty of the aerodynamic coefficient of the aircraft, solving the boundary point of the range of the aircraft re-entering the landing area, and storing the estimated value of the aerodynamic coefficient of the aircraft and the data of the boundary point of the aircraft re-entering the landing area as an offline database. And finally, constructing a deep learning network, training the deep learning network through an off-line database, and completing the design of the unpowered reentry aircraft landing area prediction method based on deep learning. The invention improves the prediction accuracy and real-time performance of the landing zone of the unpowered reentry vehicle, and has universality and expansibility.
Description
Technical Field
The invention belongs to the technical field of aircraft navigation, guidance and control, and particularly relates to a deep learning-based unpowered reentry aircraft landing area prediction method.
Background
The unpowered reentry aircraft landing area refers to an area range where the reentry aircraft can successfully land when the reentry aircraft flies in the final guide section in an unpowered manner. The aircraft landing area defines the potential flight range of the aircraft in the final guidance segment, and can provide basis for flight mission planning and target point reselection, and particularly provides basis for selecting or changing the target point when the aircraft encounters an unexpected accident. Therefore, there is a need for the study of a method for calculating the landing zone of an unpowered reentry vehicle.
At present, researchers calculate the landing area of the aircraft through numerical values or analytic methods such as trajectory optimization, guidance law design, maximum and minimum resistance profile construction and the like. However, these methods have problems that the optimization calculation time is long, or the model needs to be greatly simplified, or a plurality of tracks need to be repeatedly calculated on line, and the like, and are not suitable for on-line application.
In addition, the methods cannot be well adapted to model uncertainty possibly existing in the aircraft, and the accuracy and the practicability of the aircraft landing area calculation are greatly reduced.
Disclosure of Invention
The invention provides a deep learning-based unpowered reentry aircraft landing area prediction method for accelerating the rapidity of aircraft online landing area calculation and improving the accuracy of landing area calculation when aircraft model uncertainty exists.
The method for indicating the falling area of the unpowered reentry aircraft based on deep learning comprises the following steps:
firstly, establishing a six-degree-of-freedom model for a certain aircraft;
the six-degree-of-freedom model comprises a kinematics and dynamics equation set of the center of mass of the aircraft and a kinematics and dynamics equation set of the motion of the aircraft around the center of mass;
the kinematic and kinetic equations for the motion of the aircraft center of mass can be expressed as:
wherein x, y, z represent the position of the aircraft in the ground coordinate system; θ represents a ballistic dip angle; psi v Representing ballistic declination; m represents the aircraft mass; l, D, Y represent lift, drag and lateral force, respectively; gamma ray v Representing the roll angle. The lift L, drag D, and lateral force Y are defined as follows:
wherein Q =0.5 ρ V 2 Represents a dynamic pressure; ρ is the air density; s is the aircraft reference area; c L ,C D ,C Y Respectively lift coefficient, drag coefficient and lateral force coefficient.
The system of kinematic and kinetic equations for the motion of an aircraft about the center of mass can be expressed as:
in the formula, alpha, beta, gamma v Respectively representing an attack angle, a sideslip angle and a roll angle; w is a mx ,w my ,w mz Roll, yaw and pitch velocities are represented, respectively; m x ,M y ,M z Respectively representing three-axis aerodynamic moment of the aircraft; i is x ,I y ,I z Respectively, representing the three-axis moment of inertia.
The invention considers the model uncertainty of the aircraft aerodynamic parameters, which can be expressed as:
in the formula eta i (i = L, D, Y) represents the uncertainty of the aerodynamic coefficient, characterized by a percentage, η i ∈[-30％,+30％]，η i Uniform distribution is obeyed.
Secondly, estimating the aerodynamic coefficient of the aircraft based on the expansion state machine and the six-degree-of-freedom model;
estimating the lift coefficient related term by constructing an expansion stater:
wherein z is 1α Is an estimate of the angle of attack α of the aircraft, e Lα Error of estimation of the angle of attack alpha of the aircraft, z 2α As a lift-related term
In combination with an evaluation value of>
g α ＝[-cosαtanβsinαtanβ1]，ω＝[w mx ,w my ,w mz ] T ，β 1α ＝2w 1α ,β 2α ＝w 1α 2 ，w 1α Is the bandwidth.
Estimating the drag coefficient related term by constructing an expansion state machine:
wherein z is 1V Is an estimate of the aircraft velocity V, e LV Is the estimated error of the aircraft speed V, z 2V As a resistance-related term f D Estimated value of = -D, f sV ＝-mgsinθ，β 1V ＝2w 1V ,β 2V ＝w 1V 2 ，w 1V Is the bandwidth.
Estimating the lateral force coefficient related term by constructing an expansion state machine:
wherein z is 1β Is an estimate of the aircraft sideslip angle β, e Yβ Is the estimated error of the aircraft sideslip angle beta, z 2β As a side force related term
Estimated value of g α ＝[sinαcosα0]，β 1β ＝2w 1β ,β 2β ＝w 1β 2 ，w 1β Is the bandwidth.
The estimates of lift, drag and side force coefficients can be expressed as:
wherein,
respectively representing estimates of the lift, drag and lateral force coefficients of the aircraft.
And step three, taking the influence of uncertainty of the aerodynamic coefficient of the aircraft into consideration, solving the boundary point of the range of the aircraft re-entering the falling area, and storing the estimated value of the aerodynamic coefficient of the aircraft and the data of the boundary point of the aircraft re-entering the falling area as an off-line database.
Normal flight stateThe aircraft re-entry landing zone can then be seen approximately as an ellipse, where A (X) A ,Z A ) Points indicate maximum longitudinal points, F (X) F ,Z F ) Points represent minimum longitudinal points, B (X) B ,Z B ) Dot, E (X) E ,Z E ) The points represent the positive and negative maximum transverse travel points, C (X), respectively C ,Z C ) Dot, D (X) D ,Z D ) The points represent the positive and negative nearest convolution points, respectively. By the pair A (X) A ,Z A ) Dot, B (X) B ,Z B ) Dot, C (X) C ,Z C ) Dot, D (X) D ,Z D ) Dot, E (X) E ,Z E ) Dot sum F (X) F ,Z F ) And enveloping the points to obtain the approximate reachable region boundary of the aircraft.
The method for solving the boundary point of the aircraft reentry landing area comprises the following steps:
(1) When the attack angle with the maximum lift-drag ratio is selected and the sideslip angle is selected to be 0, the aircraft can reach the maximum longitudinal stroke point A (X) A ,Z A ) Point;
(2) When the attack angle with the maximum lift-drag ratio is selected and the sideslip angle is selected to be the maximum value or the minimum value, the aircraft can reach the maximum positive and negative traverse point B (X) B ,Z B ) Dot sum E (X) E ,Z E ) Point; when the flight direction of the aircraft deviates from the initial speed direction by more than 90 degrees, the sideslip angle is set to zero, and then the aircraft flies along the direction vertical to the initial speed, so that the aircraft can fully utilize the aerodynamic force to achieve the purpose of increasing the transverse distance.
(3) When the attack angle with the minimum lift-drag ratio is selected and the sideslip angle is selected to be the minimum value or the maximum value, the aircraft can reach the nearest gyration point C (X) C ,Z C ) Or D (X) D ,Z D ) Point;
(4) When the attack angle with the minimum lift-drag ratio is selected and the sideslip angle is selected to be 0, the aircraft can reach the minimum longitudinal stroke point F (X) F ,Z F ) And (4) point.
And then, carrying out control law design on the aircraft based on the obtained attack angle and sideslip angle instructions, and carrying out integral iteration on a six-degree-of-freedom equation of the aircraft to obtain a point for representing the boundary of the aircraft re-entering the landing zone and the boundary range of the aircraft.
And finally, randomly taking values of the uncertainty of the aerodynamic parameters in a given range, solving a large number of points of the aircraft re-entering the boundary range of the landing zone in the process, and storing the estimated value of the aerodynamic coefficient of the aircraft and the data of the points of the aircraft re-entering the boundary of the landing zone as an off-line database.
And step four, constructing a deep learning network, training the deep learning network through an off-line database, and completing the design of the unpowered reentry aircraft landing area prediction method based on deep learning.
And a full-connection network is used as a basic structure to construct a deep neural network for the prediction of the unpowered reentry aircraft landing zone.
Selecting the current flight state (flight speed V and flight position information X) of the aircraft 0 ,Y 0 ,Z 0 ) And the estimation term of the aerodynamic coefficient of the aircraft by the extended state observer
As an input to the deep-learning network, i.e. an input of the deep-learning network->
Selecting coordinates of points A, B, C, D, E and F for representing the boundary range of the aircraft re-entering the landing point area as the output of the deep learning network, namely the output of the deep learning network is designed as follows:
the off-line database trains the deep learning network in the following process:
and (5) calculating the gradient according to the loss function by adopting an Adam optimizer, and updating the parameters of the fully-connected network. And finally obtaining a deep learning network for performing reachable region boundary prediction through sufficient training iteration.
The loss function of the deep learning network is designed into a mean square error form, namely the average value of the square of the difference between the estimated value of the trimmable predictive network and the true value of the sample in a batch of data:
in the formula, N represents the size of the batch data,
an estimate representing the output of the prediction network of the reachable region boundary.
And step five, inputting the current flight state of the aircraft and the aerodynamic coefficient estimation value thereof into the trained deep learning network to obtain the predicted landing zone boundary data.
The invention has the advantages that:
(1) A deep learning-based unpowered reentry aircraft landing area prediction method adopts a deep network and a deep learning method to replace the calculation of a large number of integral iteration processes, and improves the real-time performance of aircraft landing area prediction.
(2) A prediction method for a landing area of an unpowered reentry aircraft based on deep learning is characterized in that an extended state observer is introduced to estimate model uncertainty, and prediction accuracy of the landing area range of the aircraft is improved.
(3) The method adopts the idea of time-consuming iterative computation process fitting through a deep neural network, can be applied to unpowered final pilot segments, is feasible for regional prediction of powered aircrafts and other flight segments, and has universality and expansibility.
Drawings
FIG. 1 is a flow chart of steps of a method for indicating a landing zone of an unpowered reentry vehicle based on deep learning according to the present invention;
FIG. 2 is a schematic diagram of the boundary point of the landing zone range of the unpowered reentry vehicle of the present invention;
FIG. 3 is a schematic structural diagram of a deep learning network according to the present invention;
FIG. 4 is a schematic diagram of a deep learning network training method according to the present invention;
FIG. 5 is a diagram of a flight trajectory of an aircraft in an embodiment of the present invention;
FIG. 6 is a comparison graph of results of predicted boundary points and actual boundary points of an aircraft landing zone obtained by the method of the invention in an embodiment of the invention.
Detailed Description
In order to facilitate the understanding and implementation of the present invention for those of ordinary skill in the art, the present invention is further described in detail below with reference to the accompanying drawings and examples.
Firstly, estimating the uncertainty of an aircraft model by combining an expansion state machine; then, designing an attack angle and sideslip angle guidance instruction which can reach the boundary of the falling area, carrying out six-degree-of-freedom simulation, carrying out off-line calculation of points for representing the boundary of the falling area, and storing generated data; and finally, constructing a deep neural network and combining a deep learning method, carrying out network training aiming at data generated offline, and generating an unpowered reentry aircraft landing area prediction network for online application.
A method for predicting a landing area of an unpowered reentry aircraft based on deep learning is specifically divided into the following steps as shown in figure 1:
firstly, establishing a six-degree-of-freedom model for a certain aircraft;
the six-degree-of-freedom model comprises an aircraft centroid and a kinematic and kinetic equation set moving around the centroid;
the kinematic and kinetic equations for the motion of the aircraft center of mass can be expressed as:
wherein x, y, z represent the position of the aircraft in the ground coordinate system; θ represents a ballistic dip angle; psi v Representing ballistic declination; m represents the aircraft mass; l, D, Y represent lift, drag and lateral force, respectively; gamma ray v Representing the roll angle. The lift L, drag D, and lateral force Y are defined as follows:
wherein Q =0.5 ρ V 2 Represents a dynamic pressure; ρ is the air density; s is the aircraft reference area; c L ,C D ,C Y Respectively lift coefficient, drag coefficient and lateral force coefficient.
The set of kinematic and kinetic equations for the motion of an aircraft about the center of mass can be expressed as:
in the formula, alpha, beta, gamma v Respectively representing an attack angle, a sideslip angle and a roll angle; w is a mx ,w my ,w mz Roll, yaw and pitch velocities are represented, respectively; m is a group of x ,M y ,M z Respectively representing three-axis aerodynamic moment of the aircraft; i is x ,I y ,I z Respectively, representing the three-axis moment of inertia.
The invention considers the model uncertainty of the aircraft aerodynamic parameters, which can be expressed as:
in the formula eta i (i = L, D, Y) represents the uncertainty of the aerodynamic coefficient, characterized by a percentage, η i ∈[-30％,+30％]，η i Uniform distribution is obeyed.
Secondly, estimating the aerodynamic coefficient of the aircraft based on the expansion state machine and the six-degree-of-freedom model;
estimating the lift coefficient related term by constructing an expansion stater:
wherein z is 1α Is an estimate of the angle of attack α of the aircraft, e Lα Error of estimation of the angle of attack alpha of the aircraft, z 2α As a lift-related term
Is evaluated by the evaluation unit>
g α ＝[-cosαtanβsinαtanβ1]，ω＝[w mx ,w my ,w mz ] T ，β 1α ＝2w 1α ,β 2α ＝w 1α 2 ，w 1α Is the bandwidth.
Estimating the drag coefficient related term by constructing an expansion state machine:
wherein z is 1V Is an estimate of the aircraft speed V, e LV Is the estimated error of the aircraft speed V, z 2V As a resistance-related term f D Estimated value of = -D, f sV ＝-mgsinθ，β 1V ＝2w 1V ,β 2V ＝w 1V 2 ，w 1V Is the bandwidth.
Estimating the lateral force coefficient related term by constructing an expansion state machine:
wherein z is 1β Is an estimate of the aircraft sideslip angle β, e Yβ As aircraftError in estimation of sideslip angle β, z 2β As a side force related term
Estimated value of g α ＝[sinαcosα0]，β 1β ＝2w 1β ,β 2β ＝w 1β 2 ，w 1β Is the bandwidth.
The estimates of lift, drag and side force coefficients can be expressed as:
wherein,
respectively representing estimates of the lift, drag and lateral force coefficients of the aircraft.
And step three, considering the influence of uncertainty of the aerodynamic coefficient of the aircraft, solving the boundary point of the aircraft re-entering the landing area, and storing the estimated value of the aerodynamic coefficient of the aircraft and the data of the boundary point of the aircraft re-entering the landing area as an off-line database.
Under normal flight conditions, the aircraft re-entry landing area can be approximately seen as an ellipse, where A (X) A ,Z A ) Points represent maximum longitudinal points, F (X) F ,Z F ) The points represent the minimum course points, B (X) B ,Z B ) Dot, E (X) E ,Z E ) The points represent the positive and negative maximum transverse travel points, C (X), respectively C ,Z C ) Dot, D (X) D ,Z D ) The points represent the nearest convolution points in the positive and negative directions, respectively. By pair A (X) A ,Z A ) Dot, B (X) B ,Z B ) Point, C (X) C ,Z C ) Dot, D (X) D ,Z D ) Dot, E (X) E ,Z E ) Dot sum F (X) F ,Z F ) The points are enveloped to obtain the approximate reachable region boundary of the aircraft, and a schematic diagram is shown in fig. 2.
The method can be used for solving the boundary point of the aircraft reentry landing area as follows:
(1) When the attack angle with the maximum lift-drag ratio is selected and the sideslip angle is selected to be 0, the aircraft can reach the maximum longitudinal stroke point A (X) A ,Z A ) Point;
(2) When the attack angle with the maximum lift-drag ratio is selected and the sideslip angle is selected to be the maximum value or the minimum value, the aircraft can reach the maximum transverse travel point B (X) of the positive direction and the negative direction B ,Z B ) Dot sum E (X) E ,Z E ) Point; it should be noted that a range convolution may result when the aircraft flight direction deviates from the initial speed direction by more than 90 °. Therefore, the sideslip angle zero setting strategy is designed, once the included angle between the flight direction and the initial speed direction becomes 90 degrees, the sideslip angle is set to be zero, then the aircraft flies along the direction perpendicular to the initial speed, and the aircraft can fully utilize the aerodynamic force to achieve the purpose of increasing the range.
(3) When the attack angle with the minimum lift-drag ratio is selected and the sideslip angle is selected to be the minimum value or the maximum value, the aircraft can reach the nearest gyration point C (X) C ,Z C ) Or D (X) D ,Z D ) Point;
(4) When the attack angle with the minimum lift-drag ratio is selected and the sideslip angle is selected to be 0, the aircraft can reach the minimum longitudinal stroke point F (X) F ,Z F ) And (4) point.
And then, performing control law design on the aircraft based on the obtained attack angle and sideslip angle instructions, and performing integral iteration on a six-degree-of-freedom equation of the aircraft to obtain a point for representing the boundary of the aircraft re-entering the landing zone and the boundary range of the aircraft.
In addition, because the lift force and the resistance have obvious influence on the course, the influence of uncertainty of aerodynamic parameters is fully considered when the boundary point for representing the reentry landing area of the aircraft and the boundary range of the aircraft are obtained. The uncertainty of aerodynamic parameters is randomly taken within a given range, then a large number of points of the boundary range of the aircraft re-entering the landing area in the process are obtained, and the estimated value of the aerodynamic coefficient of the aircraft and the data of the boundary points of the aircraft re-entering the landing area are stored as an off-line database.
And step four, constructing a deep learning network, training the deep learning network through an off-line database, and completing the design of the unpowered reentry aircraft landing area prediction method based on deep learning.
And thirdly, when the boundary range of the aircraft is obtained, a large number of integral iteration processes are required to be carried out, the real-time prediction performance of the boundary range of the aircraft is influenced, and in order to further improve the prediction efficiency of the boundary range of the reachable area of the aircraft, the method further combines the strong fitting capacity of deep learning and utilizes the deep full-connection network to carry out fitting on the aircraft in the large number of iteration processes, so that the calculation real-time performance can be greatly improved. Meanwhile, model uncertainty estimated by using the extended state observer is considered in network input, and therefore the accuracy of boundary range prediction of the aircraft can be further improved.
The design of the unpowered reentry aircraft landing area prediction method based on deep learning comprises network structure design, network input and output design and network training method design.
And (3) network structure design: the construction of a deep neural network for the prediction of the unpowered reentry vehicle landing zone is carried out by adopting a fully-connected network as a basic structure, the network structure is shown in figure 3, the number of hidden layers in the invention is selected to be 5, and the number of nodes of each hidden layer is set to be 20.
Network input and output design:
the network input is selected as the current flight state (speed V and aircraft position information X, Y, Z) of the aircraft and the estimation item of the extended state observer to the aerodynamic coefficient
Thus, the network input is designed to be: />
The network output is selected as the coordinates of points A, B, C, D, E and F used to characterize the boundary range of the aircraft re-entry landing area. Thus, the network output is designed to be:
designing a network training method:
the training method is as shown in fig. 4, and the parameters of the reachable area boundary prediction network are updated by calculating the gradient according to the loss function by using an Adam optimizer. Through sufficient training iteration, a deep network capable of achieving area boundary prediction can be obtained finally.
The loss function of the reachable area boundary prediction network is designed into a mean square error form, namely the mean value of the square difference between the estimated value of the trimmable prediction network and the true value of the sample in batch data:
in the formula, N represents the size of the batch data,
an estimate representing the output of the prediction network of the reachable region boundary.
Example (b):
in order to check the effectiveness of the unpowered reentry aircraft landing zone prediction method based on deep learning, the embodiment takes a certain aircraft as an embodiment to perform simulation verification.
The initial state of the aircraft in this embodiment is set as: height h 0 Speed V of =20km 0 ＝1275m/s，α＝β＝γ v ＝0°，w mx ＝w my ＝w mz =0 °/s. Bandwidth of the extended state observer: w is a 1α ＝3,w 1V ＝1,w 1β And =25. The model uncertainty is set as: eta i ＝30％(i＝x,y,z)。
The comparison between the unpowered reentry aircraft landing zone prediction method based on deep learning and the result obtained by integral iteration in the offline database is shown in fig. 5 and 6, so that the error between the predicted boundary point and the real boundary point of the reentry aircraft landing zone obtained by the method is very small, and the accuracy of the online prediction method provided by the invention is high.
For quantitative explanation, the error of the prediction result of the boundary point of the aircraft re-entering the landing area is given in table 1, and it can be found that the errors of 8 boundary points are all less than 0.5km/0.5km, so that higher prediction accuracy is achieved.
TABLE 1 accuracy statistical table of boundary prediction results of reachable regions
In addition, in this embodiment, it is counted that the calculation time consumption of the proposed prediction method is 10 -3 s, which is tens of seconds in the conventional method, and embodies the real-time property of the method of the present invention.
By combining the simulation verification of the embodiment, the real-time performance and the accuracy of the unpowered reentry aircraft landing area prediction method based on deep learning are proved.
Those skilled in the art will appreciate that the invention may be practiced without these specific details.
Claims (5)
1. A deep learning-based unpowered reentry aircraft landing area prediction method is characterized by comprising the following steps:
the method comprises the following steps of firstly, establishing a six-degree-of-freedom model for a certain aircraft, and estimating aerodynamic coefficients of the aircraft by constructing an expansion state machine;
the aircraft aerodynamic coefficients include lift, drag and lateral forces, and the estimated values of the coefficients are expressed as:
wherein,
respectively representing estimated values of lift, drag and lateral force coefficients of the aircraft; z is a radical of formula 2α As a lift-related term
L is lift, m represents aircraft mass, V is aircraft speed, β is aircraft sideslip angle, Q =0.5 ρ V 2 Represents a dynamic pressure; ρ is the air density; s is the aircraft reference area; z is a radical of formula 2V As a resistance-related term f D An estimate of = -D, D being resistance; z is a radical of 2β Is a side force related item->
Y is the lateral force;
step two, considering the influence of uncertainty of the aerodynamic coefficient of the aircraft, solving the boundary point of the range of the aircraft re-entering the landing area, and storing an estimated value of the aerodynamic coefficient of the aircraft and data of the boundary point of the aircraft re-entering the landing area as an off-line database;
the method for solving the boundary point of the aircraft re-entering landing area comprises the following steps:
(1) When the attack angle with the maximum lift-drag ratio is selected and the sideslip angle is selected to be 0, the aircraft can reach the maximum longitudinal stroke point A (X) A ,Z A ) Point;
(2) When the attack angle with the maximum lift-drag ratio is selected and the sideslip angle is selected to be the maximum value or the minimum value, the aircraft can reach the maximum positive and negative traverse point B (X) B ,Z B ) Dot sum E (X) E ,Z E ) Point; after the flight direction of the aircraft deviates from the initial speed direction by more than 90 degrees, the sideslip angle is set to zero, and then the aircraft flies along the direction vertical to the initial speed;
(3) When the attack angle with the minimum lift-drag ratio is selected and the sideslip angle is selected to be the minimum value or the maximum value, the aircraft can reach the nearest gyration point C (X) C ,Z C ) Or D (X) D ,Z D ) Point;
(4) When the attack angle with the minimum lift-drag ratio is selected and the sideslip angle is selected to be 0, the aircraft can reach the minimum longitudinal stroke point F (X) F ,Z F ) Point;
then, carrying out control law design on the aircraft based on the obtained attack angle and sideslip angle instructions, and carrying out integral iteration on a six-degree-of-freedom equation of the aircraft to obtain a point for representing the reentry of the aircraft into the landing zone boundary and the boundary range of the aircraft;
finally, randomly taking values of the uncertainty of the aerodynamic parameters in a given range, solving a large number of points of the boundary range of the aircraft re-entering the landing zone in the process, and storing the estimated value of the aerodynamic coefficient of the aircraft and the data of the boundary points of the aircraft re-entering the landing zone as an off-line database;
constructing a deep learning network, training the deep learning network through an off-line database, and completing the design of the unpowered reentry aircraft landing area prediction method based on deep learning;
selecting an estimation item of the current flight state of the aircraft and the aerodynamic coefficient of the aircraft by the extended state observer
As an input to the deep-learning network, i.e. an input of the deep-learning network->
X 0 ,Y 0 ,Z 0 And V is the position information and the speed information of the current aircraft respectively;
selecting coordinates of points A, B, C, D, E and F for representing the boundary range of the aircraft re-entering the landing point area as the output of the deep learning network, namely the output of the deep learning network is designed as follows:
the off-line database trains the deep learning network in the following process:
calculating a gradient according to a loss function by adopting an Adam optimizer, and updating parameters of the fully-connected network; through sufficient training iteration, a deep learning network for achieving area boundary prediction is finally obtained;
and step four, inputting the current flight state of the aircraft and the aerodynamic coefficient estimation value thereof into the trained deep learning network to obtain the predicted landing zone boundary data.
2. The method for predicting the landing zone of the unpowered reentry vehicle based on the deep learning of claim 1, wherein the six-degree-of-freedom model comprises a system of kinematics and dynamics equations of the center of mass of the vehicle and a system of kinematics and dynamics equations of the motion of the vehicle around the center of mass;
the kinematic and kinetic equations for the motion of the aircraft center of mass can be expressed as:
wherein x, y, z represent the position of the aircraft in the ground coordinate system; θ represents the ballistic dip; psi v Representing ballistic declination; m represents the aircraft mass; l, D, Y represent lift, drag and lateral force, respectively; gamma ray v Representing the roll angle;
the lift L, drag D, and lateral force Y are defined as follows:
wherein Q =0.5 ρ V 2 Represents a dynamic pressure; ρ is the air density; s is the aircraft reference area; c L ,C D ,C Y Respectively a lift coefficient, a drag coefficient and a lateral force coefficient;
the set of kinematic and kinetic equations for the motion of an aircraft about the center of mass can be expressed as:
in the formula, alpha, beta, gamma v Respectively representing an attack angle, a sideslip angle and a heeling angle; w is a mx ,w my ,w mz Roll, yaw and pitch velocities are represented, respectively; m x ,M y ,M z Respectively representing three-axis aerodynamic moment of the aircraft; i is x ,I y ,I z Respectively, representing the three-axis moment of inertia.
3. The method for predicting the landing zone of the unpowered reentry vehicle based on the deep learning as claimed in claim 1, wherein the estimating of the aerodynamic coefficient of the vehicle is performed by constructing an expansion state machine, specifically:
estimating the lift coefficient related term by constructing an expansion stater:
wherein z is 1α Is an estimate of the angle of attack α of the aircraft, e Lα Error of estimation of the angle of attack alpha of the aircraft, z 2α As a lift-related term
Is evaluated by the evaluation unit>
g α ＝[-cosαtanβ sinαtanβ 1]，ω＝[w mx ,w my ,w mz ] T ，β 1α ＝2w 1α ,β 2α ＝w 1α 2 ，w 1α Is the bandwidth;
estimating the drag coefficient related term by constructing an expansion state machine:
wherein z is 1V Is an estimate of the aircraft speed V, e LV Is the estimated error of the aircraft speed V, z 2V As a resistance-related term f D Estimated value of = -D, f sV ＝-mgsinθ，β 1V ＝2w 1V ,β 2V ＝w 1V 2 ，w 1V Is the bandwidth;
estimating the lateral force coefficient related term by constructing an expansion state machine:
wherein z is 1β Is an estimate of the aircraft sideslip angle β, e Yβ Is the estimated error of the aircraft sideslip angle beta, z 2β As a side force related term
Estimate of (a), g α ＝[sinαcosα0]，β 1β ＝2w 1β ,β 2β ＝w 1β 2 ，w 1β Is the bandwidth.
4. The deep learning-based unpowered reentry vehicle landing zone prediction method according to claim 1, wherein a full-connection network is adopted as a basic structure of the deep learning network.
5. The method as claimed in claim 1, wherein the loss function of the deep learning network is designed as a mean square error, which is an average of squares of differences between an estimated value of the trimmable predictive network and a true value of the sample in a batch:
in the formula, N represents the size of the batch data,
an estimate representing the output of the prediction network of the reachable region boundary. />
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211731391.5A CN115963855A (en) | 2022-12-30 | 2022-12-30 | Unpowered reentry aircraft landing area prediction method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211731391.5A CN115963855A (en) | 2022-12-30 | 2022-12-30 | Unpowered reentry aircraft landing area prediction method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115963855A true CN115963855A (en) | 2023-04-14 |
Family
ID=87363047
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211731391.5A Pending CN115963855A (en) | 2022-12-30 | 2022-12-30 | Unpowered reentry aircraft landing area prediction method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115963855A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116702439A (en) * | 2023-05-19 | 2023-09-05 | 北京理工大学 | Reentry landing point prediction method of simulation aircraft based on reentry landing point prediction model |
-
2022
- 2022-12-30 CN CN202211731391.5A patent/CN115963855A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116702439A (en) * | 2023-05-19 | 2023-09-05 | 北京理工大学 | Reentry landing point prediction method of simulation aircraft based on reentry landing point prediction model |
| CN116702439B (en) * | 2023-05-19 | 2024-02-13 | 北京理工大学 | Reentry landing point prediction method of simulation aircraft based on reentry landing point prediction model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110413000B (en) | Hypersonic aircraft reentry prediction correction fault-tolerant guidance method based on deep learning | |
| CN111351488B (en) | Intelligent trajectory reconstruction reentry guidance method for aircraft | |
| CN108227491B (en) | Intelligent vehicle track tracking control method based on sliding mode neural network | |
| CN111813146B (en) | Reentry prediction-correction guidance method based on BP neural network prediction voyage | |
| CN104778376B (en) | A kind of hypersonic gliding bullet Skipping trajectory Forecasting Methodology of near space | |
| CN114355777B (en) | Dynamic gliding method and system based on distributed pressure sensor and sectional attitude control | |
| CN114510063B (en) | Unmanned tracked vehicle and track tracking control method and system thereof | |
| CN106371312A (en) | Lifting reentry prediction-correction guidance method based on fuzzy controller | |
| CN105759609B (en) | A kind of carrier-borne aircraft autonomous landing on the ship method based on explicit non-linear Model Predictive Control | |
| CN112455720B (en) | Aerospace vehicle aerodynamic force auxiliary orbit transfer design method | |
| CN109657256A (en) | A kind of high-precision ballistic reenters nominal return trajectory emulation mode | |
| CN115963855A (en) | Unpowered reentry aircraft landing area prediction method based on deep learning | |
| Gao et al. | Dubins path‐based dynamic soaring trajectory planning and tracking control in a gradient wind field | |
| CN114460936A (en) | Off-line incremental learning-based automatic driving automobile path planning method and system | |
| CN114370793A (en) | Rocket sublevel return and vertical landing guidance method | |
| CN114637292A (en) | Vehicle track tracking robust control method and system considering obstacle avoidance | |
| Klein et al. | Moving target approach for wind-aware flight path generation | |
| CN114138022A (en) | Distributed formation control method for unmanned aerial vehicle cluster based on elite pigeon swarm intelligence | |
| CN116822362B (en) | Unmanned aerial vehicle conflict-free four-dimensional flight path planning method based on particle swarm optimization | |
| CN111006694B (en) | Design method of long-endurance inertial navigation system track generator based on track planning | |
| CN110989397B (en) | Aircraft accident search simulation method and system | |
| CN112161626B (en) | High-flyability route planning method based on route tracking mapping network | |
| CN111857112A (en) | Automobile local path planning method and electronic equipment | |
| Cheng et al. | Cross-cycle iterative unmanned aerial vehicle reentry guidance based on reinforcement learning | |
| CN117170230B (en) | Multi-mode track generation method for vertical take-off and landing unmanned aerial vehicle group |
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 |