A guidance method and system for time constraints in the atmosphere based on physical information neural network

Abstract

The invention relates to the field of intelligent guidance and control of aircraft, and discloses a guidance method and system with time constraints in the atmosphere based on a physical information neural network. The specific steps are as follows: on the basis of a proportional guidance law, a missile flight state and a remaining flight time data set under different missile initial states are constructed, then the remaining flight time calculated by the physical information neural network is used to construct a time constraint bias item, a time constraint guidance law is designed, and finally, based on a trained physical information neural network model and the time constraint guidance law, an aircraft is guided to strike a target at a desired time. The invention adopts a method combining proportional guidance with a physical information neural network to realize guidance control, which has very important engineering significance for considering the influence of air resistance and gravity and realizing the constraint on the missile attack time.

Description

Technical Field

[0001] The present invention relates to the field of intelligent guidance and control of aircraft, and in particular to a guidance method and system based on physical information neural network and time constraints within the atmosphere. Background Art

[0002] In recent years, with the rapid development of anti-missile and air defense technology, the traditional proportional guidance law can only meet the constraint of terminal miss error, but cannot meet the missile strike time constraint. More and more research is based on offset proportional guidance to achieve strike time constraint. The design of a single missile guidance law considering time constraints can enable the missile to hit the target at the expected strike time, and can give multiple missiles the same strike time to achieve multi-missile coordinated strikes, improve the terminal penetration probability of the missile, and break the interception of the enemy's air defense and anti-missile system.

[0003] The existing time-constrained guidance of a single missile can be generally divided into two categories according to whether the remaining flight time of the missile is required. The first category is the method that requires the estimation of the remaining flight time. This method usually requires the approximate solution of the remaining flight time of the missile, and then constructs the error feedback guidance law based on the remaining flight time and the expected strike time. The second category does not require the calculation of the remaining flight time of the missile, and is usually implemented by leading angle tracking, line of sight angular rate tracking or missile-target distance tracking. At present, the calculation of the remaining flight time under the assumption of constant missile speed is very mature, but there is little research on the time-constrained guidance law considering the influence of air resistance and gravity. The difficulty in designing the time-constrained guidance law considering the time-varying missile speed lies in how to accurately calculate the remaining flight time of the missile. Dongwei proposed a method for calculating the remaining flight time of a time-varying speed missile by numerical iterative integration of an improved Euler method. Kang Honglong proposed a method for calculating the future speed of the missile by numerical iterative integration in a two-dimensional plane considering only the influence of air resistance. He Shaoming proposed a method for predicting the remaining flight time based on a deep neural network. Among the above-mentioned existing methods, numerical iterative integration brings a heavy load burden to the guidance and control system; deep neural networks have a great dependence on training data, resulting in poor scalability for scenarios that have not been trained.

[0004] Therefore, it is necessary to design a guidance control law that can solve the time-constrained guidance problem in a three-dimensional combat scenario, taking into account the influence of air resistance and gravity. Summary of the invention

[0005] The purpose of the present invention is to provide a time-constrained guidance method and system within the atmosphere based on a physical information neural network to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a guidance method based on physical information neural network with time constraints in the atmosphere, comprising the following steps:

[0007] Step 1: Based on the proportional guidance law, a data set of missile flight state parameters and remaining flight time parameters at different initial states is constructed;

[0008] Step 2: Use the data set constructed in step 1 to train the neural network model;

[0009] Step 3: Based on the neural network model trained in step 2, the real-time remaining flight time of the missile is predicted, and the time constraint bias term is constructed using the expected strike time and the remaining flight time of the missile to design the time constraint guidance law.

[0010] Step 4: Based on the neural network model trained in step 2 and the time-constrained guidance law proposed in step 3, the missile is guided to hit the target at the desired time.

[0011] Preferably, the neural network model of step 2 is a neural network model that takes physical information constraints into consideration.

[0012] Preferably, the missile flight state parameters in different initial states of step 1 specifically include missile-target distance, lead angle, missile flight speed, and the remaining flight time parameter specifically is the actual remaining flight time of the missile.

[0013] Preferably: the engagement model between the missile and the target in the guidance process is specifically described as follows: the sight vector between the missile and the target is R, the missile velocity vector is described as V, and The distance between the missile and the target is described as R = ||R||, the missile's velocity is described as V = ||V||, and the rotation angular rates of the sight vector and the missile's velocity vector are respectively

[0014]

[0015]

[0016] Where a is the missile acceleration vector perpendicular to the velocity vector,

[0017] The total missile lead angle is defined as:

[0018]

[0019] The vector perpendicular to the plane formed by the line of sight vector R and the missile velocity vector V is defined as:

[0020] k=R×V / ||R×V||.

[0021] Preferably: the speed of the missile in the guidance process is affected by air resistance and gravity acceleration, specifically:

[0022]

[0023] Where T is the missile thrust, q is the air dynamic pressure, S is the missile area, m is the missile mass, g is the gravity acceleration vector, g n is the component of gravity in the direction of the missile's velocity, C D0 is the zero lift drag coefficient, K D is the induced drag coefficient.

[0024] Preferably, step 2 uses the data set constructed in step 1 and the randomly generated prior model data set to perform neural network model training, specifically including:

[0025] Step 2-1, design a physical information neural network to predict the remaining flight time of the missile. The specific expression is as follows:

[0026]

[0027] Among them, t go is the output of the physical information neural network, is the prior model, Δ(R,σ,V) is the output of the deep neural network, which is finally Together with Δ(R,σ,V), it forms the expression for the remaining flight time of the missile;

[0028] Step 2-2, the deep neural network model described in step 2-1 has two hidden layers, and the number of neurons in the hidden layer is 100;

[0029] Step 2-3, the loss function is designed as,

[0030]

[0031] in, is the remaining flight time predicted by the physical information neural network, y i is the actual remaining flight time, N is the number of training data sets, is the response of the physical information neural network to the prior model data, A priori model Calculate the remaining flight time, is the number of prior model data sets;

[0032] Step 2-4, the model training parameters of the physical information neural network are: the number of iterations is 20000, the learning rate is 0.0001, and the activation function uses the ReLU function;

[0033] Step 2-5, using the 95387713 training sample data and 234431941 prior model sample data obtained in step 1, special physical information neural network training is performed to finally obtain a trained neural network model.

[0034] Preferably, step 3 predicts the real-time remaining flight time of the missile based on the neural network model trained in step 2, constructs a time constraint bias term using the expected strike time and the real-time remaining flight time of the missile, and designs a time constraint guidance law, specifically including:

[0035] Step 3-1, set the expected missile strike time T d ;

[0036] Step 3-2, using the neural network model trained in step 2, real-time input: missile-target distance R, lead angle σ, speed V, get the output of the neural network model: the remaining flight time t of the missile go ;

[0037] Step 3-3, using T in step 3-1 d and t in step 3-2 go , construct the strike time error e = T d -t go , the construction time bias term is,

[0038]

[0039] Where κ is the time error coefficient,

[0040] Step 3-4, based on the time error bias term Ω designed in step 3-3 b , the time-constrained guidance law is designed as

[0041] A M =(NΩ r +Ω b )×V.

[0042] According to the above-mentioned guidance system with time constraints in the atmosphere based on physical information neural network, the system includes a missile kinematics iteration module, a missile remaining flight time calculation module based on physical information neural network and a missile guidance control module, wherein the missile kinematics iteration module is used to calculate the real-time position data of the missile, and then provide real-time guidance kinematics data; the missile remaining flight time calculation module based on physical information neural network is used for real-time missile remaining flight time calculation, and then provides parameter input for the design of time-constrained guidance law; the missile guidance control module solves the missile's overload instruction based on the designed time-constrained guidance law, the pre-set strike time and the real-time remaining flight time, and guides the missile to strike the target with the expected flight time.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The invention proposes a guidance method and system for time constraints in the atmosphere based on a physical information neural network. The method does not rely on the assumption of a constant missile speed, but takes into account the influence of air resistance and gravity in the atmosphere on the missile speed. At the same time, the physical information neural network model is adopted, which can have good scalability for untrained scene parameters, improves the prediction accuracy of the neural network model, predicts the remaining flight time of the missile based on the proportional guidance law through the physical information neural network, and drives a single missile to strike a target at a desired time based on the proposed time constraint bias proportional guidance law. The remaining flight time of the missile is calculated clearly and quickly, and no numerical iterative calculation process is involved, so that a large load will not be placed on the guidance control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A vector guidance model for the missile of the present invention;

[0046] Figure 2 A flow chart of a guidance method for time constraints in the atmosphere based on a physical information neural network according to the present invention;

[0047] Figure 3 It is the framework of the physical information neural network of the present invention;

[0048] Figure 4 It is the change curve of the loss function of the physical information neural network of the present invention;

[0049] Figure 5 The three-dimensional trajectory curve of the guidance control of a single missile with a strike time constraint of 60s;

[0050] Figure 6 A variation curve of the remaining flight time of a single missile using the time-constrained guidance law of the present invention;

[0051] Figure 7An overload variation curve for a single missile using the time-constrained guidance law of the present invention;

[0052] Figure 8 This is the speed change curve of a single missile. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] Example

[0055] See also Figure 1-Figure 8 , a guidance method based on physical information neural network and time constraints in the atmosphere, comprising the following steps:

[0056] Step 1: Based on the proportional guidance law, a data set of missile flight state parameters and remaining flight time parameters at different initial states is constructed;

[0057] Step 2: Use the data set constructed in step 1 to train the neural network model;

[0058] Step 3: Based on the neural network model trained in step 2, the real-time remaining flight time of the missile is predicted, and the time constraint bias term is constructed using the expected strike time and the remaining flight time of the missile to design the time constraint guidance law.

[0059] Step 4: Based on the neural network model trained in step 2 and the time-constrained guidance law proposed in step 3, the missile is guided to hit the target at the desired time.

[0060] Figure 1 The vector guidance model of the missile of the present invention is combined with Figure 1 , the engagement model between missiles and stationary targets is specifically described as:

[0061] The line of sight vector between the missile and the target is R, the velocity vector of the missile is described as V, and The distance between the missile and the target is described as R = ||R||, the missile's velocity is described as V = ||V||, and the rotation angular rates of the sight vector and the missile's velocity vector are,

[0062]

[0063]

[0064] Where a is the acceleration vector perpendicular to the missile velocity vector.

[0065] The total missile lead angle is defined as:

[0066]

[0067] The vector perpendicular to the plane formed by the line of sight vector R and the missile velocity vector V is defined as:

[0068] k=R×V / ||R×V||

[0069] Furthermore, the speed of the missile is affected by air resistance and gravity acceleration. The specific mathematical relationship is as follows:

[0070]

[0071] Where T is the missile thrust, q is the air dynamic pressure, S is the missile area, m is the missile mass, g is the gravity acceleration vector, g n is the component of gravity in the direction of the missile's velocity, C D0 is the zero lift drag coefficient, K D is the induced drag coefficient.

[0072] Furthermore, the data set of missile flight state parameters and remaining flight time parameters in different initial states is constructed based on the proportional guidance law described in step 1, specifically including:

[0073] Step 1-1, randomly initialize the initial parameters of the missile. The initial x position range of the missile is [-30km, 30km], the initial y position range is [-40km, 0km], the initial z position range is [0km, 40km], the initial trajectory inclination range is [-50°, 60°], the initial trajectory deviation range is [-90°, 90°], and the initial speed change range is [200m / s, 500m / s];

[0074] Step 1-2, the missile in step 1-1 successfully hits the target based on the proportional guidance law. The expression of the proportional guidance law is:

[0075] A PNG =NΩ r ×V

[0076] Wherein, N is the proportional guidance coefficient;

[0077] Step 1-3, construct a data set of missile flight state parameters and remaining flight time. The missile flight state parameters specifically include (R, σ, V), and finally form (R, σ, V) and remaining flight time t go At the same time, the constructed data set is normalized using the maximum-minimum principle;

[0078] Step 1-4, repeat step 1-1 to step 1-3, generate a total of 20,000 trajectories, a total of 95,387,713 sample data.

[0079] Furthermore, the step 2 uses the data set constructed in step 1 and the randomly generated prior model data set to perform neural network model training, specifically including:

[0080] Step 2-1, design a physical information neural network to predict the remaining flight time of the missile. The specific expression is as follows:

[0081]

[0082] Among them, t go is the output of the physical information neural network, is the prior model, Δ(R,σ,V) is the output of the deep neural network, which is finally Together with Δ(R,σ,V), it forms the expression for the remaining flight time of the missile;

[0083] Step 2-2, the deep neural network model described in step 2-1 has two hidden layers, and the number of neurons in the hidden layer is 100;

[0084] Step 2-3, the loss function is designed as,

[0085]

[0086] in, is the remaining flight time predicted by the physical information neural network, y i is the actual remaining flight time, N is the number of training data sets, is the response of the physical information neural network to the prior model data, A priori model Calculate the remaining flight time, is the number of prior model data sets;

[0087] Step 2-4, the model training parameters of the physical information neural network are: the number of iterations is 20000, the learning rate is 0.0001, and the activation function uses the ReLU function;

[0088] Step 2-5, using the 95387713 training sample data and 234431941 prior model sample data obtained in step 1, special physical information neural network training is performed to finally obtain a trained neural network model.

[0089] Furthermore, the neural network model obtained by training in step 2 is used to predict the remaining flight time of the missile in step 3, and the time constraint bias term is constructed using the expected strike time and the remaining flight time, and the time constraint guidance law is designed, which specifically includes:

[0090] Step 3-1, set the expected missile strike time T d ;

[0091] Step 3-2, using the neural network model trained in step 2, real-time input: missile-target distance, lead angle, speed, neural network output: the remaining flight time t of the missile go ;

[0092] Step 3-3, using T in step 3-1 d and t in step 3-2 go , giving the strike time error e = T d -t go , the construction time bias term is

[0093]

[0094] Among them, κ is the time error coefficient;

[0095] Step 3-4, based on the time error bias term Ω designed in step 3-3 b , the time-constrained guidance law is designed as

[0096] A M =(NΩ r +Ω b )×V

[0097] Step 3-5, prove the Lyapunov stability of the time-constrained guidance law designed in step 3-4, as follows:

[0098] Step 3-5-1, first give the calculation formula for the remaining flight time under the speed-varying proportional guidance law,

[0099]

[0100] in, is the future average speed of the missile;

[0101] Step 3-5-2, take the derivative of the above formula, we have

[0102]

[0103] Step 3-5-3, under the above time constraint guidance law, we can derive The above formula can be simplified to

[0104]

[0105] Step 3-5-4, construct the Lyapunov function for the strike time error:

[0106] V=0.5e 2

[0107] Step 3-5-4, take the derivative of the above formula, we have

[0108]

[0109] From the above formula, we can get: By satisfying Lyapunov stability, the strike time error can gradually converge to zero, ensuring the terminal strike time constraint of the missile.

[0110] Further, according to the neural network model trained in step 2 as described in step 4 and the time-constrained guidance law proposed in step 3, it is ensured that a single missile intercepts the target within the expected time, specifically including:

[0111] Step 4-1, obtain the flight state (R, σ, V) of the missile, input the flight state of the missile into the physical information neural network model trained in step 2, and obtain the remaining flight time of the missile;

[0112] Step 4-2, substituting the remaining flight time of the missile obtained in step 4-1 into the time-constrained guidance law designed in step 3, solving the real-time overload of a single missile, and guiding the missile to fly to the target;

[0113] Step 4-3: Return to step 4-1 until the missile hits the target.

[0114] A guidance system with time constraints in the atmosphere based on a physical information neural network, specifically comprising: a missile kinematics iteration module, a remaining flight time calculation module based on a physical information neural network, and a missile guidance control module, wherein:

[0115] Missile kinematics iteration module, used to calculate the missile's real-time position and provide real-time guidance kinematics data;

[0116] The remaining flight time calculation module based on physical information neural network is used to calculate the remaining flight time of the missile in real time and provide data input for the missile to apply the time constraint guidance law;

[0117] The missile guidance and control module calculates the missile's overload instructions in real time based on the designed time-constrained guidance law, the expected strike time, and the remaining flight time predicted by the physical information neural network.

[0118] Then, taking the time-constrained guidance of a single missile as an example, the expected strike time is 60s, and the initial parameters of the missile are as follows: x position is -12km, y position is -7km, z position is 5.1km, ballistic inclination is -10°, ballistic deviation is 35°, and the missile and air parameters are selected as follows: T = 0, S = 0.4m 2 , m = 200 kg, C D0 =0.04, K D =0.1. This embodiment only selects a set of specific aerodynamic parameters and a missile terminal parameter for simulation. For other different aerodynamic parameters and different types of missile parameters, it is only necessary to replace the relevant parameters accordingly. Figure 5-8 This is the simulation result of a single missile guidance.

[0119] like Figure 6 The figure shows the change of the remaining flight time of a single missile. Figure 4 Given the physical information neural network model training loss function change curve, from Figure 6 It can be seen that a single aircraft can hit the target within 60 seconds, achieving time-constrained guidance. The time-constrained guidance law provided by the present invention has an excellent guidance effect, wherein the value of N in step 1-2 is 3.

[0120] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0121] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A guidance method with time constraints in the atmosphere based on a physical information neural network, characterized in that: The steps include: Step 1: Based on the proportional guidance law, a data set of missile flight state parameters and remaining flight time parameters at different initial states is constructed; Step 2: Use the data set constructed in step 1 to train the neural network model; Step 3: Based on the neural network model trained in step 2, the real-time remaining flight time of the missile is predicted, and the time constraint bias term is constructed using the expected strike time and the remaining flight time of the missile to design a time constraint guidance law; Step 4: Based on the neural network model trained in step 2 and the time-constrained guidance law proposed in step 3, the missile is guided to hit the target at the desired time; Among them, the step 3 predicts the real-time remaining flight time of the missile based on the neural network model trained in step 2, constructs the time constraint bias term using the expected strike time and the real-time remaining flight time of the missile, and designs the time constraint guidance law, which specifically includes: Step 3-1, set the expected missile strike time T d ; Step 3-2, using the neural network model trained in step 2, real-time input: missile-target distance R, lead angle σ, speed V, get the output of the neural network model: the remaining flight time t of the missile go ; Step 3-3, using T in step 3-1 d and t in step 3-2 go , construct the strike time error e = T d -t go , the construction time bias term is, Where κ is the time error coefficient, Step 3-4, based on the time error bias term Ω designed in step 3-3 b , the time-constrained guidance law is designed as A M =(NΩ r +Oh b )×V。 2. The method of guiding in the atmosphere with time constraints based on a physical information neural network according to claim 1, characterized in that: The neural network model of step 2 is a neural network model that takes physical information constraints into consideration.

3. The method of guiding in the atmosphere with time constraints based on physical information neural network according to claim 2, characterized in that: The missile flight state parameters of different initial states in step 1 specifically include missile-target distance, lead angle, missile flight speed, and the remaining flight time parameter specifically refers to the actual remaining flight time of the missile.

4. The method of guiding in the atmosphere with time constraints based on physical information neural network according to claim 3 is characterized in that: The engagement model between the missile and the target during the guidance process is specifically described as follows: the line of sight vector between the missile and the target is R, the missile velocity vector is described as V, and The distance between the missile and the target is described as R = ||R||, the missile's velocity is described as V = ||V||, and the rotation angular rates of the sight vector and the missile's velocity vector are respectively Where a is the missile acceleration vector perpendicular to the velocity vector, The total missile lead angle is defined as: The vector perpendicular to the plane formed by the line of sight vector R and the missile velocity vector V is defined as: k = R×V / R×V.

5. The method of guiding in the atmosphere with time constraints based on physical information neural network according to claim 4, characterized in that: The speed of the missile during the guidance process is affected by air resistance and gravity acceleration, specifically: Where T is the missile thrust, q is the air dynamic pressure, S is the missile area, m is the missile mass, g is the gravity acceleration vector, g n is the component of gravity in the direction of the missile's velocity, C D0 is the zero lift drag coefficient, K D is the induced drag coefficient.

6. The method of guiding in the atmosphere with time constraints based on physical information neural network according to claim 5, characterized in that: The step 2 uses the data set constructed in step 1 and the randomly generated prior model data set to perform neural network model training, specifically including: Step 2-1, design a physical information neural network to predict the remaining flight time of the missile. The specific expression is as follows: Among them, t go is the output of the physical information neural network, is the prior model, Δ(R,σ,V) is the output of the deep neural network, which is finally Together with Δ(R,σ,V), it forms the expression for the remaining flight time of the missile; Step 2-2, the deep neural network model described in step 2-1 has two hidden layers, and the number of neurons in the hidden layer is 100; Step 2-3, the loss function is designed as, in, is the remaining flight time predicted by the physical information neural network, y i is the actual remaining flight time, N is the number of training data sets, is the response of the physical information neural network to the prior model data, A priori model Calculate the remaining flight time, N is the number of prior model data sets; Step 2-4, the model training parameters of the physical information neural network are: the number of iterations is 20000, the learning rate is 0.0001, and the activation function uses the ReLU function; Step 2-5, using the 95387713 training sample data and 234431941 prior model sample data obtained in step 1, special physical information neural network training is performed to finally obtain a trained neural network model.

7. A guidance system using the guidance method according to any one of claims 1 to 6, characterized in that: The system comprises a missile kinematics iteration module, a missile remaining flight time calculation module based on a physical information neural network and a missile guidance control module, wherein the missile kinematics iteration module is used to calculate the real-time position data of the missile, and then provide real-time guidance kinematics data; the missile remaining flight time calculation module based on the physical information neural network is used to calculate the real-time missile remaining flight time, and then provide parameter input for the design of the time-constrained guidance law; the missile guidance control module solves the missile's overload instruction based on the designed time-constrained guidance law, the pre-set strike time and the real-time remaining flight time, and guides the missile to strike the target with the expected flight time.

Images