CN113126643A - Intelligent robust reentry guidance method and system for hypersonic aircraft - Google Patents

Abstract

The invention belongs to the technical field of hypersonic gliding aircraft reentry trajectory planning guidance and discloses a hypersonic aircraft intelligent robust reentry guidance method and a hypersonic aircraft intelligent robust reentry guidance system, wherein the hypersonic aircraft intelligent reentry guidance method comprises the following steps: based on uncertainty system modeling, judging uncertainty and types thereof existing in the gliding reentry process of the hypersonic aerocraft, and determining an uncertainty parameter distribution form and a distribution interval; carrying out uncertainty quantitative analysis, and establishing a robust dynamic trajectory optimization model containing the statistical moment characteristics of a random system; establishing a robust track optimization numerical sample set; designing a deep neural network model architecture, carrying out model training and verifying the effectiveness of the model; and loading a deep neural network model, and intelligently updating the robust trajectory planning guidance instruction in real time. The method can reduce the robust track design time, enhance the active defense capability of the guidance instruction to the complex uncertainty, and reduce the design burden of an aircraft guidance control system.

Description

A kind of intelligent and robust reentry guidance method and system for hypersonic aircraft

technical field

The invention belongs to the technical field of re-entry trajectory planning and guidance of a hypersonic gliding aircraft, and in particular relates to an intelligent and robust re-entry guidance method and system for a hypersonic aircraft.

Background technique

At present, hypersonic glide vehicles are one of the main types of hypersonic vehicles in near space. They are usually transported to a predetermined altitude by a launch vehicle or released by a space-based platform, and re-enter the atmosphere at hypersonic speed, relying on their own high lift-to-drag ratio and aerodynamic control. Long-distance unpowered gliding, and finally complete tasks such as precise strikes on targets. Because it breaks through the conventional ballistic reentry mode, and has many advantages such as fast response speed, long flight distance, large space span, strong maneuvering penetration capability, and high strike accuracy, hypersonic gliding vehicles are regarded by various countries as near space resources competition and national competition. The technical commanding heights of strategic security.

Trajectory design is a core issue in the field of guidance and control of hypersonic aircraft. It needs to undertake the overall analysis task of the overall design stage of the aircraft, and provides a theoretical basis for the multidisciplinary design of guidance, control, penetration and thermal protection, and runs through almost the entire aircraft design process. The inherent properties of unpowered gliding and reentry of hypersonic glide vehicles make the trajectory design challenges more severe. The "ideal goal" of hypersonic vehicle trajectory design is to exceed the accuracy of the navigation error, to exceed the efficiency of the guidance period, and to obtain the global optimal trajectory online under the condition that all constraints are satisfied.

At present, due to the limited processing capability of missile-borne computing, the trajectory design framework of offline trajectory optimization and online tracking and guidance is generally adopted in engineering: the reference trajectory is optimized based on the particle dynamics model in the offline design stage and stored in the missile-borne computer; during the online flight process The guidance and control system tracks the reference trajectory and guides the aircraft to the terminal position according to the pre-planned trajectory. The traditional offline trajectory optimization is based on the deterministic nominal model, ignoring the uncertainty in the aircraft itself and the environment, so its reference trajectory often lacks robustness, which can easily lead to unpredictable fluctuations in the flight process or deviation from the design direction, thereby Significantly affect the performance of the mission, and even threaten the safety of the aircraft itself. Using technologies such as online trajectory optimization, online tracking guidance or predictive-correction guidance can compensate for the deficiencies of deterministic reference trajectories to a certain extent, but there are still new design difficulties: in order to meet real-time requirements, online trajectory optimization usually sacrifices part of the trajectory Optimality is at the expense, but considering the limitation of the amount of consumables on board, the optimal trajectory design is still very necessary; online tracking guidance is fast in calculation speed but low in guidance accuracy, and prediction-correction guidance accuracy is high but relies heavily on computing power. The hybrid guidance is an improved method under the existing architecture, but it lacks fundamental improvement in trajectory robust design and computing resource allocation, and its limited correction capability limits the performance of the aircraft.

Researches related to intelligent trajectory planning and guidance have shown that the deep learning-based trajectory controller has the potential to take over all or part of the missile-borne trajectory generation and guidance system, but the research on robust intelligent trajectory planning and guidance considering uncertainty has not attracted much attention. Since the on-board computing power is difficult to support the online learning process, the deep neural network flight controller is mainly implemented through "offline training + online application". Facing the complex multi-source uncertainty reentry trajectory optimization problem, there is a lack of high-fidelity robust dynamic trajectory optimization model and multi-condition robust trajectory design data support.

In view of the above problems, it is of great scientific significance and engineering application value to break through the traditional trajectory optimization and guidance technology framework of hypersonic vehicles and carry out research on intelligent reentry guidance methods for hypersonic vehicles based on robust trajectory optimization.

Through the above analysis, the existing problems and defects in the prior art are:

(1) The traditional offline trajectory optimization is based on the deterministic nominal model, ignoring the uncertainty in the aircraft itself and the environment, so its reference trajectory often lacks robustness, which can easily lead to unpredictable fluctuations in the flight process or deviation from the design direction, which significantly affects the performance of the mission, and even threatens the safety of the aircraft itself.

(2) Using technologies such as online trajectory optimization, online tracking guidance or prediction-correction guidance can compensate for the deficiencies of deterministic reference trajectories to a certain extent, but there are still new design difficulties: in order to meet real-time requirements, online trajectory optimization is usually based on At the cost of sacrificing part of the trajectory optimality, but considering the limitation of the amount of consumables onboard, the trajectory optimality design is still very necessary.

(3) Online tracking guidance has fast calculation speed but low guidance accuracy, and prediction-correction guidance has high accuracy but relies heavily on computing power. The hybrid guidance of the two is an improved method under the existing architecture, but it is robust to trajectory design and computational resources. The configuration lacks fundamental improvement, and its limited correction capabilities limit the performance of the aircraft.

(4) In the prior art, the research on robust intelligent trajectory planning and guidance considering uncertainty has not attracted much attention, and in the face of complex multi-source uncertainty re-entry trajectory optimization problems, there is a lack of high-fidelity robust dynamic trajectory optimization models and Multi-condition robust trajectory design data support.

The difficulty of solving the above problems and defects is as follows:

(1) Under the traditional trajectory design framework, the numerical trajectory solution efficiency often has a conflicting relationship with other key indicators such as modeling complexity, discretization accuracy, and global optimality. The much-needed flexibility in the field of hypersonic vehicle trajectory design and guidance The need for rapidity remains unmet. For the robust reentry trajectory optimization problem oriented to complex uncertainties, due to the introduction of various uncertain factors and stochastic optimization models, the number of model constraints and design variables increases exponentially, and its numerical solution will be more difficult.

(2) Since the on-board computing capability is difficult to support the online learning process, the deep neural network flight controller is mainly realized by the method of "offline training + online application", and its reliability is highly dependent on the training model and training data. Facing the complex multi-source uncertainty reentry trajectory optimization problem, there is a lack of high-fidelity robust dynamic trajectory optimization model and multi-condition robust trajectory design data support.

The significance of solving the above problems and defects is:

(1) Carry out robust trajectory design for multi-source uncertainty, which can build a reverse information flow of offline trajectory optimization and online guidance control on the basis of traditional architecture, and indirectly build a bridge between trajectory control instructions and real re-entry scenarios. It makes the relatively independent design steps on the timeline become an interactive whole, effectively improving the reliability and security of the overall design.

(2) Based on the pattern recognition ability of artificial intelligence, use deep learning technology to approximate uncertain nonlinear dynamic models, learn robust trajectory planning and guidance optimization strategies, establish intelligent robust optimal trajectory and guidance models, and realize online trajectory Prediction and path selection can effectively deal with flight problems such as target changes, terminal combat changes, and uncertain disturbances, thereby improving the flexibility of reentry flights.

SUMMARY OF THE INVENTION

In view of the problems existing in the prior art, the present invention provides an intelligent robust reentry guidance method and system for a hypersonic aircraft, in particular to an intelligent reentry guidance method and system for a hypersonic aircraft based on robust trajectory optimization.

The present invention is implemented in this way, a method for intelligent and robust re-entry guidance for a hypersonic aircraft, the method for intelligent and robust re-entry guidance for a hypersonic aircraft includes the following steps:

Step 1: Based on the uncertainty system modeling, according to data mining and expert experience, determine the uncertainty and type of the hypersonic vehicle in the process of gliding reentry, and determine the distribution form and distribution interval of the uncertainty parameters, which is the reentry process. Trajectory robust modeling and design lay the foundation;

Step 2, according to the uncertain factors obtained in step 1, carry out quantitative analysis of uncertainty based on non-embedded polynomial chaos theory, and establish a robust dynamic trajectory optimization model including the stochastic system target and statistical moment characteristics of state variables; by changing the uncertain parameters value and design a numerical solution strategy, establish a robust trajectory optimization numerical sample set for multi-source uncertain factors, and provide reliability basis and data support for the design of intelligent robust guidance models;

Step 3, take the trajectory state vector as the input and the trajectory control vector as the output, design the deep neural network model architecture; use the robust trajectory optimization numerical sample set for multi-source uncertain factors obtained in step 2 to train the model, and verify the model Effectiveness, providing feasibility analysis for online application of intelligent robust guidance model;

Step 4: Load the well-trained deep neural network model in Step 3. According to the flight status information output by the missile-borne control system including the navigation and positioning system and the attitude control system during the re-entry process of the aircraft, the intelligent real-time robust trajectory planning guidance command is finally realized. renew.

Further, in step 1, based on the uncertainty system modeling, according to data mining and expert experience, the uncertainty and its type in the process of gliding and re-entry of the hypersonic vehicle are judged, and the distribution form and distribution of the uncertainty parameters are determined. interval, including:

The cognitive uncertainty, stochastic uncertainty and numerical uncertainty in the high-speed flight dynamics system are classified and modeled. On this basis, the sources and types of uncertainty in the robust trajectory optimization design process are studied. Uncertainty modeling of initial reentry conditions, aerodynamic model data, atmospheric model parameters and the vehicle's own perturbation parameters during unpowered reentry of a supersonic gliding vehicle;

Wherein, the types of uncertain factors are:

V₀、γ₀、ψ₀分别代表飞行器状态变量地心距、经度、纬度、速度、航迹角、航向角的初始再入状态，C_L、C_D分别代表飞行器升力系数、阻力系数，ρ代表大气密度，m_V代表飞行器自身质量。Among them, ξ is a random variable vector; subscripts y ₀ , θ ₀ ,

V ₀ , γ ₀ , ψ ₀ represent the initial re-entry state of the aircraft state variables geocentric distance, longitude, latitude, speed, track angle, and heading angle, respectively, _{C L} and CD represent the lift coefficient and drag coefficient of the aircraft, ρ represents the density of the atmosphere, and m _V represents the mass of the aircraft itself.

Further, in step 1, the distribution form of the uncertain factors is assumed to be uniformly distributed and independent of each other, and the distribution range of the random model of the system and the uncertain factors is:

分别表示飞行器随机初始再入状态；

代表随机升力系数、随机阻力系数；

为随机大气密度；

为随机飞行器质量。in,

respectively represent the random initial re-entry state of the aircraft;

Represents random lift coefficient and random drag coefficient;

is the random atmospheric density;

is the random aircraft mass.

Further, in step 2, the aircraft stochastic trajectory optimization model includes stochastic dynamic differential equations, stochastic dimensionless aerodynamic forces, stochastic process constraints, stochastic initial value constraints, terminal constraints and control variables, including:

(1) Stochastic Dynamics Differential Equations

V、γ、ψ分别代表飞行器状态变量地心距、经度、纬度、速度、航迹角、航向角；Ω为地球自转角速率；y、V、t、Ω为无量纲化变量，无量纲化参数分别为R₀、V_c＝(g₀R₀)^0.5、(R₀/g₀)^0.5和(g₀/R₀)^0.5，R₀为地球平均半径，g₀为海平面重力加速度。

分别为随机无量纲升力、阻力，无量纲化参数为2m_V·g₀。Among them, y, θ,

V, γ, and ψ represent the state variables of the aircraft, namely the geocentric distance, longitude, latitude, speed, track angle, and heading angle, respectively; Ω is the angular rate of the earth's rotation; y, V, t, and Ω are dimensionless variables, which are dimensionless. The parameters are R ₀ , V _c =(g ₀ R ₀ ) ^0.5 , (R ₀ /g ₀ ) ^0.5 and (g ₀ /R ₀ ) ^0.5 , respectively, where R ₀ is the average earth radius, and g ₀ is the sea-level gravitational acceleration.

are random dimensionless lift and drag, respectively, and the dimensionless parameter is 2m _V ·g ₀ .

(2) Random aerodynamic force

Among them, S _V is the aerodynamic reference area of the aircraft.

(3) Stochastic process constraints

为热流密度，q_r为动压，n_r为过载；

q_max、n_max分别为对应的过程约束最大值。in,

is the heat flux density, q _r is the dynamic pressure, and n _r is the overload;

q _max , n _max are the corresponding process constraint maximum values, respectively.

(4) Random initial value constraint

为随机状态初值。in,

is the initial value of the random state.

(5) Terminal constraints

V_f分别为对应的终端状态设计值。Among them, X _f is the terminal state, y _f , θ _f ,

V _f are the corresponding terminal state design values, respectively.

(6) Control variables

U=[ασ] ^T ;

where α is the angle of attack and σ is the angle of inclination.

(7) Optimize the objective function

J=t _f ;

where t _f is the terminal time.

Further, in the second step, the non-embedded polynomial chaos algorithm is used to carry out quantitative analysis of uncertainty, and the statistical moment information of the state variables, target variables, process constraints and boundary constraints of the flight system is obtained, and the stochastic dynamic differential equations are transformed into high-dimensional The deterministic differential equation of , is solved, and a robust dynamic trajectory optimization model is established; wherein, the robust dynamic trajectory optimization model is in the following form:

为随机状态向量；

为随机状态微分方程转化而来的确定性微分方程，方程数目为n＝1…m^q，其中，q为随机变量向量ξ的维数，m为多项式混沌展开所采用的正交积分点数；C_μ与C_σ分别为随机过程约束的统计矩；B_μ与B_σ分别为随机边界约束的统计矩，t₀为起始时间。Among them, U(t) is the control variable, t is the time variable; J _A is the robust optimization objective, J _μ and J _σ are the statistical moments of the stochastic optimization objective, respectively, and k _μ and k _σ are the corresponding coefficients;

is a random state vector;

is a deterministic differential equation transformed from a stochastic state differential equation, the number of equations is n=1...m ^q , where q is the dimension of the random variable vector ξ, m is the number of orthogonal integration points used in the polynomial chaotic expansion; C _μ and C _σ are the statistical moments of the stochastic process constraints, respectively; B _μ and B _σ are the statistical moments of the random boundary constraints, respectively, and t ₀ is the starting time.

Further, in step 2, according to the random model of the system and the distribution of uncertain factors shown in step 1, the values of uncertain factors are randomly changed and the corresponding robust dynamic trajectory optimization model is solved to obtain a trajectory data set oriented to multi-source uncertainty, Obtain enough offline robust optimal trajectories.

Further, in step 3, the trajectory state and control data pair on the numerical discrete points are used as training data to form a robust trajectory training data set, including:

为轨迹状态变量，U＝[α σ]^T为轨迹控制变量；l＝1,2,…,L代表不同轨迹标号，d＝1,2…,D代表轨迹数值离散点数。在此基础上，利用深度学习技术逼近不确定非线性动力学模型，学习鲁棒轨迹规划和制导最优化策略，建立智能鲁棒最优轨迹和制导模型(f(X^(d)):→U^(d))。in,

is the trajectory state variable, U=[α σ] ^T is the trajectory control variable; l=1,2,...,L represents the different trajectory labels, d=1,2...,D represents the trajectory numerical discrete points. On this basis, use deep learning technology to approximate the uncertain nonlinear dynamic model, learn robust trajectory planning and guidance optimization strategies, and establish an intelligent robust optimal trajectory and guidance model (f(X ^(d) ):→U ^(d) ).

Further, in step 3, the trajectory state vector is used as the input of the deep neural network, and the trajectory control vector is used as the output of the deep neural network, and the training and testing of the deep neural network is carried out to establish the state and control optimality of the uncertain nonlinear flight dynamics model. mapping, including:

In the Python environment, the Keras deep learning package is used to create a DNN. The Keras framework runs with GPU-accelerated Tensorflow as the backend; the trajectory data set is divided into training set and test set according to an appropriate ratio; the deep neural network consists of an input layer, an output layer and multiple hidden layers, the created DNN network is a sequential model, and adjacent layers are connected to each other; the Adam optimization algorithm with excellent performance is used, the loss function adopts the mean square error, and the hidden layer activation function uses ReLU function, each time a certain batch of data is passed in to calculate the loss and update the network parameters to obtain the best training results; finally, the trajectory generated by the optimal control is compared with the trajectory generated by the state quantity driven by the DNN to verify the validity of the model .

Further, in step 4, the deep neural network model with complete training is loaded, according to the flight state information output by the missile-borne control system including the navigation and positioning system and the attitude control system during the re-entry process of the aircraft, to realize the robust trajectory planning guidance instruction. Smart real-time updates, including:

作为网络输入，以控制量U^(d)＝[α σ]^T作为输出，实现鲁棒轨迹规划制导指令智能实时更新。Load the well-trained deep neural network model into the aircraft, and use the real-time flight state quantity output by the missile-borne control system including the navigation and positioning system and the attitude control system during the re-entry process of the aircraft.

As the input of the network, the control variable U ^(d) = [α σ] ^T is used as the output to realize the intelligent real-time update of the guidance instructions for robust trajectory planning.

Another object of the present invention is to provide an intelligent and robust re-entry guidance system for a hypersonic aircraft applying the intelligent and robust re-entry guidance method for a hypersonic aircraft. The intelligent and robust re-entry guidance system for a hypersonic aircraft includes:

Uncertainty modeling module, used for system analysis based on uncertainty, based on data mining and expert experience to judge the uncertainty and its types in the process of gliding and re-entry of hypersonic vehicle, and determine the distribution form and distribution of uncertainty parameters interval;

The uncertainty quantitative analysis module is used to carry out the uncertainty quantitative analysis based on the non-embedded polynomial chaos theory according to the obtained uncertainty factors,

The trajectory optimization model building module is used to establish a robust dynamic trajectory optimization model including the stochastic system objective and the statistical moment characteristics of state variables;

The numerical sample set building module is used to establish a robust trajectory optimization numerical sample set for multi-source uncertain factors by changing the uncertain parameter values and designing a numerical solution strategy;

The neural network model architecture design module is used to design the deep neural network model architecture with the trajectory state vector as the input and the trajectory control vector as the output;

The model training module is used to use the obtained robust trajectory optimization numerical sample set for multi-source uncertain factors to train the model and verify the validity of the model;

The command real-time update module is used to load the well-trained deep neural network model. According to the flight status information output by the missile-borne control system including the navigation and positioning system and the attitude control system during the re-entry process of the aircraft, it can realize robust trajectory planning and guidance command intelligence. Live Update.

Combined with all the above technical solutions, the advantages and positive effects of the present invention are as follows: the intelligent reentry guidance method for hypersonic aircraft based on robust trajectory optimization provided by the present invention is first based on uncertainty system modeling, and based on data mining and Expert experience judges the uncertainties and types in the process of gliding and re-entry of hypersonic vehicles, and determines the distribution form and distribution interval of uncertainty parameters. A robust dynamic trajectory optimization model based on statistical moment information of state variables, by changing the uncertain parameter values and designing a numerical solution strategy, a robust trajectory optimization numerical sample set for multi-source uncertain factors is established; then the trajectory state vector is used as input, Taking the trajectory control vector as the output, a deep neural network model was designed and the sample set was used for training and verification; finally, the fully trained deep neural network model was loaded, and the flight state output by the missile-borne control system such as the navigation and positioning system and the attitude control system was used. information to achieve intelligent real-time update of robust trajectory planning guidance instructions. The invention can realize intelligent online planning of robust re-entry trajectory against multi-source uncertain factors, greatly reduce robust trajectory design time, effectively enhance the active defense capability of guidance instructions against complex uncertainties, thereby effectively reducing the aircraft guidance and control system Design burden.

The invention analyzes the uncertainty parameter distribution form and distribution interval of the re-entry process, so as to realize the robust re-entry trajectory intelligent online planning for multi-source uncertain factors, and then solves the corresponding robust dynamic trajectory optimization model to obtain sufficient The offline robust optimal trajectory is based on the trajectory state and the control data pair on the numerical discrete points as the training data to complete the training of the neural network, and obtain the nonlinear mapping relationship between the state quantity and the control quantity. Compared with the traditional The trajectory optimization method effectively enhances the guidance command's active defense capability against complex multi-source uncertainties, and reduces the design burden of the aircraft guidance and control system.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the following will briefly introduce the accompanying drawings that need to be used in the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

FIG. 1 is a flowchart of an intelligent robust reentry guidance method for a hypersonic aircraft provided by an embodiment of the present invention.

FIG. 2 is a schematic diagram of an intelligent robust reentry guidance method for a hypersonic aircraft provided by an embodiment of the present invention.

FIG. 3 is a structural block diagram of an intelligent robust reentry guidance system for a hypersonic aircraft provided by an embodiment of the present invention.

In the figure: 1. Uncertainty modeling module; 2. Uncertainty quantitative analysis module; 3. Trajectory optimization model building module; 4. Numerical sample set building module; 5. Neural network model architecture design module; 6. Model training module; 7. Instruction real-time update module.

FIG. 4 is a schematic diagram of the robustness verification of the re-entry flight height of the intelligent guidance model provided by the embodiment of the present invention.

FIG. 5 is a schematic diagram of the longitude robustness verification of the re-entry flight of the intelligent guidance model provided by the embodiment of the present invention.

FIG. 6 is a schematic diagram of the latitude robustness verification of the re-entry flight of the intelligent guidance model provided by the embodiment of the present invention.

FIG. 7 is a schematic diagram of the robustness verification of the reentry flight speed of the intelligent guidance model provided by the embodiment of the present invention.

FIG. 8 is a schematic diagram of the robustness verification of the re-entry flight track angle of the intelligent guidance model provided by the embodiment of the present invention.

FIG. 9 is a schematic diagram of the robustness verification of the re-entry flight heading angle of the intelligent guidance model provided by the embodiment of the present invention.

FIG. 10 is a schematic diagram of an application of intelligent robust reentry guidance for a hypersonic aircraft provided by an embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

In view of the problems existing in the prior art, the present invention provides an intelligent and robust reentry guidance method and system for a hypersonic aircraft. The present invention will be described in detail below with reference to the accompanying drawings.

As shown in FIG. 1 , the intelligent robust reentry guidance method for a hypersonic aircraft provided by an embodiment of the present invention includes the following steps:

S101, based on uncertainty system modeling, according to data mining and expert experience, to judge the uncertainty and its type in the process of gliding and re-entry of the hypersonic vehicle, and to determine the distribution form and distribution interval of uncertainty parameters;

S102, according to the uncertain factors obtained in S101, carry out quantitative analysis of uncertainty based on non-embedded polynomial chaos theory, and establish a robust dynamic trajectory optimization model including the stochastic system target and statistical moment characteristics of state variables; Design a numerical solution strategy, and establish a robust trajectory optimization numerical sample set for multi-source uncertain factors;

S103, take the trajectory state vector as input and the trajectory control vector as output, design a deep neural network model architecture; use the robust trajectory optimization numerical sample set for multi-source uncertain factors obtained in S102 to train the model, and verify the validity of the model ;

S104, load the deep neural network model fully trained in S103, and realize intelligent real-time update of robust trajectory planning guidance instructions according to the flight status information output by the missile-borne control system including the navigation and positioning system and the attitude control system during the re-entry process of the aircraft.

The principle diagram of the intelligent robust reentry guidance method for a hypersonic aircraft provided by the embodiment of the present invention is shown in FIG. 2 .

As shown in FIG. 3 , the intelligent robust reentry guidance system for a hypersonic aircraft provided by an embodiment of the present invention includes:

Uncertainty modeling module 1 is used to analyze uncertainties based on the system, judge the uncertainties and types of hypersonic vehicles in the process of gliding and re-entry based on data mining and expert experience, and determine the distribution form of uncertainty parameters and the relationship between them. distribution interval;

Uncertainty quantitative analysis module 2 is used to carry out uncertainty quantitative analysis based on non-embedded polynomial chaos theory according to the obtained uncertainty factors,

The trajectory optimization model building module 3 is used to establish a robust dynamic trajectory optimization model including the stochastic system objective and the statistical moment characteristics of the state variables;

The numerical sample set establishment module 4 is used to establish a robust trajectory optimization numerical sample set for multi-source uncertain factors by changing the uncertain parameter values and designing a numerical solution strategy;

The neural network model architecture design module 5 is used to design the deep neural network model architecture with the trajectory state vector as the input and the trajectory control vector as the output;

The model training module 6 is used for using the obtained robust trajectory optimization numerical sample set for multi-source uncertain factors to perform model training, and to verify the validity of the model;

The command real-time update module 7 is used to load the well-trained deep neural network model, according to the flight status information output by the missile-borne control system including the navigation and positioning system and the attitude control system during the re-entry process of the aircraft, to realize the robust trajectory planning guidance command Smart real-time updates.

The technical solutions of the present invention will be further described below in conjunction with the embodiments.

The technical solution of the present invention will be described in detail by taking the CAV-H aircraft as a specific example.

As shown in Figure 2, the intelligent reentry guidance method for hypersonic aircraft based on robust trajectory optimization includes the following steps:

Step 1: Based on uncertainty system modeling, according to data mining and expert experience, determine the uncertainty and type of the hypersonic vehicle in the process of gliding and re-entry, and determine the distribution form and distribution interval of uncertainty parameters.

In this embodiment, the cognitive uncertainty, random uncertainty, numerical uncertainty, etc. in the high-speed flight dynamics system are classified and modeled. On this basis, the source of uncertainty in the process of robust trajectory optimization design and The uncertainty modeling is carried out around the initial reentry conditions, aerodynamic model data, atmospheric model parameters, and the perturbation parameters of the aircraft itself during the unpowered reentry process of the hypersonic glide vehicle. The specific types of uncertain factors used are:

V₀、γ₀、ψ₀分别代表飞行器状态变量地心距、经度、纬度、速度、航迹角、航向角的初始再入状态，C_L、C_D分别代表飞行器升力系数、阻力系数，ρ代表大气密度，m_V代表飞行器自身质量。Among them, ξ is a random variable vector, which obeys a uniform distribution; the subscripts y ₀ , θ ₀ ,

V ₀ , γ ₀ , ψ ₀ represent the initial re-entry state of the aircraft state variables geocentric distance, longitude, latitude, speed, track angle, and heading angle, respectively, _{C L} and CD represent the lift coefficient and drag coefficient of the aircraft, ρ represents the density of the atmosphere, and m _V represents the mass of the aircraft itself.

The distribution form of uncertain factors specifically adopted in this embodiment is assumed to be uniformly distributed and independent of each other, and the system random model and the distribution range of uncertain factors are:

分别表示飞行器随机初始再入状态；

代表随机升力系数、随机阻力系数；

为随机大气密度；

为随机飞行器质量；in,

respectively represent the random initial re-entry state of the aircraft;

Represents random lift coefficient and random drag coefficient;

is the random atmospheric density;

is the mass of the random aircraft;

V₀＝7200m·s^-1，γ₀＝-2°，ψ₀＝58°，其中地球平均半径R₀＝6378000m；The initial re-entry state of the aircraft y ₀ =100km+R ₀ , θ ₀ =160°,

V ₀ =7200m·s ^-1 , γ ₀ =-2°, ψ ₀ =58°, where the average earth radius R ₀ =6378000m;

其中

马赫数

声速v_S≈295.188m/s，α为攻角；The two-variable aerodynamic coefficient model and nonlinear least squares method are used to fit and calculate the lift coefficient and drag coefficient of the aircraft, and the expressions are as follows:

in

Mach number

The speed of sound v _S ≈ 295.188m/s, and α is the angle of attack;

其中，海平面大气密度ρ₀＝1.2258kg/m³，海拔高度h＝y-R₀，系数β₀＝1.3785×10^-4m^-1；Using a general exponential atmospheric model, the atmospheric density

Among them, the sea level air density ρ ₀ =1.2258kg/m ³ , the altitude h=yR ₀ , the coefficient β ₀ =1.3785×10 ^-4 m ^-1 ;

Aircraft mass m _V =907.2kg.

Step 2: Carry out quantitative analysis of uncertainty based on non-embedded polynomial chaos theory, establish a robust dynamic trajectory optimization model including stochastic system objectives and statistical moment characteristics of state variables, design numerical solution strategies, and establish robust solutions for multi-source uncertain factors. Rod Trajectory Optimization Numerical Sample Collection.

The adopted aircraft stochastic trajectory optimization model includes stochastic dynamic differential equations, stochastic dimensionless aerodynamic forces, stochastic process constraints, stochastic initial value constraints, terminal constraints, control variables, etc. The details are as follows:

The stochastic dynamic differential equation is:

V、γ、ψ分别代表飞行器状态变量地心距、经度、纬度、速度、航迹角、航向角；Ω＝7.2722×10^-5rad/s为地球自转角速率；y、V、t、Ω为无量纲化变量，无量纲化参数分别为R₀、V_c＝(g₀R₀)^0.5、(R₀/g₀)^0.5和(g₀/R₀)^0.5，地球平均半径R₀＝6378000m，海平面重力加速度g₀＝9.80665m/s²。

分别为随机无量纲升力、阻力，无量纲化参数为2m_V·g₀。Among them, y, θ,

V, γ, and ψ represent the state variables of the aircraft's geocentric distance, longitude, latitude, speed, track angle, and heading angle, respectively; Ω=7.2722×10 ^-5 rad/s is the angular rate of the earth's rotation; y, V, t, Ω are dimensionless variables, and the dimensionless parameters are R ₀ , V _c =(g ₀ R ₀ ) ^0.5 , (R ₀ /g ₀ ) ^0.5 and (g ₀ /R ₀ ) ^0.5 , and the average earth radius R ₀ = 6378000m, sea level gravitational acceleration g ₀ =9.80665m/s ² .

are random dimensionless lift and drag, respectively, and the dimensionless parameter is 2m _V ·g ₀ .

The random aerodynamic force is:

Among them, the aircraft aerodynamic reference area S _V =0.4839m ² .

The random process constraints are:

为热流密度，q_r为动压，n_r为过载；

q_max、n_max分别为对应的过程约束最大值，

q_max＝400kPa，n_max＝6。in,

is the heat flux density, q _r is the dynamic pressure, and n _r is the overload;

q _max , n _max are the corresponding maximum process constraints, respectively,

q _max =400 kPa, n _max =6.

The random initial value constraint is:

The terminal constraints are:

V_f＝1000m·s^-1。Among them, y _f =20km+R ₀ , θ _f =236°,

V _f =1000 m·s ^-1 .

The control variables are:

U=[ασ] ^T (8)

Among them, the angle of attack α∈[10°, 20°], the angle of inclination σ∈[-80°, 80°].

The optimization objective function is:

J=t _f (9)

where t _f is the terminal time.

Then, the non-embedded polynomial chaos algorithm is used for quantitative analysis of uncertainty, and the statistical moment information of the state variables, target variables, process constraints, and boundary constraints of the flight system is obtained, and the stochastic dynamic differential equation is transformed into a high-dimensional deterministic differential equation. After solving, the finally established robust dynamic trajectory optimization model is in the following form:

为随机状态向量；

为随机状态微分方程转化而来的确定性微分方程，方程数目为n＝1…m^q，其中，q为随机变量向量ξ的维数，本例中q＝10，m为多项式混沌展开所采用的正交积分点数，本例中m＝6；C_μ与C_σ分别为随机过程约束的统计矩；B_μ与B_σ分别为随机边界约束的统计矩，t₀为起始时间。Among them, U(t) is the control variable, t is the time variable; J _A is the robust optimization objective, J _μ and J _σ are the statistical moments of the stochastic optimization objective, respectively, and k _μ and k _σ are the corresponding coefficients;

is a random state vector;

It is a deterministic differential equation transformed from a stochastic state differential equation. The number of equations is n=1...m ^q , where q is the dimension of the random variable vector ξ, in this example q=10, m is the polynomial chaos expansion used The number of orthogonal integration points of , m=6 in this example; C _μ and C _σ are the statistical moments of the stochastic process constraints respectively; B _μ and B _σ are the statistical moments of the random boundary constraints, respectively, and t ₀ is the starting time.

Considering the requirements of computational accuracy and computational burden, 6 Gaussian sampling points are used in the second-order non-embedded polynomial chaotic expansion in the simulation analysis; the robust dynamic optimization model is solved by particle swarm algorithm, the particle population size is 50, and the maximum iteration The number of times is 100, the number of discrete nodes is 100, and the number of integration points is 500.

According to the system random model and the distribution of uncertain factors shown in formula (1) and formula (2), the values of uncertain factors are randomly changed and the corresponding robust dynamic trajectory optimization model is solved to obtain 4000 offline robust optimal trajectories.

Step 3: Take the trajectory state vector as input and the trajectory control vector as output, design a deep neural network model architecture, use the robust trajectory optimization numerical sample set for multi-source uncertainties obtained in step 2 to train the model, and verify the model effectiveness.

其中，

为轨迹状态变量，U＝[α_cσ_c]^T为轨迹控制变量；l＝1,2,…,L代表不同轨迹标号，d＝1,2…,D代表轨迹数值离散点数。在此基础上，利用深度学习技术逼近不确定非线性动力学模型，学习鲁棒轨迹规划和制导最优化策略，建立智能鲁棒最优轨迹和制导模型(f(X^(d)):→U^(d))。Specifically, the trajectory state and control data pairs on the numerical discrete points are used as training data to form a robust trajectory training data set, and by collecting the state and control pairs, a robust trajectory training data set is formed.

in,

is the trajectory state variable, U=[α _c σ _c ] ^T is the trajectory control variable; l=1,2,...,L represents different trajectory labels, d=1,2...,D represents the number of discrete points of the trajectory value. On this basis, use deep learning technology to approximate the uncertain nonlinear dynamic model, learn robust trajectory planning and guidance optimization strategies, and establish an intelligent robust optimal trajectory and guidance model (f(X ^(d) ):→U ^(d) ).

In this example, 4000 robust optimal trajectories are optimized in MATLAB, and each trajectory contains 500 numerical discrete points, and then a numerical sample set of robust trajectory optimization oriented to multi-source uncertain factors is obtained. The robust state in the data set is the same as The number of control data pairs is 4000×500 pairs.

Before the deep neural network training, in order to improve the network training efficiency and enable the neural network to train faster and better, normalize the trajectory state variables in the data set, and compress the data size to [0, 1] In the range of , the normalization method is:

In this embodiment, the data set is randomly divided according to the ratio of 75% of the training set and 25% of the test set, wherein the training set is used for training the deep neural network. After many iterations, when the value of the loss function reaches the required error or the maximum iteration When the number of times, network training is completed; the test set is used to verify and test the performance of the trained network, and the evaluation indicators include mean square error (MSE), mean absolute error (MAE), and accuracy.

The deep neural network consists of an input layer, an output layer and multiple hidden layers. The deep neural network in this embodiment is constructed using the Keras2.3.1 deep learning package in the Python3.8.3 environment, and the Keras framework is based on GPU-accelerated Tensorflow2. 2.0 runs as the back-end, defining the network model as a sequential model, and the adjacent layers are fully connected to each other.

为，包含6个状态参数，y、θ、

V、γ、ψ分别代表飞行器状态变量地心距、经度、纬度、速度、航迹角、航向角。Input layer: the state quantity of the aircraft at time d

is, contains 6 state parameters, y, θ,

V, γ, and ψ represent the state variables of the aircraft, namely the center-to-center distance, longitude, latitude, speed, track angle, and heading angle, respectively.

Hidden layer: Determine the appropriate number of hidden layers and the number of nodes in each layer through general experience and experimental results of repeated training; the activation function selects the ReLU function that can avoid the problem of gradient disappearance in most cases and converge faster, expressing The formula is:

Output layer: Since the output control quantity U ^(d) = [ασ] ^T , it does not involve the classification problem, but only outputs the predicted value of the network, and the output layer does not need to call the activation function.

The measurement of the network output error is the loss function. The larger the loss value, the worse the prediction accuracy of the network. In this embodiment, the mean square error is used as the loss function, and the expression is:

Among them, C represents the error size of each batch, X ^d represents the input state quantity, U _s represents the real value of the control quantity, U _p represents the network output value, and N _b represents the size of a batch of data provided to the network for training each time.

In the data set used in this embodiment for training and testing the network, the number of pairs of robust state and control data is 2×10 ⁶ pairs. Considering the large data set and limited computing resources of the computer, the data cannot be transferred in one time. network, and only relying on one data to update the parameters at a time will cause the loss function to fluctuate greatly during the training process, resulting in the model not converging, so each time a certain batch of data sets are imported into the network for training, this example selects N _b =1024.

When configuring the model, select the excellent Adam optimization algorithm to train the network to make the network converge faster and better. The algorithm parameter settings follow the default values.

The setting of the number of network training epochs also affects the predictive ability of the model. Considering the computational complexity issue, the more epochs of training, the more accurate the results may be, but the corresponding training time also increases. In this example, first set the number of training rounds to 200 to select the appropriate hidden layer structure, and increase the number of training rounds under the optimal structure to make the network fully converge, and save the model to predict the control amount of the aircraft.

In the selection of the hidden layer structure of the network, by fixing the batch size of each training to 1024 and the number of training rounds to 200, the influence of different hidden layer structures on the prediction accuracy of the network is compared. Table 1 compares and analyzes the mean square error (MSE) size and accuracy on the training set and test set obtained with different numbers of hidden layers and nodes.

Table 1 MSE and accuracy results under different hidden layer structures

It can be seen from Table 1 that a relatively shallow network with a limited number of units will reduce the ability of the network to approximate the optimal control structure, resulting in insufficient training; increasing the number of units will reduce the Tr-MSE and Te-MSE values and increase the accuracy. Aspects have a positive effect, and more layers of the network tend to bring more benefits. However, when the network scale is too large, the computational complexity will increase accordingly, the network convergence speed will become slower, and the improvement of training result indicators will not be obvious. After the network is finally mounted on the aircraft, the calculation amount of forward prediction will also increase with the depth. If the size of the network is too large, it may cause the network to be over-trained, resulting in over-fitting. In this embodiment, a structure in which the hidden layer is 6/64 is selected according to the computing power of the computer and the training index of the network.

In order to determine the impact of different training rounds on the accuracy of deep neural networks, Table 2 lists the same hidden layer structure, that is, the number of hidden layers is 6, the number of nodes in each layer is 64, and the same batch size of 1024, different training rounds. Count the obtained training set and test set evaluation index results.

Table 2 Results of network evaluation indicators under different training rounds

It can be seen from Table 2 that in the initial stage, with the continuous increase of the number of training rounds, the error and accuracy of the training set and test set show an overall improvement trend, but when the number of training rounds is large, the training results on the test set tend to be better. Sex is no longer obvious, but instead shows constant fluctuations within a certain margin of error. In addition, the training time of the network also increases with the number of epochs. In this example, the DNN model with 1000 training rounds is finally selected and saved for calling and testing.

Step 4: Test the performance of the DNN-driven control scheme. In the Windows 10 operating system 8G memory i5-7300HQ CPU environment, the DNN completes the trajectory prediction of 500 discrete points in only 0.04s, and has the ability to meet the intelligent real-time update of robust trajectory planning guidance instructions.

In order to verify the robustness of the robust trajectory planning guidance command, considering the range of uncertain factors such as re-entry initial value and atmospheric parameters used in the above-mentioned offline robust optimization design stage, the DNN model is used to predict the robust trajectory control vector U _C , and solve the dynamic equation based on the fourth-order Runge-Kutta integral algorithm, and calculate the corresponding trajectory state vector X _C . On this basis, increase the value range of the original uncertainty factors by 15%, 20%, and 25% respectively, and calculate the corresponding DNN control vector and state vector in turn to obtain

For the re-entry trajectory states obtained in the above four cases, the robustness of the trajectory planning guidance of the DNN model is analyzed, and the obtained aircraft trajectory state variable results are shown in Figures 4 to 9, respectively. It can be seen that due to the use of a robust trajectory optimization numerical sample set for multi-source uncertainty factors for training, after increasing the value range of the original uncertainty factors by 15%, 20%, and 25%, the DNN model-driven The trajectory state vectors are almost coincident, and the guidance command maintains good robustness.

To sum up, the present invention provides an intelligent reentry guidance method for hypersonic aircraft based on robust trajectory optimization, and its application schematic diagram is shown in FIG. 10 . Compared with the traditional trajectory optimization method, this method can enhance the active defense capability of the guidance command against complex multi-source uncertainties, and at the same time effectively reduce the design burden of the aircraft guidance and control system.

In the above-mentioned embodiments, it may be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented in whole or in part in the form of a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer may be a general purpose computer, special purpose computer, computer network, or other programmable device. The computer instructions may be stored in or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions may be downloaded from a website site, computer, server or data center Transmission to another website site, computer, server, or data center by wire (eg, coaxial cable, fiber optic, digital subscriber line (DSL), or wireless (eg, infrared, wireless, microwave, etc.)). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, a data center, or the like that includes an integration of one or more available media. The usable media may be magnetic media (eg, floppy disks, hard disks, magnetic tapes), optical media (eg, DVD), or semiconductor media (eg, Solid State Disk (SSD)), among others.

The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited to this. Any person skilled in the art is within the technical scope disclosed by the present invention, and all within the spirit and principle of the present invention Any modifications, equivalent replacements and improvements made within the scope of the present invention should be included within the protection scope of the present invention.

Claims (10)

1. An intelligent robust reentry guidance method for a hypersonic aircraft is characterized by comprising the following steps:

based on uncertainty system modeling, judging uncertainty and types thereof existing in the gliding reentry process of the hypersonic aerocraft according to data mining and expert experience, and determining the distribution form and distribution interval of uncertainty parameters;

carrying out uncertainty quantitative analysis based on a non-embedded polynomial chaos theory according to the obtained uncertainty factors, and establishing a robust dynamic trajectory optimization model containing the statistical moment characteristics of a random system; establishing a robust track optimization numerical sample set aiming at the multi-source uncertain factors by changing uncertain parameter values and designing a numerical solving strategy;

designing a deep neural network model architecture by taking the track state vector as input and the track control vector as output; performing model training by using the obtained robust track optimization numerical sample set aiming at the multi-source uncertain factors, and verifying the effectiveness of the model;

and loading a deep neural network model with complete training, and intelligently updating the robust trajectory planning guidance instruction in real time according to flight state information output by a missile-borne control system including a navigation positioning system and an attitude control system in the reentry process of the aircraft.

2. The hypersonic aircraft intelligent robust reentry guidance method of claim 1, wherein the modeling based on the uncertainty system, judging the uncertainty and the type thereof existing in the process of hypersonic aircraft gliding reentry according to data mining and expert experience, and determining the distribution form and the distribution interval of the uncertainty parameters comprises:

carrying out classification modeling aiming at cognitive uncertainty, random uncertainty and numerical uncertainty in a high-speed flight dynamics system, researching uncertainty sources and types in a robust trajectory optimization design process on the basis, and carrying out uncertainty modeling around initial reentry conditions, pneumatic model data, atmospheric model parameters and perturbation parameters of an aircraft in the unpowered reentry process of the hypersonic glide aircraft;

wherein the uncertain factor types are:

wherein xi is a random variable vector; subscript y₀、θ₀、

V₀、γ₀、ψ₀Respectively representing the initial reentry states of the aircraft state variables such as the earth's center distance, longitude, latitude, speed, track angle and course angle, C_L、C_DRespectively represents the lift coefficient and the drag coefficient of the aircraft, rho represents the atmospheric density, m_VRepresenting the aircraft's own mass.

3. The hypersonic aircraft intelligent robust reentry guidance method according to claim 1, wherein the uncertainty distribution form is assumed to be uniformly distributed and independent from each other, and the system stochastic model and the uncertainty distribution range are as follows:

wherein,

respectively representing the random initial reentry state of the aircraft;

representing a random lift coefficient and a random drag coefficient;

is a random atmospheric density;

is a random aircraft mass.

4. The hypersonic aircraft intelligent robust reentry guidance method of claim 1, wherein the aircraft random trajectory optimization model comprises random dynamical differential equations, random dimensionless aerodynamic forces, random process constraints, random initial value constraints, terminal constraints and control variables, including:

(1) differential equation of stochastic dynamics

Wherein y, theta,

V, gamma and psi respectively represent the geocentric distance, longitude, latitude, speed, track angle and heading angle of the aircraft state variables; omega is the rotation angular rate of the earth; y, V, t and omega are dimensionless variables, R is the dimensionless parameter₀、V_c＝(g₀R₀)^0.5、(R₀/g₀)^0.5And (g)₀/R₀)^0.5，R₀Is the mean radius of the earth, g₀Is seaPlanar gravitational acceleration;

respectively random dimensionless lift force and resistance force, and dimensionless parameters of 2m_V·g₀；

(2) Random aerodynamic force

Wherein S is_VAn aerodynamic reference area for the aircraft;

(3) random process constraints

Wherein,

is the heat flow density, q_rIs dynamic pressure, n_rIs overloaded;

q_max、n_maxrespectively corresponding process constraint maximum values;

(4) random initial value constraint

Wherein,

is a random state initial value;

(5) terminal constraints

Wherein, X_fIs in the terminal state, y_f、θ_f、

V_fRespectively corresponding terminal state design values;

(6) controlled variable

U＝[ασ]^T；

Wherein alpha is an attack angle and sigma is an inclination angle;

(7) optimizing an objective function

J＝t_f；

Wherein, t_fIs the terminal time.

5. The hypersonic aircraft intelligent robust reentry guidance method according to claim 1, characterized in that a non-embedded polynomial chaotic algorithm is adopted for uncertainty quantitative analysis to obtain statistical moment information of flight system state variables, target variables, process constraints and boundary constraints, and meanwhile, a stochastic dynamics differential equation is converted into a high-dimensional deterministic differential equation to be solved, and a robust dynamic trajectory optimization model is established; wherein the robust dynamic trajectory optimization model is in the form of:

wherein U (t) is a control variable, and t is a time variable; j. the design is a square_AFor robust optimization of the objective, J_μAnd J_σSeparately optimizing the target statistical moment, k, for random_μAnd k is_σIs the corresponding coefficient;

is a random state vector;

derived for random state differential equation transformationQualitative differential equation with n being 1 … m^q(ii) a Wherein q is the dimension of a random variable vector xi, and m is the number of orthogonal integration points adopted by the polynomial chaotic expansion; c_μAnd C_σStatistical moments that are respectively random process constraints; b is_μAnd B_σStatistical moments, t, respectively, of random boundary constraints₀Is the starting time.

6. The hypersonic aircraft intelligent robust reentry guidance method according to claim 1 is characterized in that according to the system random model and uncertainty factor distribution shown in claims 2 and 3, uncertainty factor values are changed randomly and a corresponding robust dynamic trajectory optimization model is solved to obtain a trajectory data set and obtain a sufficient offline robust optimal trajectory.

7. The hypersonic aircraft intelligent robust reentry guidance method of claim 1, wherein the trajectory state and control data pairs on the numerical discrete points are used as training data to form a robust trajectory training data set, and the robust trajectory training data set comprises:

wherein,

for the track state variable, U ═ α σ]^TIs a trajectory control variable; l is 1,2, …, L represents different track labels, D is 1,2 …, and D represents track numerical discrete points; on the basis, an uncertain nonlinear dynamics model is approximated by using a deep learning technology, a robust track planning and guidance optimization strategy is learned, and an intelligent robust optimal track and guidance model (f (X) is established^(d)):→U^(d))。

8. The hypersonic aircraft intelligent robust reentry guidance method of claim 1, wherein the deep neural network is trained and tested by taking a track state vector as the input of the deep neural network and a track control vector as the output of the deep neural network, and uncertain nonlinear flight dynamics model state and control optimality mapping is established, comprising:

under a Python environment, using a Keras deep learning package to create DNN, and operating a Keras framework by taking Tensorflow accelerated based on a GPU as a back end; dividing a track data set into a training set and a testing set according to a proper proportion; the deep neural network consists of an input layer, an output layer and a plurality of hidden layers, the created DNN network is a sequential model, and adjacent layers are connected with each other; adopting an Adam optimization algorithm with excellent performance, adopting a mean square error as a loss function, adopting a ReLU function as a hidden layer activation function, transmitting a certain batch of data each time to calculate loss and update network parameters to obtain an optimal training result; and finally, comparing the track generated by the optimal control with the track generated by the state quantity generated by the DNN drive, and verifying the validity of the model.

9. The hypersonic aircraft intelligent robust reentry guidance method of claim 1, wherein the loading of the deep neural network model with complete training realizes intelligent real-time updating of the robust trajectory planning guidance instruction according to flight state information output by a missile-borne control system including a navigation positioning system and an attitude control system in the reentry process of the aircraft, and comprises the following steps:

loading the deep neural network model which is completely trained on the aircraft, and using the real-time flight state quantity output by the missile-borne control system including the navigation positioning system and the attitude control system in the reentry process of the aircraft

As network input, to control the quantity U^(d)＝[α σ]^TAnd as output, intelligent real-time updating of the robust track planning guidance instruction is realized.

10. A hypersonic aircraft intelligent robust reentry guidance system for executing the hypersonic aircraft intelligent robust reentry guidance method according to any one of claims 1 to 9, wherein the hypersonic aircraft intelligent robust reentry guidance system comprises:

the uncertainty modeling module is used for judging uncertainty and types thereof existing in the gliding reentry process of the hypersonic flight vehicle according to data mining and expert experience based on uncertainty system analysis and determining the distribution form and the distribution interval of uncertainty parameters;

the uncertainty quantitative analysis module is used for developing uncertainty quantitative analysis based on a non-embedded polynomial chaos theory according to the obtained uncertainty factors;

the track optimization model building module is used for building a robust dynamic track optimization model containing the random system target and the state variable statistical moment characteristics;

the numerical sample set establishing module is used for establishing a robust track optimization numerical sample set aiming at the multi-source uncertain factors by changing uncertain parameter values and designing a numerical solving strategy;

the neural network model architecture design module is used for designing a deep neural network model architecture by taking the track state vector as input and taking the track control vector as output;

the model training module is used for performing model training by using the obtained robust track optimization numerical sample set aiming at the multi-source uncertain factors and verifying the effectiveness of the model;

and the instruction real-time updating module is used for loading a deep neural network model with complete training and realizing intelligent real-time updating of the robust track planning guidance instruction according to flight state information output by a missile-borne control system including the navigation positioning system and the attitude control system in the reentry process of the aircraft.

Images