Intelligent collaborative guidance method and system based on BP neural network with time and space constraints

Abstract

The space-time constraint intelligent collaborative guidance method and system based on BP neural network belong to the field of collaborative guidance. It solves the existing intelligent collaborative attack target scenario based on estimating the remaining flight time, which is limited to stationary targets. For the scenario of collaborative attack on highly maneuverable targets, there is a low accuracy of estimating the remaining flight time, which leads to low accuracy of highly maneuverable targets. The present invention takes the ballistic data of a single missile under proportional guidance as a training set, trains the neural network, establishes a mapping from the missile state, the target maneuverability characteristics to the remaining flight time, and inverts the trained neural network in the collaborative guidance law design to form a learning method for the initial scenario and can intelligently predict the remaining flight time online, construct the space-time collaborative guidance law of each missile according to the predicted remaining flight time of each missile flying to the maneuverable target, and realize the collaborative attack task of multiple missiles on the maneuverable target. It is mainly used in the game process between missile groups and maneuverable targets.

Description

Technical Field

[0001] The invention belongs to the field of cooperative guidance. Background Art

[0002] In recent years, in order to improve the strike effectiveness, cooperative guidance technology has received more and more attention. Cooperatively guided multiple missiles can be regarded as a multi-agent system that can cooperate with each other to achieve mission objectives based on simple control laws. The effectiveness of multiple missiles in coordinated operations is better than that of a single high-tech and high-cost aircraft. In addition, a multi-agent system composed of multiple missiles will also have stronger combat capabilities than a single aircraft. Among them, it is a very challenging problem for multiple missiles in different initial conditions to achieve simultaneous strikes on targets.

[0003] At present, the coordinated guidance laws for simultaneous multi-missile attacks can be divided into two categories. The first category is to specify the expected attack time for the coordinated guidance system. Min-Jea Tahk et al. proposed a non-singular sliding mode guidance law to enable missiles to hit targets at a specified time; Xiaodong Liu et al. designed a guidance law for specifying attack time and attack angle based on adaptive sliding mode. However, it is obviously difficult to specify the same attack time for multiple missiles under different initial conditions. In addition, there is no interaction between the missiles, which reduces the sensitivity of the multi-missile coordinated guidance system to the environment. Another method is for the missiles in the system to coordinate with each other and align the attack time. Min-Jea Tahk et al. expanded the proportional guidance law and designed a multi-missile coordinated attack guidance law. However, this guidance law requires each missile to have global information of the entire system, which is centralized control and has a heavy communication burden. Subsequently, Jialing Zhou et al. designed a distributed coordinated attack guidance law with finite time convergence based on Lyapunov theory. Abhinav Sinha et al. designed a super-helical sliding mode guidance law to make the guidance instructions smoother. Based on the consistency theory, Li Wen et al. selected the relative distance and approach speed between the aircraft and the target as coordination variables, and designed a distributed time-coordinated three-dimensional guidance method for unpowered aircraft under time-varying speed. Based on the proportional guidance law, Jeon et al. introduced the error term of the remaining time between missiles into the navigation ratio, and realized the coordinated guidance of multiple missiles by reducing the variance of the remaining flight time. Zhao Shiyu et al. applied coordination variables and coordination functions to the study of coordinated guidance, selected the remaining time of the missile as the coordination variable, and realized the salvo attack of multiple missiles. Zhang Youan et al. introduced the "Leader-Follower" formation control method in the study of coordinated guidance, and defined the concepts of leading missiles and led missiles in the missile group. Zhao Qilun et al. used the multi-agent second-order consistency algorithm to control the led missile under the "leader-led missile" model, thereby realizing the tracking of the led missile by the led missile and completing the coordinated guidance task of multiple missiles. Guidance laws based on neural networks or deep learning networks require optimized samples for training. The number of samples determines to a certain extent the combat environment in which the intelligent guidance law can be applied, that is, some combat environments that are not in the samples may make the guidance law invalid. In order to simplify the guidance law design and improve the robustness of the guidance law, Brian Gaudet et al. designed the terminal guidance law of the interceptor based on meta-reinforcement learning. Nan Ying et al. improved the memory pool generation method of the traditional deep Q network (DQN) and proposed a mid-course penetration control model for ballistic missiles based on the Markov decision process. Reinforcement learning technology does not require the samples required by deep learning, nor does it need to design the structure of the controller. It can autonomously explore actions that achieve the maximum reward value.

[0004] The above intelligent guidance law is only for a single missile, not a "group of missiles". Hu Shiyou et al. summarized the current status of the development of intelligent precision guidance technology for missile weapons and analyzed the development trend, pointing out that conducting research on information interaction and collaborative processing technology between multiple missiles and realizing distributed intelligent collaboration of groups are important directions for the development of intelligent precision guidance technology for missile weapons in the future. Shao et al. used radial basis function (RBF) neural network to estimate target motion information in real time and reduce coupling interference in three-dimensional motion models. As a result, the guidance law can adapt to changes in parameters and improve the robustness of the guidance law. Li et al. proposed a guidance law for intercepting non-maneuvering targets with fall angle and lateral acceleration command constraints. The guidance law combines the sliding mode control method with the wavelet neural network to improve the robustness of the system and reduce the system's chattering. Based on the Lyapunov stability theory, Lai et al. proposed a neural adaptive fast terminal sliding mode dynamic surface control method based on nonlinear extended state observer to ensure the stability of the system. In order to improve the terminal guidance accuracy, Li Jie et al. derived a sliding mode guidance law based on zeroing the line of sight angular rate according to the sliding mode variable structure theory, used the RBF neural network to estimate the switching term gain online, and simulated and compared the designed RBF sliding mode guidance law with the sliding mode guidance law and the proportional guidance law. In order to solve the problems of line of sight angular rate chattering and ignoring the dynamic characteristics of the autopilot when intercepting high-speed and highly maneuvering targets, Tong Tingshuai et al. proposed a sliding mode guidance law based on RBF neural network gain adjustment by combining variable structure control theory and neural network. Rajagopalan et al. analyzed in detail the feasibility and advantages of using neural network to design guidance law. Xu Yang used the long-short term memory (LSTM) network to design an intelligent guidance law with anti-interference ability for infrared air-to-air missiles under the conditions of target maneuvering and infrared decoy interference. Filici et al. designed an online neural network guidance law considering the randomness of disturbances. Once disturbances occur, the neural network will perform online compensation. The guidance law problem under multi-constraint and uncertain conditions is studied. A possible idea is to first relax the constraints and establish a rough uncertain model, and use the existing control theory to design a guidance law with constrained variables and uncertain parameters; then build a simulation system, perform a large number of guidance simulations under the desired constraints and uncertain environments, and obtain labeled samples; then use the powerful learning and reasoning capabilities of artificial intelligence such as neural networks or reinforcement learning to adaptively estimate the constrained variables and uncertain parameters, thereby obtaining an intelligent guidance law under multiple constraints and uncertain conditions. It is also possible to directly build a suitable deep learning neural network based on the main characteristic parameters that affect guidance under constraints and uncertain environments, and obtain an intelligent guidance law through offline and online linear learning methods.Some other literatures generate a series of dynamic virtual guidance points according to the requirements and actual positions of the aircraft formation, and use the real-time state as input to train the neural network guidance model online based on model predictive control to complete the aircraft formation control. Some other literatures propose to use a separate control strategy for the guidance loop and attitude loop of the aircraft, and use the L1 guidance algorithm to design a flight controller based on an adaptive neural network to effectively improve the tracking accuracy and anti-interference ability of the aircraft.

[0005] Traditional methods for estimating the remaining flight time require some strict restrictions, such as a small lead angle, pure proportional guidance, and the target to be struck is a stationary target. Therefore, for scenarios of time-space coordinated strikes on highly maneuverable targets, the accuracy of such estimation methods decreases. To address this problem, the present invention proposes a new technical solution to improve the accuracy of the estimation of the remaining flight time. Summary of the invention

[0006] The purpose of the present invention is to solve the problem that the existing intelligent collaborative strike target scenario based on estimating the remaining flight time is limited to stationary targets. For the scenario of collaborative strike on highly maneuverable targets, there is a low accuracy in estimating the remaining flight time, which leads to low accuracy of highly maneuverable targets. The present invention provides a spatiotemporal constraint intelligent collaborative guidance method and system based on BP neural network.

[0007] The space-time constraint intelligent collaborative guidance method based on BP neural network includes the following steps:

[0008] Step 1: Build a training set:

[0009] In different simulation scenarios, the single-missile guidance model simulates the maneuvering target using the proportional guidance of a given single missile, and obtains the trajectory data at each sampling time in each simulation scenario as samples to form a training set;

[0010] Each sample includes: the relative distance R of the projectile and the target as the network input data, the sight angle q of the projectile and the target, and the derivative of the relative distance Line of sight angular velocity Missile Speed ​​V m , missile velocity lead angle η m , target speed V t and target acceleration a t , and the remaining flight time t of the missile to the maneuvering target as the network output data go ;

[0011] Step 2: Network training:

[0012] Using the training set to train the BP neural network, and obtaining the trained BP neural network;

[0013] Step 3: Use the trained BP neural network to predict the remaining flight time of each missile in the swarm to the maneuvering target:

[0014] The relative distance R, the sight angle q, and the approach speed of each missile in the missile group are calculated. Line of sight angular velocity Missile Speed ​​V m , missile velocity lead angle η m , target speed V t and target acceleration a t As the input of the trained BP neural network, the remaining flight time of each missile in the missile group to the maneuvering target is predicted;

[0015] Step 4: Based on the predicted remaining flight time of each missile to the maneuvering target, construct the spatiotemporal coordinated guidance law of each missile, so as to realize the coordinated strike mission of multiple missiles against the maneuvering target.

[0016] Preferably, the parameters in the simulation scene include an initial relative distance between the missile and the target, an initial sight angle between the missile and the target, an initial ballistic inclination angle of the missile, and a target acceleration.

[0017] Preferably, the initial relative distance between the missile and the target varies from 8,000 meters to 12,000 meters, the initial sight angle between the missile and the target varies from -180° to 180°, the initial ballistic inclination angle of the missile varies from -80° to 80°, and the target acceleration varies from -4g to 4g.

[0018] Preferably, a new simulation scenario is formed by changing one or more parameter values ​​in the simulation scenario.

[0019] Preferably, the hidden layer of the BP neural network adopts the Tanh activation function, and the output layer adopts the linear output function.

[0020] Preferably, in step 4, the implementation method of constructing the spatiotemporal coordinated guidance law of each missile according to the predicted remaining flight time of each missile flying to the maneuvering target includes:

[0021] Step 41: Calculate the remaining flight time error ξ of each missile flying to the maneuverable target based on the predicted remaining flight time of all missiles flying to the maneuverable target i ; Among them, ξ i is the remaining flight time error of the i-th missile flying to the maneuvering target;

[0022] Step 42: According to ξ i , get the time-varying navigation gain of the ith missile

[0023] Step 43: According to Construct the spatiotemporal coordinated guidance law a of the i-th missilemi .

[0024] Preferably,

[0025] Among them, a ij is the communication matrix between the i-th missile and the j-th missile, and a ij The value is 0 or 1. ij = 0, indicating that the ith missile and the jth missile cannot communicate. ij =1 indicates that the ith missile can communicate with the jth missile, t goj is the predicted remaining flight time of the jth missile to the maneuvering target, j = 1, 2, 3...n, t goi is the predicted remaining flight time of the ith missile to the maneuvering target, i = 1, 2, 3...n, n is the total number of missiles in the swarm.

[0026] Preferably,

[0027] Where N2 is the navigation ratio of the cooperative guidance law, μ i ,k 1i , k 2i and μ i are the first, second and third adjustable parameters of the ith missile, sig(·) is a nonlinear function, and sign(·) is the sign function.

[0028] Preferably,

[0029] Among them, V mi is the missile speed of the ith missile, is the line-of-sight angular velocity of the ith missile, V ci is the approach speed of the ith missile, q i is the sight angle of the ith missile, q fi is the expected terminal sight angle of the ith missile, is the estimated remaining flight time of the ith missile to the maneuvering target, R i is the relative distance between the ith missile and the target, is the relative distance derivative of the ith missile to the target.

[0030] A spatiotemporal constraint intelligent collaborative guidance system based on a BP neural network comprises a storage device, a processor, and a computer program stored in the storage device and executable on the processor. The processor executes the computer program to implement a spatiotemporal constraint intelligent collaborative guidance method based on a BP neural network.

[0031] Advantages of the present invention:

[0032] The present invention uses BP neural network to predict the remaining flight time in the terminal guidance process of missile, and then realizes the spatiotemporal constraint collaborative guidance of maneuvering targets. In the design of intelligent collaborative guidance process, in order to obtain the data required for network training, a large number of simulations are carried out in the scenario of single missile attacking maneuvering targets, and sample features are selected from the ballistic data to establish a training set. On this basis, the network is trained, and a mapping relationship from missile state to remaining flight time is established, and this network is applied to the spatiotemporal collaborative guidance process.

[0033] In the process of multiple missiles attacking targets in coordination, information such as missile flight status and target flight status will affect the coordination process. In order to accurately estimate the remaining flight time of the coordinated guidance process, the selected neural network input data vector should contain as much information as possible that affects the remaining time. The missile flight status information selected by the present invention includes the missile-target relative distance R, the missile-target line of sight angle q, and the missile-target relative distance derivative. Line of sight angular velocity Missile Speed ​​V m and the missile velocity lead angle η m , the target flight state includes the target speed V t and target acceleration a t The present invention trains a BP neural network about the remaining flight time to improve the estimation accuracy of the remaining flight time. The present invention is different from the prior art in that:

[0034] First, the present invention utilizes the universal approximation characteristics of neural networks to establish a mapping from missile status, target maneuverability characteristics to remaining flight time, effectively avoiding the shortcomings of small lead angle assumptions in traditional methods.

[0035] Secondly, the present invention can realize the spatiotemporal constraint coordinated guidance of maneuvering targets, and does not require a complex dynamic iterative parameter adjustment process.

[0036] The present invention verifies the correctness and effectiveness of the intelligent collaborative guidance method proposed in the present invention when striking highly maneuverable targets through simulations in different scenarios and comparison with the traditional remaining flight time estimation method. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the principle of the spatiotemporal constraint intelligent collaborative guidance method based on BP neural network of the present invention;

[0038] Figure 2 This is a simulation diagram of the proportional guidance of a given single missile for a maneuvering target; Figure 2 a is the trajectory diagram of the projectile and target, Figure 2 b is the relative distance diagram of projectile and target;

[0039] Figure 3 This is a simulation result diagram of the remaining flight time prediction using the BP neural network trained by the present invention in scenario 1 of the verification test; wherein, Figure 3 a is the trajectory diagram of three missiles and the maneuvering target; Figure 3 b is the relative distance diagram of the three missiles; Figure 3 c is the time-varying navigation gain diagram of the three missiles, Figure 3 d is the remaining flight time curve of the three missiles at each time predicted by the trained BP neural network. Figure 3 e is the time-space coordinated guidance law curve of the three missiles, Figure 3 f is the missile-target sight angle curve of the three missiles;

[0040] Figure 4 This is the simulation result diagram of scenario 1 of the verification test using the traditional remaining flight time estimation method; among them, Figure 4 a is the trajectory diagram of three missiles and the maneuvering target. Figure 4 b is the relative distance diagram of the three missiles; Figure 4 c is the time-varying navigation gain diagram of the three missiles, Figure 4 d is the remaining flight time curve of the three missiles at each time predicted by the trained BP neural network. Figure 4 e is the time-space coordinated guidance law curve of the three missiles, Figure 4 f is the missile-target sight angle curve of the three missiles;

[0041] Figure 5 This is a simulation result diagram of the remaining flight time prediction using the BP neural network trained by the present invention in scenario 2 of the verification test; wherein, Figure 5 a is the trajectory diagram of three missiles and the maneuvering target; Figure 5 b is the relative distance diagram of the three missiles; Figure 5 c is the time-varying navigation gain diagram of the three missiles, Figure 5 d is the remaining flight time curve of the three missiles at each time predicted by the trained BP neural network. Figure 5 e is the time-space coordinated guidance law curve of the three missiles, Figure 5 f is the missile-target sight angle curve of the three missiles;

[0042] Figure 6 This is the simulation result diagram of scenario 2 of the verification test using the traditional remaining flight time estimation method; among them, Figure 6 a is the trajectory diagram of three missiles and the maneuvering target. Figure 6 b is the relative distance diagram of the three missiles; Figure 6 c is the time-varying navigation gain diagram of the three missiles, Figure 6d is the remaining flight time curve of the three missiles at each time predicted by the trained BP neural network. Figure 6 e is the time-space coordinated guidance law curve of the three missiles, Figure 6 f is the missile-target line-of-sight angle curve of the three missiles.

[0043] Attached Figures 3 to 6 In the figure, X and Y are the horizontal and vertical coordinates of the missile position in the plane coordinate system under the top-down perspective, M1 to M3 are missile 1, missile 2 and missile 3, t is time, and T is the maneuvering target. DETAILED DESCRIPTION

[0044] 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.

[0045] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0046] Research on multi-constrained space-time collaborative guidance process:

[0047] As an important machine learning method and intelligent information processing model, neural network has been widely used in various fields such as intelligent control, guidance law design, and aircraft formation in recent years. Multilayer feedforward neural network has also been proved to be able to express any function when there are enough neurons. At the same time, neural network can also be regarded as a deposition unit of prior knowledge. The solution of remaining flight time can be directly obtained by training the neural network to respond to the stimulus signal of missile state. BP algorithm is still the most widely used neural network training method due to its solid theoretical basis, rigorous derivation, and clear physical concept. The neural network model is used to establish the mapping relationship between the coordinated flight state of multiple missiles and the remaining flight time. The network input is the flight state of missiles and targets at different times, and the network output is the remaining flight time of the missile flying to the maneuvering target. In the process of multiple missiles coordinating to attack targets, information such as missile flight state and target flight state will affect the cooperative process. In order to accurately estimate the remaining flight time of the cooperative guidance process, the selected neural network input vector should contain as much information as possible that affects the remaining time. At the same time, in order to make the established neural network have stronger generalization ability, it is necessary to collect collaborative guidance process data and remaining flight time data under different initial or flight conditions. These data sets are used to train the neural network. The neural network input is the collaborative guidance process data under different initial or flight conditions, and the neural network output is the actual remaining flight time t of a single missile under proportional guidance. go .

[0048] The trajectory data of a single missile under proportional guidance is used as a training set to train the network. In the collaborative guidance process, the trained network is used to estimate the remaining flight time online, thereby improving the accuracy of the remaining flight time. Figure 1 The spatiotemporal constraint intelligent collaborative guidance method based on BP neural network described in this embodiment comprises the following steps:

[0049] Step 1: Build a training set:

[0050] In different simulation scenarios, the single-missile guidance model simulates the maneuvering target using the proportional guidance of a given single missile, and obtains the trajectory data at each sampling time in each simulation scenario as samples to form a training set;

[0051] The data in each sample in the training set is obtained by simulation under a two-dimensional single-missile guidance model. Taking into account the complexity of multi-missile attacks, for a given guidance law and the same flight speed, the guidance principle of a single missile relative to the target in the coordinated process is similar. In order to reduce the complexity of the neural network model and improve the training efficiency of the neural network model, a single-missile proportional guidance is used for simulation of maneuvering targets, that is, the guidance law is shown as follows.

[0052]

[0053] Among them, N1 is the proportional guidance ratio, V m is the missile speed, is the line of sight angular velocity;

[0054] At the same time, in order to improve the generalization ability of the neural network to different flight states, it is necessary to collect experimental data under different initial flight states, so the range of simulation initial conditions is given as shown in Table 1.

[0055] Table 1 Initial parameter settings for simulation scenarios

[0056]

[0057]

[0058] By changing the values ​​of one or more parameters in the simulation scene, a new simulation scene is formed to achieve scene changes and improve the generalization ability of the neural network for different flight states. The initial relative distance between the missile and the target changes from 8000 to 12000 meters at intervals of 1000 meters, the initial sight angle of the missile changes from -180° to 180° at intervals of 10°, and the initial ballistic inclination of the missile changes from -80° to 80° at intervals of 10°. g is the acceleration of gravity, g = 9.8m / s 2 After a large number of simulations, the training data set obtained is as follows Figure 2 shown.

[0059] Given the missile and target flight speeds, the sample data taken needs to reflect the relationship between the current missile state and the true value of the remaining flight time, so the main data features considered, that is, the missile state, are shown in Table 2. Each sample in the training set includes: the missile-target relative distance R as network input data, the missile-target line of sight angle q, and the missile-target relative distance derivative Line of sight angular velocity Missile Speed ​​V m , missile velocity lead angle η m , target speed V t and target acceleration a t , and the remaining flight time t of the missile to the maneuvering target as the network output data go .

[0060] Table 2 BP neural network input and output vectors

[0061]

[0062] Step 2: Network training:

[0063] The BP neural network is trained using the training set to obtain the trained BP neural network. In specific applications, the network input data in each sample is input, namely: the relative distance R between the projectile and the target, the sight angle q between the projectile and the target, and the derivative of the relative distance between the projectile and the target. Line of sight angular velocity Missile Speed ​​V m , missile velocity lead angle η m , target speed V t and target acceleration a t As the input of the BP neural network; the network output data in the sample, that is, the remaining flight time t of the missile to the maneuvering target go It is trained as the input and output of the BP neural network.

[0064] Step 3: Use the trained BP neural network to predict the remaining flight time of each missile in the swarm to the maneuvering target:

[0065] The relative distance R, the sight angle q, and the approach speed of each missile in the missile group are calculated. Line of sight angular velocity Missile Speed ​​V m , missile velocity lead angle η m , target speed V t and target acceleration a t As the input of the trained BP neural network, the remaining flight time of each missile in the missile group to the maneuvering target is predicted;

[0066] Step 4: Based on the predicted remaining flight time of each missile to the maneuvering target, construct the spatiotemporal coordinated guidance law of each missile, so as to realize the coordinated strike mission of multiple missiles against the maneuvering target.

[0067] In order to reduce the difference between the remaining flight time under single-missile proportional guidance and multi-missile coordinated guidance, and to improve the estimation accuracy of the remaining flight time as much as possible, it is necessary to keep the navigation ratio in the coordinated guidance law consistent with the proportional guidance navigation coefficient used when obtaining training data. That is, N1 = N2, where N1 is the proportional guidance navigation ratio shown in formula (1), and N2 is the navigation ratio of the coordinated guidance law. The expression of the space-time coordinated guidance law is shown in formula (4).

[0068] Furthermore, the implementation methods of constructing the spatiotemporal coordinated guidance law of each missile include:

[0069] Step 41: Calculate the remaining flight time error ξ of each missile flying to the maneuvering target based on the predicted remaining flight time of all missiles flying to the maneuvering target i ; Specifically,

[0070]

[0071] Among them, ξi is the remaining flight time error of the ith missile flying to the maneuvering target, a ij is the communication matrix between the i-th missile and the j-th missile, and a ij The value is 0 or 1. ij = 0, indicating that the ith missile and the jth missile cannot communicate. ij =1 indicates that the ith missile can communicate with the jth missile, t goj is the predicted remaining flight time of the jth missile to the maneuvering target, j = 1, 2, 3...n, t goi is the predicted remaining flight time of the ith missile to the maneuvering target, i = 1, 2, 3, ... n, n is the total number of missiles in the swarm;

[0072] Step 42: According to ξ i , get the time-varying navigation gain of the ith missile Specifically,

[0073]

[0074] Where N2 is the navigation ratio of the cooperative guidance law, μ i ,k 1i , k 2i and μ i are the first, second and third adjustable parameters of the ith missile, sig(·) is a nonlinear function, and sign(·) is the sign function;

[0075] Step 43: According to Construct the spatiotemporal coordinated guidance law a of the i-th missile mi ;

[0076]

[0077] V mi is the missile speed of the ith missile, is the line-of-sight angular velocity of the ith missile, V ci is the approach speed of the ith missile, q i is the sight angle of the ith missile, q fi is the expected terminal sight angle of the ith missile, is the estimated remaining flight time of the ith missile to the maneuvering target, R i is the relative distance between the ith missile and the target, is the relative distance derivative of the ith missile to the target.

[0078] Specific implementation method 2: The spatiotemporal constraint intelligent collaborative guidance system based on BP neural network described in this implementation method includes a storage device, a processor, and a computer program stored in the storage device and executable on the processor. The processor executes the computer program to implement a spatiotemporal constraint intelligent collaborative guidance method based on BP neural network.

[0079] Principle analysis:

[0080] The invention aims at the game process between air-to-air missiles and maneuverable targets, analyzes the research status of intelligent methods in cooperative guidance control, then takes the limitations of traditional cooperative guidance methods as the entry point, utilizes the universal approximation characteristics of neural networks, takes the ballistic data of a single missile under proportional guidance as a training set, trains the neural network, establishes a mapping from missile state, target maneuverability to remaining flight time, and inverts the trained neural network in the design of cooperative guidance laws, forms a learning method for initial scenes and can intelligently predict the remaining flight time online, constructs the time-space cooperative guidance laws of each missile according to the predicted remaining flight time of each missile flying to the maneuverable target, and realizes the cooperative strike task of multiple missiles on the maneuverable target.

[0081] Verification test:

[0082] The spatiotemporal constraint intelligent collaborative guidance method based on BP neural network was simulated and verified. Two different simulation scenarios were given respectively, and compared with the traditional remaining flight time estimation method to verify the adaptability and effectiveness of the collaborative guidance method using neural network to estimate the remaining flight time for different scenarios.

[0083] 1. Simulation 1: Verification test within simulation time t

[0084] Table 3 Initial parameters of simulation scenario 1

[0085]

[0086]

[0087] 2. Simulation 2: Verification test within simulation time t

[0088] Table 4 Initial parameters of simulation scenario 2

[0089]

[0090] Tables 5 and 6 below respectively give the miss distances and attack times of three missiles when the remaining flight time is estimated using the BP neural network of the present invention and the traditional estimation method, as follows.

[0091] Table 5 Miss-target amount and attack time using BP in two simulation scenarios

[0092]

[0093] Table 6 Miss distance and attack time using traditional estimation method in two simulation scenarios

[0094]

[0095] For the two simulation scenarios, the missile-target trajectory, missile-target relative distance, time-varying navigation gain, remaining flight time, time-space collaborative guidance law, and line-of-sight angle change curves of the three missiles under the intelligent collaborative guidance proposed by the present invention are shown in Figure 2. Figure 3 and Figure 5 Table 5 shows the miss distance and attack time of three missiles under the remaining flight time using the method of the present invention in two simulation scenarios. Figure 4 and Figure 6 The simulation results of the traditional remaining flight time estimation method in two scenarios are given, and Table 6 gives the corresponding miss distance and attack time.

[0096] from Figure 3 a, 3b and Figure 5 It can be seen from 5a and 5b that for two different simulation scenarios, the remaining flight time t of the missile flying to the maneuvering target is predicted based on the BP neural network. go The intelligent space-time collaborative guidance law can ensure that the missiles hit the target at the same time. Figure 3 c and Figure 5 c shows the change curve of the time-varying navigation gain in the guidance law. It can be seen that in the initial stage t go When the consistency error is large, the guidance law generates a large navigation gain to eliminate the consistency error. go When the agreement is reached, the guidance law uses a smaller navigation gain to hit the target. Figure 3 d and Figure 5 d shows that the t of the three missiles at the initial moment of the simulation go The differences between them are large, but they can converge to the same in a very short time under the action of the cooperative guidance law, and the curve is smooth and continuous without chattering, which shows that the BP neural network can be used to predict t go accuracy and rationality. Figure 3 e and Figure 5 e gives the time-space coordinated guidance law curves of the three missiles, that is, the line-of-sight normal acceleration command curve. The larger navigation gain in the initial stage leads to a larger acceleration in order to eliminate the consistency error. As the remaining flight time tends to be consistent, the line-of-sight normal acceleration amplitude also quickly converges to a reasonable range. Finally Figure 3 f and Figure 5 f shows the change of the sight angle. It can be seen that the sight angles of the three missiles converge to their expected values ​​at the end of the attack.

[0097] Through the comparative simulation of two different remaining flight time estimation methods, it can be seen that the collaborative guidance effect is better when the neural network is used to estimate the remaining flight time, and the consistency error of the collaborative guidance using the traditional estimation method is larger, which ultimately leads to a large hit time error. This is because some prerequisites of the traditional method of estimating the remaining flight time cannot be met in the maneuvering target scenario, resulting in excessive estimation errors. Finally, Tables 5 and 6 give the miss distances and attack times of the three missiles, indicating that under the designed neural network combined with the spatiotemporal collaborative guidance law, the three missiles can hit the target with a small miss distance and the guidance time is consistent. It can be seen that under the intelligent method, the spatiotemporal collaborative guidance attack of multiple air-to-air missiles on air maneuvering targets has achieved a more ideal effect.

[0098] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that features described in conjunction with individual embodiments may be used in other described embodiments.

Claims

1. A spatiotemporal constraint intelligent collaborative guidance method based on BP neural network, characterized in that: The guidance method comprises the following steps: Step 1: Build a training set: In different simulation scenarios, the single-missile guidance model simulates the maneuvering target using the proportional guidance of a given single missile, and obtains the trajectory data at each sampling time in each simulation scenario as samples to form a training set; Each sample includes: the relative distance R of the projectile and the target as the network input data, the sight angle q of the projectile and the target, and the derivative of the relative distance , line of sight angular velocity , Missile Speed ​​V m , missile velocity lead angle η m , target speed V t and target acceleration a t , and the remaining flight time t of the missile to the maneuvering target as the network output data go ; Step 2: Network training: Using the training set to train the BP neural network, and obtaining the trained BP neural network; Step 3: Use the trained BP neural network to predict the remaining flight time of each missile in the swarm to the maneuvering target: The relative distance R, the sight angle q, and the approach speed of each missile in the missile group are calculated. , line of sight angular velocity , Missile Speed ​​V m , missile velocity lead angle η m , target speed V t and target acceleration a t As the input of the trained BP neural network, the remaining flight time of each missile in the missile group to the maneuvering target is predicted; Step 4: construct the spatiotemporal coordinated guidance law of each missile according to the predicted remaining flight time of each missile to the maneuverable target, so as to realize the coordinated strike mission of multiple missiles against the maneuverable target; Among them, the implementation methods of constructing the spatiotemporal coordinated guidance law of each missile include: Step 41: Calculate the remaining flight time error ξ of each missile flying to the maneuvering target based on the predicted remaining flight time of all missiles flying to the maneuvering target i ; Among them, ξ i is the remaining flight time error of the i-th missile flying to the maneuvering target; Step 42: According to ξ i , get the time-varying navigation gain of the ith missile ; Step 43: According to , construct the spatiotemporal coordinated guidance law a of the i-th missile mi ; ； Among them, a ij is the communication matrix between the i-th missile and the j-th missile, and a ij The value is 0 or 1. ij = 0, indicating that the ith missile and the jth missile cannot communicate. ij =1 indicates that the ith missile can communicate with the jth missile, t goj is the predicted remaining flight time of the jth missile to the maneuvering target, j = 1, 2, 3...n, t goi is the predicted remaining flight time of the ith missile to the maneuvering target, i = 1, 2, 3, ... n, n is the total number of missiles in the swarm; ； Where N2 is the navigation ratio of the cooperative guidance law, k 1i , k 2i and μ i are the first, second and third adjustable parameters of the ith missile, sig(·) is a nonlinear function, and , sign(·) is the sign function; ； Among them, V mi is the missile speed of the ith missile, is the line-of-sight angular velocity of the ith missile, V ci is the approach speed of the ith missile, q i is the sight angle of the ith missile, q fi is the expected terminal sight angle of the ith missile, is the estimated remaining flight time of the ith missile to the maneuvering target, R i is the relative distance between the ith missile and the target, is the relative distance derivative of the ith missile to the target.

2. The time-space constraint intelligent collaborative guidance method based on BP neural network according to claim 1 is characterized in that: The parameters in the simulation scenario include the initial relative distance between the missile and the target, the initial sight angle between the missile and the target, the initial ballistic inclination angle of the missile, and the target acceleration.

3. The time-space constraint intelligent collaborative guidance method based on BP neural network according to claim 2 is characterized in that: The initial relative distance between the missile and the target varies from 8,000 meters to 12,000 meters, the initial sight angle between the missile and the target varies from -180° to 180°, the initial ballistic inclination of the missile varies from -80° to 80°, and the target acceleration varies from -4g to 4g.

4. The time-space constraint intelligent collaborative guidance method based on BP neural network according to claim 2 is characterized in that: By changing one or more parameter values ​​in a simulation scenario, a new simulation scenario is formed.

5. The time-space constraint intelligent collaborative guidance method based on BP neural network according to claim 1 is characterized in that: The hidden layer of the BP neural network uses the Tanh activation function, and the output layer uses a linear output function.

6. The spatiotemporal constraint intelligent collaborative guidance system based on BP neural network is characterized by: The invention comprises a storage device, a processor, and a computer program stored in the storage device and executable on the processor, characterized in that: The processor executes the computer program to implement the method according to any one of claims 1 to 5.

Images