A morphing aircraft predictive correction guidance method based on deep learning

Abstract

The application provides a morphing aircraft predictive correction guidance method based on deep learning, and belongs to the field of aircraft guidance and control. The method first trains a flight performance prediction model of the morphing aircraft, wherein the input of the model is the speed, guidance parameters and morphing parameters of the morphing aircraft, and the output is the range capability value and the cross-range capability value; in each guidance period, based on the prediction model, a nonlinear optimization problem is constructed with the cross-range capability as the optimization performance index and the required range as the constraint, and is solved to obtain the current guidance parameters and morphing parameters of the morphing aircraft, and then the current attack angle and the amplitude of the roll angle are obtained, so that the predictive correction guidance of the morphing aircraft is realized. The application can make the guidance decision of the morphing aircraft in flight, which takes into account the morphing instruction and the attack angle and roll angle instruction, so that the morphing aircraft can maximize the cross-range under the premise of reaching the target point, and the cross-range maneuvering capability of the morphing aircraft can be effectively improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of aircraft guidance and control technology, and particularly relates to a morphing aircraft predictive correction guidance method based on deep learning. BACKGROUND

[0002] A morphing aircraft is a kind of aircraft that changes the wing type, wing span and other structures during flight. Unlike traditional aircraft that can only be used for a single purpose in a certain environment, a morphing aircraft can change its shape according to the needs of multiple factors such as environment, profile and combat mission. Through morphing, the aircraft can effectively increase the range and reduce fuel consumption, can adapt to wide-speed-range and long-range flight missions, and has better task adaptability. Reentry flight refers to the process of an aircraft returning to the earth's atmosphere and accurately reaching the predetermined position, which faces strict force and thermal load environmental conditions. In order to better exert the flight performance of the morphing aircraft and meet the flight constraints, a reentry guidance law suitable for the morphing aircraft needs to be designed.

[0003] Existing methods mainly focus on the trajectory planning method of the morphing aircraft, and there are few reentry predictive correction guidance methods for the morphing aircraft. The predictive correction guidance method of the conventional aircraft cannot be directly applied to the reentry guidance of the morphing aircraft because it does not consider the influence of morphing on the aerodynamic performance. In order to comprehensively consider the influence of morphing factors and guidance instructions on the flight mission, it is crucial to design a morphing aircraft predictive correction guidance method. SUMMARY

[0004] The purpose of the present application is to overcome the shortcomings of the prior art and provide a morphing aircraft predictive correction guidance method based on deep learning. The present application can make a guidance decision that takes into account the morphing instructions and the angle of attack and bank angle instructions in real time during the flight of the morphing aircraft, so that the morphing aircraft can maximize the lateral range under the premise of reaching the target point, effectively improve the lateral maneuverability of the morphing aircraft, and thus improve the maneuverability and anti-interception ability of the morphing aircraft.

[0005] The present application provides a morphing aircraft predictive correction guidance method based on deep learning, which comprises the following steps:

[0006] A morphing aircraft flight performance prediction model is trained, the input of the model is the speed, guidance parameters and morphing parameters of the morphing aircraft, and the output is the range capability value and lateral range capability value of the morphing aircraft;

[0007] In each guidance period, based on the morphing aircraft flight performance prediction model, a nonlinear optimization problem is constructed with the lateral range capability as the optimization performance index and the required range as the constraint;

[0008] Solving the nonlinear optimization problem, obtaining the current guidance parameters and the deformation parameters of the morphing aircraft, and further obtaining the current attack angle and the amplitude of the roll angle of the morphing aircraft, so as to realize the predictive correction guidance of the morphing aircraft.

[0009] In a specific embodiment of the present application, the morphing aircraft flight performance prediction model adopts a neural network model with four full connection layers, the first layer of the model adopts an elu function, and the second layer and the third layer adopt a tanh function.

[0010] In a specific embodiment of the present application, the method further comprises:

[0011] Based on the relationship between the attack angle of the morphing aircraft and the speed, the roll angle corridor of the morphing aircraft under each aerodynamic configuration is established; the specific steps are as follows:

[0012] 1) Establish the relationship between the attack angle of the morphing aircraft and the speed;

[0013] Wherein, the expression of the attack angle of the morphing aircraft about the speed during flight is:

[0014]

[0015] Wherein, α is the attack angle of the morphing aircraft, V is the speed of the morphing aircraft; V b ,V a are respectively preset speed segmentation points, V b is less than the initial speed of the morphing aircraft, V a is greater than the terminal speed V f of the reentry flight of the morphing aircraft; is the attack angle value with the maximum lift-drag ratio; α m is the initial value of the attack angle;

[0016] 2) Establish the roll angle corridor of the morphing aircraft under each aerodynamic configuration;

[0017] Wherein, the reentry flight of the morphing aircraft under each aerodynamic configuration is divided into an initial descent segment, a balanced flight segment and a terminal capability management segment; for any aerodynamic configuration:

[0018] The initial descent segment contains from the reentry point to the first valley bottom point of the trajectory, and the speed range is V≥V b ; the expression of the upper limit ν 1,max and the lower limit ν 1,min of the roll angle corridor of the initial descent segment is:

[0019]

[0020] Wherein, k is the deformation coefficient, representing the deformation amount of the morphing aircraft; ν initthe constant bank angle taken for the initial descent segment;

[0021] the minimum bank angle of the balanced flight segment is 0, and the velocity range is [V a ,V b ], the balanced flight segment contains three process constraints of overload constraint, dynamic pressure constraint and heat flow constraint, and the expressions are as follows:

[0022]

[0023] wherein, respectively represent the heat flow, overload and dynamic pressure borne by the morphing aircraft at time t, respectively represent the maximum heat flow, maximum dynamic pressure and maximum overload constraint that the morphing aircraft can bear; S ref ,m0,g0 respectively represent the reference area, mass and gravitational acceleration of the morphing aircraft; C L (α,V,k),C D (α,V,k) respectively represent the lift coefficient and drag coefficient of the morphing aircraft; ρ is the atmospheric density at the current altitude of the morphing aircraft; C1 is a heat flow calculation constant; V c is the first cosmic speed; ρ0 is the atmospheric density at sea level; R d represents the curvature radius of the leading edge of the morphing aircraft;

[0024] then the air density of the minimum geocentric distance corresponding to each constraint in formula (3) is:

[0025]

[0026] wherein, r Q ,r p ,r n respectively represent the minimum geocentric distance corresponding to the heat flow constraint, the overload constraint and the dynamic pressure constraint;

[0027] take the minimum value of r Q ,r n ,r p in formula (4) as the minimum flight geocentric distance, and the expression is as follows:

[0028] r min (V)=min{r Q ,r q ,r n} (5)

[0029] then calculate the maximum bank angle of the balanced flight segment according to the quasi-balanced flight condition:

[0030]

[0031] wherein, g is the gravitational acceleration at the altitude of the morphing aircraft, ν2,max (V, k) represents the maximum roll angle calculated according to the quasi-equilibrium glide condition;

[0032] then the upper bound of the roll angle corridor of the equilibrium glide segment is 2,max and the lower bound of the roll angle corridor of the equilibrium glide segment is 2,min is expressed as:

[0033]

[0034] The terminal energy management segment satisfies the terminal state constraint, and the speed range is V ∈ [V f , V a ];

[0035] In combination with the quasi-equilibrium glide condition, the terminal equilibrium glide roll angle v QEGC,f is calculated by the expression:

[0036]

[0037] where r f , v f respectively represent the terminal geocentric distance and the terminal speed of reentry; the expression of the upper bound of the roll angle corridor of the terminal energy management segment v 3,max (V, k) and the lower bound of the roll angle corridor of the terminal energy management segment v 3,min (V, k) is:

[0038]

[0039] where k ν is the contraction coefficient.

[0040] In one specific embodiment of the present application, the method further comprises:

[0041] Based on the roll angle corridor of the morphing aircraft in each aerodynamic configuration, a training set for training the flight performance prediction model of the morphing aircraft is established, and the specific steps are as follows:

[0042] 1) Based on the value range of the guidance parameter ω and the morphing coefficient k, the flight trajectory simulation results of the morphing aircraft are obtained under different combinations of the guidance parameter ω and the morphing coefficient k, and the trajectory information is sampled;

[0043] Wherein, in the simulation of each flight trajectory, according to the guidance parameter ω and the roll angle corridor, the roll angle v = |v cmd | is obtained:

[0044] |v cmd | = (1-ω) v min (k) + ωv max (k) (10)

[0045] where ω is in the range [0, 1], and v max (k),ν min (k) are the upper and lower bounds of the roll angle corridor corresponding to the aerodynamic configuration with deformation coefficient k, respectively;

[0046] According to the flight speed and formula (1), the attack angle a is calculated;

[0047] Using the roll angle and the attack angle for simulation, the flight trajectory simulation results under the corresponding ω and k combination are obtained; then, by sampling, the basic trajectory information sequence of the flight trajectory is recorded, and the trajectory information includes: the height r i , the longitude l i , the latitude f i , the speed V i , the trajectory inclination angle q i , the trajectory deflection angle s i and the flight time t i , wherein the subscript i represents the i-th sampling point;

[0048] 2) Based on the trajectory information obtained in step 1), the flight performance indicators of each sampling point on each flight trajectory are calculated, including: the flight range capability of the flight vehicle and the flight lateral range capability of the flight vehicle;

[0049] wherein the flight range capability of the i-th sampling point is calculated as follows:

[0050]

[0051] wherein DR i represents the flight range capability of the i-th sampling point, S i represents the flight range of the i-th sampling point, and S f represents the distance flown by the flight vehicle along the entire trajectory;

[0052] Let the coordinate difference between the i-th sampling point and the terminal point in the trajectory coordinate be:

[0053]

[0054] wherein l f , f f represent the longitude and latitude of the terminal point of this trajectory, and A xi , A yi represent the longitude difference and latitude difference between the i-th sampling point and the terminal point in the trajectory coordinate, respectively;

[0055] Then the flight lateral range capability of the i-th sampling point is:

[0056] CR i = |A yi cos(s i )-Axi sin(σ i )| (13)

[0057] 3) using the results of steps 1) and 2), constructing a training set;

[0058] uniformly sampling in the set speed range [V f , V0] to obtain a uniformly distributed speed sequence V uniform , wherein V f is the terminal velocity of reentry flight, and V0 is the initial velocity of reentry flight;

[0059] On each flight trajectory obtained in step 1), according to the original V i , CR i , DR i data sequence, interpolation is performed to obtain the corresponding values of the aircraft range capability and the aircraft lateral range capability on the speed sequence V uniform , and the aircraft range capability sequence and the aircraft lateral range capability sequence composed of the interpolation results are denoted as CR uniform , DR uniform , respectively;

[0060] using V uniform , CR uniform , DR uniform to construct a training set; wherein the input of a single training sample of the training set is any speed value of the morphing aircraft in V uniform in each flight trajectory, the guidance parameter ω corresponding to the flight trajectory and the deformation coefficient k, and the output is the aircraft range capability value and the aircraft lateral range capability value corresponding to the speed value under the flight trajectory, which are obtained from CR uniform , DR uniform , respectively;

[0061] Then the input and output of the training set are denoted as:

[0062] X = {x j,s |1≤j≤42000,j∈N,s∈{1,2,3}}, Y = {y j,s |1≤j≤42000,j∈N,s∈{1,2}},

[0063] wherein X represents the input data set of the training set, Y represents the output data set of the training set, j represents the jth training sample, s represents the sth dimension, x j,s represents the value of the sth dimension of the input data in the jth training sample, and y j,s represents the value of the sth dimension of the output data in the jth training sample.

[0064] In one specific embodiment of the present application, the method further comprises:

[0065] Before training the morphing aircraft flight performance prediction model using the training set, the data of each dimension of the training set is normalized, and the expression is calculated as follows:

[0066]

[0067] wherein, represents the normalized value of the input data of the s-th dimension of the j-th training sample, represents the normalized value of the output data of the s-th dimension of the j-th training sample; x mean,s ,x std,s respectively represent the mean and variance of the input data of the s-th dimension of the training sample; y mean,s ,y std,s respectively represent the mean and variance of the output data of the s-th dimension of the training sample.

[0068] In one specific embodiment of the present application, the nonlinear optimization problem is constructed with the lateral range capability as the optimization performance index and the required range as the constraint, which includes:

[0069] According to the position of the morphing aircraft and the position of the handover point, the required range d is calculated desired , and the expression is calculated as follows:

[0070] d desired = R·arccos [cos (φ msl )·cos (φ tgt )·cos (λ msl -λ tgt )+sin (φ msl )·sin (φ tgt )] (16)

[0071] wherein, λ msl , λ tgt are the longitude of the morphing aircraft and the target point, respectively, and φ msl , φ tgt are the latitude of the morphing aircraft and the target point, respectively;

[0072] Let the predicted value output by the morphing aircraft flight performance prediction model be CR net (V, ω, k), DR net (V, ω, k), then the nonlinear optimization problem with equality constraints is constructed as follows:

[0073]

[0074] wherein, ω min , ω max respectively represent the minimum and maximum values of the guidance parameters; k min , kmax These represent the minimum and maximum values ​​of the deformation coefficient, respectively.

[0075] In a specific embodiment of the present invention, solving the nonlinear optimization problem to obtain the current guidance parameters and deformation parameters of the deformable aircraft includes:

[0076] 1) Based on the optimization problem of equation (17), construct the augmented Lagrangian function:

[0077] L=CR net (V,ω,k)+λ(DR net (V,ω,k)-d desired )+ρ||DR net (V,ω,k)-d desired || 2 (18)

[0078] Where λ and ρ are Lagrange multipliers, and ρ is a positive number;

[0079] 2) Set the initial iteration values ​​ω0, k0, λ0;

[0080] 3) Combining the trained flight performance prediction model of the deformable aircraft, the gradient ascent method is used to optimize the variables. Let the variable obtained in the nth iteration be ω. n ,k n ,λ n ,but:

[0081]

[0082] in, All were calculated by the trained flight performance prediction model of the deformable aircraft;

[0083] 4) Update the augmentation coefficient λ n+1 =λ n +ρ(DR net (V,ω,k)-d desired );

[0084] 5) Judgment: If |ω n+1 -ω n |≤∈ ω ,|k n+1 -k n |≤∈ k , where ∈ ω ,∈ k If the convergence threshold is met, proceed to step 6); otherwise, let n = n + 1 and return to step 3).

[0085] 6) Determine ω*=ω n ,k*=k n To seek the root;

[0086] The calculated k* is used as the deformation parameter command k of the guidance law output. cmd .

[0087] In a specific embodiment of the present invention, obtaining the current angle of attack and roll angle amplitude of the deformable aircraft includes:

[0088] 1) The angle of attack command α is calculated based on the current speed of the deformable aircraft and equation (1). cmd ;

[0089] 2) Calculate the tilt angle amplitude based on the guidance parameter ω*. The calculation expression is as follows:

[0090] ν cmd |=(1-ω*)ν min (k)+ω*ν max (k) (20).

[0091] In one specific embodiment of the present invention, the method further includes:

[0092] Set the horizontal stroke threshold to δCR thre If the horizontal stroke error contraction coefficient is K, then the sign of the tilt angle in each subsequent guidance cycle is obtained by the following method:

[0093] 1) Let the current heading angle of the aircraft be σ, calculate the lateral distance CR of the target point relative to the current heading of the aircraft. tgt The expression is as follows:

[0094] CR tgt =(φ tgt -φ)sinσ-(λ tgt -λ)cosσ (21)

[0095] 2) Let the tilt angle command obtained in the previous guidance cycle be ν. last The transverse error is calculated using the following expression:

[0096]

[0097] Where, the sign function is the sign function; if the aircraft is in the first guidance cycle, then ν last The sign is positive, and the horizontal error recording value δCR is set. record =δCR;

[0098] 3) Calculate the tilt angle flip sign, the expression is as follows:

[0099]

[0100] 4) Calculate the tilt angle command and update the lateral travel record value;

[0101] If FLAG = True, the lateral error obtained in Step 2 is taken as the new lateral record value δCR record = δCR, and the roll angle command is v cmd = - sign(v last ) | v cmd |;

[0102] If FLAG = False, the lateral record value is not updated, and the roll angle command is v cmd = sign(v last ) | v cmd |.

[0103] The characteristics and benefits of the present application are:

[0104] (1) The present application is based on the typical aerodynamic configuration of the morphing aircraft, and according to the flight process constraints, the roll angle corridor of the morphing aircraft is generated with a fixed angle of attack section, which facilitates the parameterization of the control section of the morphing aircraft and simplifies the prediction and correction guidance decision space.

[0105] (2) The present application constructs a dataset of flight performance indicators of the morphing aircraft and trains a flight performance prediction deep learning model, which can take into account the influence of morphing decisions and control decisions on the overall flight task during online decision-making, has good real-time performance, high guidance accuracy, and high practical value.

[0106] (3) The method of the present application can be widely applied to various task scenarios of morphing aircraft, and has important reference value for accurate guidance and control of morphing aircraft. BRIEF DESCRIPTION OF DRAWINGS

[0107] Figure 1 is a flowchart of a morphing aircraft prediction and correction guidance method based on deep learning according to an embodiment of the present application;

[0108] Figure 2 is a roll angle corridor diagram of a morphing aircraft under a typical configuration according to an embodiment of the present application;

[0109] Figure 3 is a flight performance prediction model training Loss diagram according to an embodiment of the present application;

[0110] Figure 4 is a flight trajectory curve diagram of a morphing aircraft according to an embodiment of the present application.

[0111] Figure 5 is a height curve diagram of a morphing aircraft over time according to an embodiment of the present application.

[0112] Figure 6is a time-varying curve diagram of an angle of attack command of a morphing aircraft in one specific embodiment of the present application.

[0113] Figure 7 is a time-varying curve diagram of a roll angle command of a morphing aircraft in one specific embodiment of the present application.

[0114] Figure 8 is a time-varying curve diagram of a morphing coefficient of a morphing aircraft in one specific embodiment of the present application. DETAILED DESCRIPTION

[0115] The present application proposes a morphing aircraft predictive correction guidance method based on deep learning, which will be described in detail below in combination with the drawings and embodiments.

[0116] The embodiment of the present application proposes a morphing aircraft predictive correction guidance method based on deep learning, which comprises:

[0117] A morphing aircraft flight performance prediction model is trained, the input of the model being the speed, guidance parameters and morphing parameters of the morphing aircraft, and the output being the range capability value and the cross-range capability value of the morphing aircraft;

[0118] In each guidance cycle, a nonlinear optimization problem is constructed based on the morphing aircraft flight performance prediction model, with the cross-range capability as the optimization performance index and the required range as the constraint;

[0119] The nonlinear optimization problem is solved to obtain the current guidance parameters and morphing parameters of the morphing aircraft, and further to obtain the current angle of attack and roll angle amplitude of the morphing aircraft, so as to realize the predictive correction guidance of the morphing aircraft.

[0120] In one specific embodiment of the present application, the overall flow of the morphing aircraft predictive correction guidance method based on deep learning is as shown in Figure 1 The method comprises the following steps:

[0121] 1) Based on the relationship between the angle of attack of the morphing aircraft and the speed, the roll angle corridor of the morphing aircraft under each aerodynamic configuration is established; the specific steps are as follows:

[0122] 1-1) The relationship between the angle of attack of the morphing aircraft and the speed is established.

[0123] In this embodiment, the expression of the angle of attack of the morphing aircraft with respect to the speed during flight is set as:

[0124]

[0125] Wherein, α is the angle of attack of the morphing aircraft, V is the speed of the morphing aircraft; V b ,V aare preset speed segment points, V b is required to be less than the initial speed of the morphing aircraft, V a is required to be greater than the terminal speed of the morphing aircraft in reentry flight, V f ; the same numerical value is adopted for each aerodynamic configuration of the morphing aircraft, V b = 4500 m / s, V a = 3000 m / s. is the angle of attack value at which the lift-drag ratio is maximum; a m is the initial value of the angle of attack, which is usually taken as a large value to adapt to the low air density in high altitude at the initial stage of flight, and the value in the embodiment is 16 deg.

[0126] 1-2) Establish the roll angle corridor of the morphing aircraft in each aerodynamic configuration.

[0127] In the embodiment, the reentry flight of the morphing aircraft in each aerodynamic configuration is divided into an initial descent segment, a balanced flight segment, and a terminal capability management segment; for any aerodynamic configuration:

[0128] a) The initial descent segment contains from the reentry point to the first valley bottom point of the trajectory, in which segment, the aerodynamic force is very insufficient, and the main task of the segment is to ensure that the maximum heat peak value occurring in the segment is less than the set allowable value. Given that the maneuvering capability of the aircraft in the initial descent segment is very limited, the roll angle of the aircraft is not adjusted in this segment, and a constant roll angle is taken, and the speed range of the segment is V b . The expression of the upper limit v 1,max and the lower limit v 1,min of the roll angle corridor of the initial descent segment is:

[0129]

[0130] wherein k is a morphing coefficient, representing the morphing amount of the morphing aircraft, such as different sweepback angles of a variable sweep wing aircraft, and in one specific embodiment of the present application, the value range of k is [1, 4], v init is the constant roll angle taken in the initial descent segment, and in one specific embodiment of the present application, the value is 30 deg.

[0131] b) The design goal of the balanced flight segment is to reasonably complete energy dissipation under the premise of meeting strict process constraints.

[0132] In the embodiment, the minimum roll angle of the balanced flight segment is taken as 0, and the speed range is [V a , V b ]. The balanced flight segment mainly considers three process constraints of overload constraint, dynamic pressure constraint, and heat flow constraint, and the expression is as follows:

[0133]

[0134] wherein, Q, G, V represent the heat flux, overload and dynamic pressure that the morphing aircraft can withstand at time t, Qmax, Gmax, Vmax represent the maximum heat flux, maximum dynamic pressure and maximum overload constraints that the morphing aircraft can withstand; S ref S, m0, g0 represent the reference area, mass and gravitational acceleration of the morphing aircraft; C L (α, V, k), C D (α, V, k) represent the lift coefficient and drag coefficient of the morphing aircraft, which are related to the angle of attack α, the speed V and the morphing coefficient k, wherein α is calculated by formula (1); ρ is the atmospheric density at the current height of the aircraft; C1 is a heat flux calculation constant, which is 110311.7; V c is the first cosmic speed, which is 7900 m / s; ρ0 is the atmospheric density at sea level, which is 1.225 kg / m 3 ; R d represents the curvature radius of the leading edge of the morphing aircraft, which is related to the parameters of the aircraft itself, and in this embodiment, it is 0.002.

[0135] Then the air density of the minimum geocentric distance corresponding to each constraint in formula (3) is:

[0136]

[0137] wherein, r Q , r p , r n represent the minimum geocentric distances corresponding to the heat flux constraint, the overload constraint and the dynamic pressure constraint, which can be obtained from the expression or the interpolation table of the air density changing with the height. As can be seen from formula (4), under the condition that the relationship between the angle of attack and the speed is determined and the morphing parameter k is constant, r Q , r p , r n are only related to the speed V.

[0138] In this embodiment, the minimum value of r Q , r n , r p in formula (4) is taken as the minimum flight geocentric distance, and the expression is as follows:

[0139] r min (V)=min{r Q , r q , r n} (5)

[0140] Then the maximum roll angle of the balanced flight segment is calculated according to the quasi-equilibrium gliding condition (QEGC):

[0141]

[0142] Where g is the gravitational acceleration at the altitude of the deformable aircraft, ν 2,max (V,k) represents the maximum tilt angle calculated based on the quasi-equilibrium gliding conditions.

[0143] The upper limit ν of the tilt angle corridor for the balanced flight segment 2,max and the lower realm ν 2,min Represented as:

[0144]

[0145] c) The terminal energy management section needs to satisfy the terminal state constraints. The speed range of this section is V∈[V f V a ].

[0146] This embodiment focuses on the quasi-equilibrium flight condition (QEGC) and the terminal equilibrium flight tilt angle ν. QEGC,f The calculation expression is:

[0147]

[0148] Where, r f ,v f These represent the terminal distance from the Earth's center and the terminal velocity of the reentry flight, respectively. In this embodiment, r is taken as... f =30km,V f =2000m / s. Upper limit ν of the tilt angle corridor in the terminal energy management section. 3,max (V,k) and lower bound ν 3,min The expression for (V,k) is:

[0149]

[0150] Where, k ν The shrinkage coefficient is 4 in this embodiment. The above formula can ensure that V a At the speed level, the tilt angle corridors of the terminal energy management segment and the balanced flight segment can be effectively connected.

[0151] A schematic diagram of a typical downward tilting side corridor configuration of a deformable aircraft in a specific embodiment of the present invention is shown below. Figure 2 As shown in the figure. The figure shows a schematic diagram of the tilt angle corridor when the deformation coefficient k = 1. The upper boundary of the tilt angle corridor is given by ν. 1,max (V,k),ν 2,max (V,k),ν 3,max (V,k) is formed by splicing along the velocity direction, denoted as ν. max (V,k); the lower boundary of the sloping corridor is given by ν. 1,min (V,k),ν 2,min (V,k),ν 3,min(V,k) is formed by splicing along the velocity direction, denoted as ν. min (V,k).

[0152] 2) Based on the results of step 1), establish a training set; the specific steps are as follows:

[0153] 2-1) Based on the range of values ​​of guidance parameter ω and deformation coefficient k, the flight trajectory simulation results of deformable aircraft are obtained under different combinations of guidance parameter ω and deformation coefficient k, and ballistic information is sampled from the flight trajectory.

[0154] In this embodiment, the initial state of the aircraft re-entry process is kept unchanged, and the values ​​of the guidance parameter ω and the deformation coefficient k are uniformly selected within their respective ranges. Then, the simulation results of the corresponding flight trajectory are obtained under each combination of guidance parameter ω and deformation coefficient k. Several trajectories with different ranges and different lateral maneuver distances can be obtained (wherein, each trajectory has the same initial state and the guidance parameter and deformation coefficient remain unchanged during flight).

[0155] In the simulation of each flight trajectory, based on the guidance parameter ω and the roll angle corridor obtained in step 1), the roll angle ν = |ν| can be obtained. cmd |:

[0156] |ν cmd |=(1-ω)ν min (V,k)+ων max (V,k) (10)

[0157] In a specific embodiment of the present invention, ω ranges from [0,1], and k ranges from [1,4], representing the deformation coefficient, i.e., different configurations of the aircraft. ν max (V,k),ν min (V,k) represent the upper and lower bounds of the tilt angle corridor corresponding to the aerodynamic configuration with deformation coefficient k.

[0158] Based on the flight speed and the relationship between the angle of attack and speed of the deformable aircraft established in step 1), i.e., equation (1), the angle of attack α can be obtained. Using the calculated tilt angle and angle of attack in the aircraft simulation process, the flight trajectory simulation results under the corresponding combinations of ω and k can be obtained. In this embodiment, by traversing different deformation coefficients k and control profile weight values ​​ω, several flight trajectories can be obtained.

[0159] Furthermore, by sampling and recording the basic ballistic information sequence for each flight trajectory, the ballistic information includes: the altitude r of each sampling point. i Longitude λ i latitude φ i Speed ​​V i Ballistic inclination angle θ i Ballistic deflection angle σi and flight time t i (The sampling frequency in this embodiment is 1 second), where the subscript i represents the i-th sampling point. In a specific embodiment of the present invention, 21 values ​​are taken from the range of ω and 4 values ​​are taken from the range of k, resulting in simulation results of 84 trajectories. In the simulation of each trajectory, the simulation step size is 1 second, the starting speed is 5000 m / s, and the ending speed is 2000 m / s.

[0160] 2-2) Based on the ballistic information obtained in step 2-1), calculate the flight performance index of each sampling point on each flight trajectory.

[0161] In this embodiment, the range capability and lateral range capability of the aircraft at each sampling point of each flight trajectory can be obtained from the ballistic information of that sampling point.

[0162] Wherein, the aircraft's range capability refers to the distance an aircraft can travel from the current moment to a location that meets the handover speed and altitude requirements, under a certain guidance and aerodynamic configuration selection strategy. In this embodiment, the expression for calculating the aircraft's range capability at the i-th sampling point is as follows:

[0163]

[0164] Among them, DR i S represents the range capability of the aircraft at the i-th sampling point. i S represents the range reached by the aircraft at the i-th sampling point. f This indicates the distance the aircraft has traveled along the entire trajectory.

[0165] The lateral range capability of an aircraft refers to the lateral flight distance of an aircraft from the current moment to the moment it meets the handover speed and altitude requirements under certain guidance and aerodynamic configuration selection strategies. In other words, it is the distance between the terminal handover point and the current trajectory line of the aircraft.

[0166] In this embodiment, let the coordinate difference between the i-th sampling point and the terminal point in the trajectory coordinates be:

[0167]

[0168] Where, λ f ,φ f This represents the longitude and latitude of the terminal point of this trajectory, Δ. xi ,Δ yi These represent the longitude difference and latitude difference between the i-th sampling point and the terminal point in the trajectory coordinates, respectively.

[0169] The lateral range capability of the aircraft at the i-th sampling point is:

[0170] CR i =|Δ yicos(σ i )-Δ xi sin(σ i (13)

[0171] 2-3) Using the results of steps 2-1) and 2-2), construct a training set.

[0172] In this embodiment of the invention, within the set speed range [V] f ,V0](V f V0 is the reentry flight termination velocity, and V0 is the reentry flight initial velocity. In a specific embodiment of the present invention, multiple points (500 points in a specific embodiment of the present invention) are uniformly sampled in the range [2000, 5000] m / s to obtain a uniformly distributed velocity sequence V0. uniform (There are no restrictions on the range of values; try to include as many as possible.)

[0173] On each flight trajectory obtained in step 2-1), based on the original data sequence V i ,CR i ,DR i Interpolation yields the aircraft's range and lateral range capabilities in the velocity sequence V. uniform The corresponding values ​​on the above, the interpolation results forming the aircraft range capability sequence and the aircraft lateral range capability sequence are respectively denoted as CR uniform ,DR uniform .

[0174] Then use V uniform ,CR uniform ,DR uniform Building the training set:

[0175] In this embodiment, the input of a single training sample in the training set is the deformable aircraft in V for each flight trajectory. uniform Given any velocity value in the data, the corresponding guidance parameter ω and deformation coefficient k for that flight trajectory, the outputs are the aircraft's range capability and lateral range capability for that velocity value under that flight trajectory, respectively derived from CR. uniform ,DR uniform The data is obtained from the data. In one specific embodiment of the present invention, a total of 84*500=42000 training samples are generated (84 trajectories, each trajectory has 500 sampling points, corresponding to 500 velocity values). The input dimension of each training sample is 3 and the output dimension is 2.

[0176] In this embodiment, the input and output of the training set are denoted as:

[0177] X={x j,s |1≤j≤42000,j∈N,s∈{1,2,3}},Y={y j,s|1≤j≤42000,j∈N,s∈{1,2}}

[0178] Where X represents the input data set of the training set, Y represents the output data set of the training set, j represents the j-th training sample, s represents the s-th dimension, and x j,s Let y represent the value of the s-th dimension of the input data in the j-th training sample. j,s This represents the value of the s-th dimension of the output data in the j-th training sample.

[0179] 3) Use the training set obtained in step 2) to train the flight performance prediction model of the deformable aircraft, and obtain the trained flight performance prediction model of the deformable aircraft; the specific steps are as follows:

[0180] 3-1) Preprocess the training set data obtained in step 2).

[0181] In this embodiment, the data of each dimension of the training set obtained in step 2) is normalized, and the calculation expression is as follows:

[0182]

[0183] in, This represents the normalized value of the input data in dimension s of the j-th training sample. x represents the normalized value of the output data of the j-th training sample in the s-th dimension; mean,s ,x std,s Let y represent the mean and variance of the input data in the s-th dimension of the training samples, respectively; mean,s ,y std,s Let represent the mean and variance of the output data in the s-th dimension of the training sample, respectively.

[0184] 3-2) Construct a flight performance prediction model for deformable aircraft.

[0185] This embodiment uses a neural network model with four fully connected layers as the flight performance prediction model for deformable aircraft. The model has an input dimension of 3 (velocity V, guidance parameter ω, and deformation parameter k), a hidden layer dimension of 128, and an output dimension of 2 (range capability value and lateral range capability value). During the forward propagation process, the ELU function and the tanh function are used to introduce the nonlinear properties of the neural network. The ELU function is used in the first layer, the tanh function in the second and third layers, and there is no activation function in the output layer. The training loss function is as follows:

[0186]

[0187] Among them, Y pre This represents the output data predicted by the model. This represents the true values ​​of the output data in the training set.

[0188] The initial learning rate used in training was 0.002, and a learning rate decay method was adopted. The optimizer was the Adam optimizer, and its parameters were the default parameters of the torch library. During training, the learning rate (lr) was reduced to 0.8 times its original value every two epochs, for a total of 200 epochs. The batch size for training data was 32.

[0189] The trained morphing aircraft flight performance prediction model is denoted as net, and the output calculated by the model based on the input V, ω, k values ​​is denoted as CR. net (V,ω,k) and DR net (V,ω,k), since this model is composed of a neural network, it has the characteristic of backpropagation differentiation, and the partial derivatives of the flight performance with respect to ω and k can be calculated based on the input values ​​of V,ω,k. This completes the offline portion of this embodiment.

[0190] 4) During online guidance, within each guidance cycle, a nonlinear optimization problem is first constructed, with transverse range capability as the optimized performance index and the required range as the constraint; specifically as follows:

[0191] The required range d is calculated based on the deformable aircraft's own position and the handover point position. desired The calculation expression is as follows:

[0192] d desired =R·arccos[cos(φ) msl )·cos(φ tgt )·cos(λ msl -λ tgt )+sin(φ msl )·sin(φ tgt (16)

[0193] Where, λ msl ,λ tgt The longitudes of the morphing aircraft and the target point are φ, respectively. msl ,φ tgt These are the latitudes of the transforming aircraft and the target point, respectively.

[0194] Let the predicted value of the prediction model trained in step 3) be CR. net (V,ω,k),DR net Given (V,ω,k), the nonlinear optimization problem with equality constraints is constructed as follows:

[0195]

[0196] Where, ω min ,ωmax These represent the minimum and maximum values ​​of the guidance parameters, respectively, which are taken as 0 and 1 in this embodiment; k min ,k max represents the minimum and maximum values ​​of the deformation coefficient, respectively. In this embodiment, they are taken as 1 and 4. 5) Solve the nonlinear optimization problem established in step 4).

[0197] This embodiment uses the augmented Lagrange function method to solve the nonlinear optimization problem in step 4), as detailed below:

[0198] 5-1) Based on the optimization problem shown in equation (17), construct the augmented Lagrangian function:

[0199] L=CR net (V,ω,k)+λ(DR net (V,ω,k)-d desired )+ρ||DR net (V,ω,k)-d desired || 2 (18)

[0200] Where λ and ρ are Lagrange multipliers, and ρ is a positive number.

[0201] 5-2) Set the initial iteration values ​​ω0, k0, λ0. In this embodiment, ω0 = 0.5, k0 = 2.5, λ0 = 0.5.

[0202] 5-3) Combining the model trained in step 3), optimize the variables using the gradient ascent method. Let the variable obtained in the nth iteration be ω. n ,k n ,λ n .

[0203]

[0204] in, All of these are calculated from the model trained in step 3).

[0205] 5-4) Update the augmentation coefficient λ n+1 =λ n +ρ(DR net (V,ω,k)-d desired ).

[0206] 5-5) Judgment: If |ω n+1 -ω n |≤∈ ω ,|k n+1 -k n |≤∈ k , where ∈ ω ,∈ kIf the convergence threshold is met, proceed to step 5-6; otherwise, let n = n + 1 and return to step 5-3).

[0207] 5-6) Determine ω*=ω n ,k*=k n To seek the root.

[0208] The calculated k* is used as the deformable parameter command k of the guidance law output. cmd Then proceed to step 6).

[0209] 6) Based on the results of step 5), calculate the angle of attack command and the yaw angle command; the specific steps are as follows:

[0210] 6-1) The angle of attack command α is calculated based on the current velocity of the deformable aircraft and equation (1). cmd .

[0211] 6-2) Calculate the tilt angle amplitude based on the guidance parameter ω* obtained in step 5). The calculation expression is:

[0212] |ν cmd |=(1-ω*)ν min (k)+ω*ν max (k) (20)

[0213] Furthermore, the method described in this embodiment also includes:

[0214] 7) Set the transverse threshold to δCR thre If the horizontal stroke error contraction coefficient is K, then the sign of the tilt angle in each subsequent guidance cycle is obtained by the following method:

[0215] 7-1) Let the current heading angle of the aircraft be σ, calculate the lateral distance CR of the target point relative to the current heading of the aircraft. tgt The expression is as follows:

[0216] CR tgt =(φ tgt -φ)sinσ-(λ tgt -λ)cosσ (21)

[0217] 7-2) Let the tilt angle command obtained in the previous guidance cycle be ν. last The transverse error is calculated using the following expression:

[0218]

[0219] Here, the sign function is the sign function. If the aircraft is in the first guidance cycle, then ν last The sign is positive by default, and the horizontal error recording value δCR is set. record =δCR.

[0220] 7-3) Calculate the tilt angle flip mark, the expression is as follows:

[0221]

[0222] 7-4) Calculate the tilt angle command and update the lateral travel record value.

[0223] If FLAG = True, then the horizontal error obtained in Step 2 will be used as the new horizontal record value δCR. record =δCR, the yaw angle command is ν cmd =-sign(ν last )|ν cmd |

[0224] If FLAG = False, the lateral travel record value will not be updated, and the tilt angle command will be ν. cmd =sign(ν last )|ν cmd |

[0225] At this point, the angle of attack command (step 6), bank angle command (step 6), and deformation parameter command (step 5) for each guidance cycle are obtained. Within this guidance cycle, the aircraft executes the obtained angle of attack command, bank angle command, and deformation parameter command. At the start of the next guidance cycle, steps 4), 5), and 6) are repeated to perform a new round of command calculations. This process is repeated until the aircraft reaches the designated area and completes its flight mission.

[0226] Furthermore, the following Python simulation illustrates the process of the deep learning-based predictive correction guidance method for deformable aircraft according to this invention. Figure 3 This is a loss graph for training a flight performance prediction model in a specific embodiment of the present invention. For example... Figure 3 As shown, a total of 100 epochs were conducted during the training process. The loss function of a single epoch decreased from 0.003 to below 0.0001, indicating that the flight performance prediction model has a good fitting effect.

[0227] In the online phase of the method described in this embodiment, the initial altitude, longitude, latitude, speed, trajectory inclination angle, and trajectory deflection angle of the aircraft are set to 56km, 0deg, 0deg, 5000m / s, 0deg, and 90deg, respectively. The longitude and latitude of the target point are set to 28deg and 5deg, respectively. The altitude range of the target point is [29, 31]km. The final velocity of the aircraft is 2000m / s, the guidance period is 1s, and the transverse range threshold is δCR. thre =10km, the horizontal error shrinkage coefficient is K=2. Figure 4 This is a flight trajectory curve of a deformable aircraft in a specific embodiment of the present invention. For example... Figure 4As shown, using the method described in this embodiment, the deformable aircraft can reach a predetermined target point. Figure 5 This is a graph showing the change in altitude of a deformable aircraft over time in a specific embodiment of the present invention. Figure 5 As shown, the morphing aircraft reached the predetermined target altitude of 30.233 km, which meets the target altitude requirement. Figure 6 This is a graph showing the change of angle-of-attack command of a deformable aircraft over time in a specific embodiment of the present invention; Figure 7 This is a graph showing the change of tilt angle command of the deformable aircraft over time in a specific embodiment of the present invention; Figure 8 This is a graph showing the deformation coefficient of the deformable aircraft over time in a specific embodiment of the present invention. In this embodiment, the aircraft underwent seven rollovers during flight.

Claims

1. A prediction and correction guidance method for deformable aircraft based on deep learning, characterized in that, include: A flight performance prediction model for a deformable aircraft is trained. The inputs of the model are the speed, guidance parameters, and deformation parameters of the deformable aircraft, and the outputs are the range capability and lateral range capability of the deformable aircraft. Within each guidance cycle, based on the flight performance prediction model of the deformable aircraft, a nonlinear optimization problem is constructed with lateral range capability as the optimized performance index and the required range as the constraint. Solving the nonlinear optimization problem yields the current guidance parameters and deformation parameters of the deformable aircraft, which in turn provides the current angle of attack and roll angle amplitudes of the deformable aircraft, enabling predictive correction guidance for the deformable aircraft. The method further includes: Based on the relationship between the angle of attack and velocity of the morphing aircraft, a roll angle corridor for the morphing aircraft under each aerodynamic configuration is established; the specific steps are as follows: 1) Establish the relationship between the angle of attack and velocity of the transforming aircraft; The expression for the angle of attack of the transforming aircraft during flight, with respect to velocity, is as follows: Where α is the angle of attack of the transforming aircraft, and V is the velocity of the transforming aircraft; V b V a These are the preset velocity segment points, V b Less than the initial velocity of the transforming aircraft, V a The reentry terminal velocity V greater than that of the deformable aircraft f ; The angle of attack value with the highest lift-to-drag ratio; α m The initial value is the angle of attack; 2) Establish the tilt angle corridor for the deformable aircraft under each aerodynamic configuration; The reentry flight of the deformable aircraft under each aerodynamic configuration is divided into an initial descent phase, a balanced flight phase, and a terminal capability management phase; for any aerodynamic configuration: The initial descent segment includes the path from the reentry point to the first trough of the trajectory, with a velocity range of V ≥ V. b ; Initial descent segment tilt angle upper limit ν of the corridor 1,max and the lower realm ν 1,min The expression is: Where k is the deformation coefficient, representing the amount of deformation of the deformable aircraft; ν init A constant tilt angle is used for the initial descent segment; The minimum tilt angle of the balanced flight segment is 0, and the speed range is [V]. a V b The balanced flight segment includes three process constraints: overload constraint, dynamic pressure constraint, and thermal flow constraint, expressed as follows: in, n(t) and p(t) represent the heat flux, overload, and dynamic pressure experienced by the deformable aircraft at time t, respectively. p max ,n max These represent the maximum heat flux, maximum dynamic pressure, and maximum overload constraint that the morphing aircraft can withstand; S ref m0 and g0 represent the reference area, mass, and gravitational acceleration of the deformable aircraft, respectively; C L (α,V,k),C D (α, V, k) represent the lift coefficient and drag coefficient of the deformable aircraft, respectively; ρ is the atmospheric density at the current altitude of the deformable aircraft; C1 is the heat flux calculation constant; V c R is the first cosmic velocity; ρ0 is the atmospheric density at sea level; d This represents the radius of curvature of the leading edge of the deformable aircraft; Then the air density corresponding to the minimum geocentric distance for each constraint in equation (3) is: ρ(r p )=2p max V -2 Where, r Q ,r p ,r n These represent the minimum geocentric distances corresponding to thermal flux constraints, overload constraints, and dynamic pressure constraints, respectively. Take r in equation (4) Q ,r n ,r p The minimum value in the range is taken as the minimum flight geocentric distance, and the expression is as follows: r min (V)=min{r Q ,r q ,r n } (5) Then, calculate the maximum roll angle during the balanced flight segment based on the quasi-balanced flight conditions: Where g is the gravitational acceleration at the altitude of the deformable aircraft, ν 2,max (V,k) represents the maximum tilt angle calculated based on the quasi-equilibrium gliding conditions; The upper limit ν of the tilt angle corridor for the balanced flight segment 2,max and the lower realm ν 2,min Represented as: The terminal energy management section satisfies the terminal state constraints, with a speed range of V∈[V]. f V a ]; Based on the quasi-equilibrium flight conditions, the terminal equilibrium flight tilt angle ν QEGC,f The calculation expression is: Where, r f ,v f These represent the terminal geocentric distance and terminal velocity of the reentry flight, respectively; the upper limit ν of the tilt corridor during the terminal energy management phase. 3,max (V,k) and lower bound ν 3,min The expression for (V,k) is: Where, k ν This is the shrinkage coefficient.

2. The method according to claim 1, characterized in that, The flight performance prediction model for the deformable aircraft adopts a neural network model with four fully connected layers. The first layer of the model uses the ELU function, and the second and third layers use the tanh function.

3. The method according to claim 1, characterized in that, The method further includes: Based on the roll angle corridor of the morphing aircraft under each aerodynamic configuration, a training set is established for training the flight performance prediction model of the morphing aircraft. The specific steps are as follows: 1) Based on the range of values ​​of guidance parameter ω and deformation coefficient k, the flight trajectory simulation results of deformable aircraft are obtained under different combinations of guidance parameter ω and deformation coefficient k, and ballistic information is sampled from the flight trajectory. In the simulation of each flight trajectory, the roll angle ν = |ν| is obtained based on the guidance parameter ω and the roll angle corridor. cmd |: |n cmd |=(1-ω)ν min (k)+ων max (k) (10) Where ω takes values ​​in the range [0,1], ν max (k),ν min (k) represents the upper and lower bounds of the tilt angle corridor corresponding to the aerodynamic configuration with deformation coefficient k; The angle of attack α is calculated based on the flight speed and equation (1); Simulations are performed using the tilt angle and angle of attack to obtain the flight trajectory simulation results under the corresponding combinations of ω and k. Then, the basic ballistic information sequence of this flight trajectory is recorded by sampling, and the ballistic information includes: the height r of each sampling point. i Longitude λ i latitude φ i Speed ​​V i Ballistic inclination angle θ i Ballistic deflection angle σ i and flight time t i Where the subscript i represents the i-th sampling point; 2) Based on the ballistic information obtained in step 1), calculate the flight performance indicators of each sampling point on each flight trajectory, including: the range capability and the lateral range capability of the aircraft; The formula for calculating the range capability of the aircraft at the i-th sampling point is as follows: DR i =S f -S i Among them, DR i S represents the range capability of the aircraft at the i-th sampling point. i S represents the range reached by the aircraft at the i-th sampling point. f This indicates the distance the aircraft has traveled along the entire trajectory; Let the coordinate difference between the i-th sampling point and the terminal point in the trajectory coordinates be: Where, λ f ,φ f This represents the longitude and latitude of the terminal point of this trajectory, Δ. xi ,Δ yi These represent the longitude difference and latitude difference between the i-th sampling point and the terminal point in the trajectory coordinates, respectively; The lateral range capability of the aircraft at the i-th sampling point is: CR i =|D yi cos(σ i )-D xi sin(σ i )| (13) 3) Using the results of steps 1) and 2), construct a training set; Within the set speed range [V f Uniform sampling is performed within [V0] to obtain a uniformly distributed velocity sequence V. uniform , where V f V0 is the reentry flight termination velocity, and V0 is the reentry flight initial velocity. On each flight path obtained in step 1), based on the original V i ,CR i ,DR i The data sequence was interpolated to obtain the aircraft's range and lateral range capabilities in the velocity sequence V. uniform The corresponding values ​​on the above, the interpolation results forming the aircraft range capability sequence and the aircraft lateral range capability sequence are respectively denoted as CR uniform ,DR uniform ; Using V uniform ,CR uniform ,DR uniform Construct a training set; wherein, the input of a single training sample in the training set is the deformable aircraft in V for each flight trajectory. uniform Given any velocity value in the data, the corresponding guidance parameter ω and deformation coefficient k for that flight trajectory, the outputs are the aircraft's range capability and lateral range capability for that velocity value under that flight trajectory, respectively derived from CR. uniform ,DR uniform Obtain from; The input and output of the training set are denoted as follows: X={x j,s |1≤j≤42000,j∈N,s∈{1,2,3}},Y={y j,s |1≤j≤42000,j∈N,s∈{1,2}}, Where X represents the input data set of the training set, Y represents the output data set of the training set, j represents the j-th training sample, s represents the s-th dimension, and x j,s Let y represent the value of the s-th dimension of the input data in the j-th training sample. j,s This represents the value of the s-th dimension of the output data in the j-th training sample.

4. The method according to claim 3, characterized in that, The method further includes: Before training the flight performance prediction model of the deformable aircraft using the training set, the data for each dimension of the training set is normalized, and the calculation expression is as follows: in, This represents the normalized value of the input data in dimension s of the j-th training sample. x represents the normalized value of the output data of the j-th training sample in the s-th dimension; mean,s ,x std,s Let y represent the mean and variance of the input data in the s-th dimension of the training samples, respectively; mean,s ,y std,s Let represent the mean and variance of the output data in the s-th dimension of the training sample, respectively.

5. The method according to claim 4, characterized in that, The construction of the nonlinear optimization problem, which uses lateral range capability as the performance index and the required range as the constraint, includes: The required range d is calculated based on the deformable aircraft's own position and the handover point position. desired The calculation expression is as follows: d desired =R arccos[cos(φ msl )·cos(φ tgt )·cos(λ msl -l tgt )+sin(φ msl )·sin(φ tgt )] (16) Where, λ msl ,λ tgt The longitudes of the morphing aircraft and the target point are φ, respectively. msl ,φ tgt These are the latitudes of the deformable aircraft and the target point, respectively. Let the predicted value output by the flight performance prediction model of the deformable aircraft be CR. net (V,ω,k),DR net Given (V,ω,k), the nonlinear optimization problem with equality constraints is constructed as follows: Where, ω min ,ω max These represent the minimum and maximum values ​​of the guidance parameters, respectively; k min ,k max These represent the minimum and maximum values ​​of the deformation coefficient, respectively.

6. The method according to claim 5, characterized in that, Solving the nonlinear optimization problem to obtain the current guidance parameters and deformation parameters of the deformable aircraft includes: 1) Based on the optimization problem of equation (17), construct the augmented Lagrangian function: L=CR net (V,ω,k)+λ(DR net (V,ω,k)-d desired )+ρ||DR net (V,ω,k)-d desired || 2 (18) Where λ and ρ are Lagrange multipliers, and ρ is a positive number; 2) Set the initial iteration values ​​ω0, k0, λ0; 3) Combining the trained flight performance prediction model of the deformable aircraft, the gradient ascent method is used to optimize the variables. Let the variable obtained in the nth iteration be ω. n ,k n ,λ n ,but: Among them, DR net (V,ω,k), All were calculated by the trained flight performance prediction model of the deformable aircraft; 4) Update the augmentation coefficient λ n+1 =λ n +ρ(DR net (V,ω,k)-d desired ); 5) Judgment: If |ω n+1 -ω n |≤∈ ω ,|k n+1 -k n |≤∈ k , where ∈ ω ,∈ k If the convergence threshold is met, proceed to step 6); otherwise, let n = n + 1 and return to step 3). 6) Determine ω*=ω n ,k*=k n To seek the root; The calculated k* is used as the deformation parameter command k of the guidance law output. cmd .

7. The method according to claim 6, characterized in that, The process of obtaining the current angle of attack and roll angle amplitude of the deformable aircraft includes: 1) The angle of attack command α is calculated based on the current speed of the deformable aircraft and equation (1). cmd ; 2) Calculate the tilt angle amplitude based on the guidance parameter ω*. The calculation expression is as follows: |n cmd |=(1-ω*)n min (k)+ω*ν max (k) (20).

8. The method according to claim 7, characterized in that, The method further includes: Set the horizontal stroke threshold to δCR thre If the horizontal stroke error contraction coefficient is K, then the sign of the tilt angle in each subsequent guidance cycle is obtained by the following method: 1) Let the current heading angle of the aircraft be σ, calculate the lateral distance CR of the target point relative to the current heading of the aircraft. tgt The expression is as follows: CR tgt =(φ tgt -φ)sinσ-(λ tgt -λ)cosσ (21) 2) Let the tilt angle command obtained in the previous guidance cycle be ν. last The transverse error is calculated using the following expression: Where, the sign function is the sign function; if the aircraft is in the first guidance cycle, then ν last The sign is positive, and the horizontal error recording value δCR is set. record =δCR; 3) Calculate the tilt angle flip sign, the expression is as follows: 4) Calculate the tilt angle command and update the lateral travel record value; If FLAG = True, then the horizontal error obtained in Step 2 will be used as the new horizontal record value δCR. record =δCR, the yaw angle command is ν cmd =-sign(ν last )|ν cmd |; If FLAG = False, the lateral travel record value will not be updated, and the tilt angle command will be ν. cmd =sign(ν last )|ν cmd |

Images