Air defense and anti-guide intelligent interception method and system

Abstract

The invention discloses an air defense and anti-guide intelligent interception method and system, and belongs to the technical field of air defense and anti-guide, and the method comprises the steps: obtaining asynchronous collection data, closest to the current moment t, of a multi-source sensor for each interception target, and splicing the asynchronous collection data into asynchronous time sequence data; identifying the type feature of the intercepted target; an asynchronous attention mechanism and a Transform model are adopted, and the position and speed of the intercepted target at the moment t are obtained; obtaining a threat value of the intercepted target; according to the threat value of the intercepted target and the position and speed of the moment t, a physical constraint diffusion model is used for selecting one of a short-term trajectory, a medium-term trajectory or a long-term trajectory for predicting the intercepted target; and generating a plurality of interception schemes for intercepting all the interception targets, and selecting an optimal scheme to perform interception operation. The method can improve the prediction precision of the position of the intercepted target, shortens the decision time, intelligently distributes the intercepted resources, and is suitable for a complex and changeable battlefield environment.

Description

Technical Field

[0001] The present invention belongs to the technical field of air defense and antimissile, and particularly relates to an air defense and antimissile intelligent interception method and system. Background Art

[0002] With the rapid development of modern air defense and antimissile technologies, incoming targets show characteristics of high speed, mobility, intelligence, and clustering, posing higher requirements for the situation awareness, trajectory prediction, and interception decision-making capabilities of air defense and antimissile systems. Currently, mainstream air defense and antimissile systems mainly use traditional algorithms such as Kalman filtering and extended Kalman filtering for trajectory prediction. These methods perform well in dealing with linear or weakly non-linear moving targets, but when facing hypersonic and highly maneuverable targets, the prediction accuracy drops significantly, making it difficult to provide reliable data support for subsequent interception decisions.

[0003] In the field of interception decision-making, existing systems mostly adopt rule-based expert systems or simple artificial intelligence methods. These methods usually rely on preset decision rules and fixed cooperation modes. Although they can play a role in specific scenarios, in the face of complex and changing modern battlefield environments, especially when confronting intelligent weapon systems, their adaptability and decision-making effectiveness are significantly insufficient. In recent years, deep learning technologies have received extensive attention and application in the field of air defense and antimissile, but existing deep learning methods still have obvious deficiencies: pure data-driven models lack necessary physical constraints and may produce prediction results that violate physical laws; the synchronous computing mechanism of traditional model architectures such as Transformer limits the real-time performance of the system; multi-agent collaborative decision-making lacks effective causal reasoning capabilities, resulting in poor interpretability of the decision-making process.

[0004] In addition, at the system implementation level, existing intelligent air defense and antimissile systems generally suffer from problems such as low computing efficiency, poor real-time performance, and limited scalability. Especially when facing large-scale saturation attacks, the response speed and decision-making quality of the system are difficult to meet the combat requirements. At the same time, complex deployment environments and high computing costs also restrict the widespread application of intelligent technologies in the field of air defense and antimissile.

[0005] Therefore, a new type of air defense and antimissile intelligent interception method and system is proposed, which can effectively solve the above technical problems and improve the situation awareness, prediction decision-making, and collaborative interception capabilities of the system in complex battlefield environments. The present invention precisely aims at the deficiencies of the existing technology and proposes an air defense and antimissile intelligent interception system and method based on a physically constrained diffusion model and asynchronous causal Transformer, providing a new technical route for improving the intelligence level of air defense and antimissile systems. Summary of the Invention

[0006] The present invention aims at the deficiencies in the prior art and provides an air defense and antimissile intelligent interception method and system, which can improve the accuracy of predicting the position of an interception target, shorten the decision-making time, and intelligently allocate interception resources, and is applicable to complex and changeable battlefield environments.

[0007] The present invention provides the following technical solutions:

[0008] In a first aspect, an air defense and antimissile intelligent interception method is provided, including:

[0009] For each interception target, obtain asynchronous acquisition data of multi-source sensors closest to the current moment t, and splice them into asynchronous time series data. The multi-source sensors include several radar sensors, optoelectronic sensors, and infrared detectors;

[0010] Based on the acquired pictures of the optoelectronic sensors, identify the type characteristics of the interception target; based on the asynchronous time series data, use an asynchronous attention mechanism and a Transformer model to obtain the position and speed of the interception target at moment t;

[0011] According to the type characteristics, position, and speed of the interception target at moment t, obtain the threat value of the interception target;

[0012] According to the threat value, position, and speed of the interception target at moment t, use a physical constraint diffusion model to alternatively predict the short-term trajectory, medium-term trajectory, or long-term trajectory of the interception target;

[0013] Based on the predicted trajectories, threat values of each interception target, and the performance of all interception missiles, generate multiple interception plans for intercepting all interception targets, and preferentially select the optimal plan in the interception plans for interception operations.

[0014] Optionally, the step of using an asynchronous attention mechanism and a Transformer model to obtain the position and speed of the interception target at moment t based on the asynchronous time series data is specifically as follows:

[0015] For the asynchronous time series data from multi-source sensors Use an asynchronous attention mechanism to dynamically adjust the weight A of the multi-source sensor data ij ;

[0016]

[0017] w(|t i -t j |) = exp(-γ|t i -t j |)

[0018] w(|t i -t k |) = exp(-γ|·i -t k |)

[0019] wherein, t i 、t j and t k are respectively the sampling times of the i, j, and k sensors closest to the current time t, and are respectively the observed values of the i, j, and k sensors at the corresponding sampling time t i 、t j and t k , γ is the time decay parameter, and are respectively the priority weights of the data points and , M ij and M ik are the causal masks of the data points and . When t j >t i or t k >t i , both M ij and M ik are 0, otherwise they are 1; is the query vector of the data point , and are respectively the key vectors of the data points and ;

[0020] Input the weight A ij and the asynchronous time series data into the Transformer model to generate the position x t and velocity v t of the intercept target at time t;

[0021]

[0022] wherein, v tj is the observed velocity of the j sensor at the corresponding sampling time t j .

[0023] Optionally, obtaining the threat value of the intercept target according to the type characteristics, position, and velocity of the intercept target at time t specifically includes:

[0024] Concatenate the position and velocity of the intercept target at time t and the type characteristics of the intercept target to form a joint feature vector h joint ;

[0025] h joint =Concat(hembed , v t , x t )

[0026] Among them, v t and x t are respectively the position and velocity of the intercepted target at time t, and h embed is the type feature of the intercepted target,

[0027] The joint feature vector is mapped to the threat value T of the intercepted target u through a multi-layer perceptron u ;

[0028] T u = W2·ReLU(W1h joint + b1)+ b2

[0029] Among them, W1 and b1 are the weights of two dimensions for extracting non-linear interaction features, W2 and b2 are the weights of two dimensions for outputting the highlighted threat value T u ;

[0030] Optionally, according to the threat value of the intercepted target, the position and velocity at time t, use the physical constraint diffusion model to alternatively predict the short-term trajectory, medium-term trajectory or long-term trajectory of the intercepted target; specifically,

[0031] Compare the threat value of the intercepted target and the velocity at time t with the preset interval threshold, and alternatively select short-term prediction, medium-term prediction or long-term prediction, and obtain the corresponding prediction interval Δt based on the selected prediction trajectory length;

[0032] Take the position x t and velocity v t of the intercepted target at time t as the initial state, and use the physical constraint diffusion model to predict the trajectory of the intercepted target according to the corresponding prediction interval Δt;

[0033] The physical constraint diffusion model is:

[0034]

[0035] Among them, x t+Δt is the predicted position of the intercepted target at the next moment, β t is the diffusion coefficient; ∈ is a noise vector sampled from the standard normal distribution, and f(x t , v t , Δt) is the physical constraint term;

[0036]

[0037] Among them, Faero (x t , v t , Δt) is the aerodynamic force, C d is the air resistance coefficient, ρ is the atmospheric density, and S is the reference area of the intercepted target;

[0038] F gravity (x t , Δt) = mg

[0039] wherein, F gravity (x t , Δt) is the gravitational force, m is the weight of the intercepted target, and g is the acceleration due to gravity.

[0040] Optionally, after being trained, the physical constraint diffusion model performs position update. When the physical constraint diffusion model is trained, it simultaneously satisfies the energy conservation constraint loss function and the momentum conservation loss function;

[0041] The energy conservation constraint loss function L E is:

[0042]

[0043] The momentum conservation loss function L M is:

[0044]

[0045] wherein, N is the number of sampling points during the training process, E t is the predicted total energy at time t, E0 is the total energy at the initial time, p t is the momentum at time t, and p0 is the initial momentum.

[0046] Optionally, based on the predicted trajectories, threat values of each intercepted target, and the performance of all intercepting missiles, multiple interception schemes for intercepting all intercepted targets are generated, and the optimal scheme among the interception schemes is preferentially selected for the interception operation, specifically including:

[0047] For each intercepted target, based on the threat value of the intercepted target and the performance of all types of intercepting missiles, the resource allocation function A(I i ) is used to obtain the allocation probability of each type of intercepting missile for the current intercepted target;

[0048]

[0049] P(I i ) = w4R interceptor + w5v interceptor + w6M interceptor

[0050] Among them, T u is the threat value of the intercepted target u, and P(I i ) is the performance score of the interceptor missile I i . R interceptor , v interceptor and M interceptor are the range, speed and maneuverability of the interceptor missile I i ; w4, w5 and w6 are the weights corresponding to R interceptor , v interceptor and M interceptor respectively, and n is the total number of interceptor missile types;

[0051] All types of interceptor missiles that meet the set allocation probability threshold for each intercepted target are used as feasible solutions, and the launch time and interception position of each feasible solution are obtained;

[0052] The feasible solutions of all intercepted targets are combined to obtain an initial interception plan set for intercepting all intercepted targets, and the interception plans that do not meet the set launch requirements are excluded;

[0053] For all the interception plans after exclusion, an optimization algorithm is used to find the optimal plan for interception operations.

[0054] Optionally, the step of using all types of interceptor missiles that meet the set allocation probability threshold for each intercepted target as feasible solutions and obtaining the launch time and interception position of each feasible solution is specifically:

[0055] For each intercepted target, based on the predicted trajectory of the intercepted target, the motion process of the intercepted target is fitted to a uniformly accelerated motion, and a first uniformly accelerated model is constructed;

[0056]

[0057] The interception motion of each type of interceptor missile for the current intercepted target is simplified to a uniformly accelerated motion, and a second uniformly accelerated model is constructed;

[0058]

[0059] At the interception moment, the intercepted target and the interceptor missile need to satisfy the following formula, and the Newton-Raphson iteration method is used to solve the launch time and interception position of the interceptor missile for the current intercepted target for the following formula.

[0060]

[0061] Among them, t intercept is the interception moment of the interceptor missile, x target (t intercept ) and y target (t intercept) is the interception position coordinate of the target, t x is the flight time, x interceptor (t intercept ) and y interceptor (t intercept ) are the position coordinates of the interceptor missile, a x,target , a y,target , v x,target (0) and v y,target (0) are respectively the acceleration and initial velocity of the target to be intercepted in the x and y directions, a x , a y , v x0 and v yo are respectively the acceleration and initial velocity of the interceptor missile in the x and y directions; t launch is the launch time of the interceptor; x launch and y launch are the launch positions of the interceptor, x target (0) and y target (0) are the initial positions of the target to be intercepted.

[0062] Optionally, for all the intercepted schemes after exclusion, an optimization algorithm is used to find the optimal scheme for interception operations, specifically including: evaluating in the way of simulated annealing to find the optimal scheme that minimizes the objective function O;

[0063]

[0064] where C, E, and R are respectively the interception cost, interception error, and resource utilization rate of the current interception scheme, w c , w e and w r are respectively the weight of the interception cost, the weight of the interception error, and the weight of the resource utilization rate; C max , E max and R ideal are respectively the set maximum interception cost, interception error, and resource utilization rate.

[0065] Optionally, according to the threat value of the target to be intercepted, the position and velocity at time t, when using the physical constraint diffusion model to predict one of the short-term trajectory, medium-term trajectory, or long-term trajectory of the target to be intercepted, the prediction length of the short-term trajectory is less than 5 seconds, the prediction length of the medium-term trajectory is 5 - 15 seconds, and the prediction length of the long-term trajectory is greater than 15 seconds.

[0066] In the second aspect, an air defense and antimissile intelligent interception system is provided, including:

[0067] A data acquisition module, which is used to obtain asynchronous acquisition data of multi-source sensors closest to the current moment t for each interception target, and splice them into asynchronous time series data. The multi-source sensors include several radar sensors, optoelectronic sensors, and infrared detectors;

[0068] An ACT module, which is used to identify the type characteristics of the interception target based on the acquired pictures of the optoelectronic sensors; and based on the asynchronous time series data, adopt an asynchronous attention mechanism and a Transformer model to obtain the position and speed of the interception target at moment t;

[0069] A threat value acquisition module, which is used to obtain the threat value of the interception target according to the type characteristics, position, and speed of the interception target at moment t;

[0070] A PCD trajectory prediction module, which is used to select and predict the short-term trajectory, medium-term trajectory, or long-term trajectory of the interception target according to the threat value, position, and speed of the interception target at moment t by using a physical constraint diffusion model;

[0071] An interception plan generation module, which is used to generate multiple interception plans for intercepting all interception targets based on the predicted trajectories, threat values of each interception target, and the performance of all interception missiles, and preferentially select the optimal plan in the interception plans for interception operations.

[0072] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0073] By introducing a physical constraint diffusion model (PCD), the present invention integrates the physical motion laws of missiles (such as Newton's laws of motion, law of conservation of energy, etc.) into the diffusion model to more accurately simulate and predict the trajectories of targets, especially in complex hypersonic environments, significantly reducing prediction errors; the present invention uses an asynchronous update attention mechanism, priority-driven information flow, and time-dependent causal masks to quickly process asynchronous sensor data and generate decisions, greatly shortening the decision-making time, meeting the requirements of rapid confrontation, and ensuring that effective interception decisions can be made within an extremely short time. Through a multi-level interaction mechanism and an optimized decision-making algorithm, the present invention can intelligently allocate interception resources according to information such as the threat level, position, and speed of different targets, and realize the cooperative interception of multiple interception missiles. By combining physical constraints and an asynchronous update attention mechanism, the present invention can not only adapt to various different threat scenarios, but also automatically adjust algorithm parameters and decision-making strategies in complex environments to ensure efficient operation in different combat scenarios (including hypersonic target interception, saturation attack defense, intelligent cluster confrontation, etc.). In addition, through innovative technologies such as adaptive joint optimization, dynamic weight adjustment, asynchronous parallel computing, and distributed deployment, the present application has stronger adaptability and demonstrates excellent performance under different environments and task requirements. Description of the Drawings

[0074] Figure 1 It is the system flow chart of an air defense and antimissile intelligent interception method of the present invention;

[0075] Figure 2 It is the performance comparison chart of the air defense and antimissile intelligent interception system of the present invention;

[0076] Figure 3 It is the structural framework diagram of an air defense and antimissile intelligent interception system of the present invention. Specific implementation manners

[0077] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and cannot be used to limit the protection scope of the present invention.

[0078] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the term "comprising" and any deformation thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0079] Embodiment 1

[0080] As Figure 1 shown, an air defense and antimissile intelligent interception method is provided, including the following steps:

[0081] S1: For each interception target, obtain the asynchronous acquisition data of multi-source sensors closest to the current moment t, and splice them into asynchronous time series data. The multi-source sensors include several radar sensors, optoelectronic sensors and infrared detectors.

[0082] Radar Sensors: Radars with different frequency bands and powers are deployed at different locations. Long-range radars have a large detection range and can detect incoming missiles at relatively long distances, providing early warning information for the system. Although their accuracy is relatively low, they play a crucial role in monitoring a large airspace. Short-range radars, on the other hand, have a high resolution and can provide more accurate position, velocity, and acceleration information when the target is approaching, helping to precisely track the target's dynamics in the later stage. Radar sensors are hierarchically deployed based on different frequency band and power characteristics: Long-range early warning radars (usually using L / S frequency bands) are preferentially deployed in strategic depth areas (such as borderlines, coastlines, or high-altitude observation points). Through high-power transmission and low-frequency characteristics, they achieve wide-area air situation coverage (detection radius can reach more than 500 kilometers), and can provide early warning signals 120 - 300 seconds before the enemy missile enters the airspace. Their deployment azimuth angle is optimized to eliminate terrain shielding effects; Short-range fire control radars (mostly using X / Ku frequency bands) are densely configured around key protected targets (such as within 50 kilometers of military bases and core cities). Relying on high-precision beamforming and pulse compression technologies, they achieve centimeter-level positioning accuracy and millisecond-level data refresh rate when the target enters the terminal interception stage (distance < 50 kilometers), ensuring real-time trajectory locking of hypersonic targets. The two types of radars form a stepped detection network through spatial distribution and frequency band complementarity, and combined with terrain matching algorithms and electromagnetic compatibility optimization, effectively balance the detection range, accuracy, and anti-interference ability.

[0083] Infrared Detectors: They mainly detect the thermal signals of incoming missiles, are sensitive to infrared radiation, and can play a unique advantage in complex electromagnetic interference environments, assisting radars in target tracking. Especially for incoming missile targets with high temperature characteristics, they can provide additional position and motion information, improving the reliability and accuracy of target detection. Using multi-spectral infrared focal plane arrays (wavelength coverage of 3 - 5μm and 8 - 12μm dual bands), they are deployed in a distributed networked manner at blind spot compensation nodes of radars (such as low-altitude defense positions, back slopes of mountains, etc.), focusing on covering areas where radars are vulnerable to electromagnetic interference or low-altitude penetration paths. The field of view angle of a single detector is designed to be 120°×90°. Through a time-space composite scanning mechanism, it realizes omnidirectional heat source tracking. The sampling rate of the infrared radiation characteristics of hypersonic targets (surface temperature ≥ 800°C) is not less than 200Hz, and a non-uniformity correction algorithm is used to eliminate environmental thermal noise, ensuring that ballistic heat trace coordinates with sub-second delay (accuracy ≤ 0.1mrad) can still be provided in complex electromagnetic environments. After data fusion with radars, the target positioning error can be reduced to within 3 meters.

[0084] Optoelectronic sensor: It can obtain the images and optical signals of the target through optical and electronic means. When the target approaches, it provides high-resolution target detail information for the system, such as the shape and contour of the target, further assisting in the precise identification and positioning of the target. It consists of a high-resolution visible light / short-wave infrared (SWIR) dual-mode imaging system and is deployed according to the defense depth gradient: The long-range optoelectronic tower (erected at a height of 50-100 meters) is equipped with a large-aperture optical lens (focal length ≥ 1000mm) and an adaptive optical image stabilization system to achieve pixel-level contour recognition of targets 100 kilometers away (resolution ≤ 0.05m / 10km); The short-range optoelectronic unit (integrated into the interception platform) uses a staring focal plane array to continuously capture high-definition images of the target at a frame rate of 30fps, and combines deep learning feature extraction algorithms to real-time identify the target type (warhead / decoy). All optoelectronic nodes achieve microsecond-level time synchronization through a fiber optic time synchronization network, and use multi-view stereo vision algorithms to reconstruct the three-dimensional motion trajectory of the target. After cross-verifying with infrared / radar data, the false alarm rate caused by interference such as clouds and smoke can be eliminated (false alarm probability < 0.1%).

[0085] The asynchronous acquisition data of multi-source sensors obtained usually need to go through preprocessing operations, and the preprocessing operations include:

[0086] Noise removal: Advanced filtering algorithms such as extended Kalman filtering are used to process the data from different sensors to remove measurement noise. By establishing a state estimation model, according to the dynamic characteristics of the system and the measurement equation, the noise is filtered from the original data, thereby improving the quality and reliability of the data.

[0087] Data fusion: The data from different sensors are fused and processed to integrate the advantages of different sensors and generate a unified target information description. For example, the distance and speed information of the radar are fused with the thermal signal information of the infrared detector to more comprehensively describe the state of the target. The data fusion algorithm may perform weighted processing on the data according to the accuracy, reliability of different sensors and their correlation to ensure that the fused data more accurately reflects the true state of the target.

[0088] Time synchronization: Since the sampling times and data transmission delays of different sensors may be different, this part will perform time synchronization operations on the data of different sensors and unify them to a common time reference. This involves time calibration of data with different timestamps to ensure the consistency of data from different sensors in the time dimension, so that subsequent algorithms can accurately process time series information.

[0089] Spatial calibration: Convert the data obtained by different sensors at different positions and angles into a unified spatial reference system, eliminate the data deviation caused by the differences in sensor positions and perspectives, and ensure that all data represents the position and motion information of the target in the same spatial framework.

[0090] Stitch the asynchronously acquired data into asynchronously time-series data t i 、t j and t k are the sampling times of sensors i, j, and k closest to the current time t, respectively. and are the observation values of sensors i, j, and k at the corresponding sampling times t i 、t j and t k respectively.

[0091] S2: Based on the pictures collected by the optoelectronic sensor, identify the type characteristics of the interception target; based on the asynchronously time-series data, use the asynchronous attention mechanism and the Transformer model to obtain the position and speed of the interception target at time t.

[0092] Specifically: Step S21: Based on the pictures collected by the optoelectronic sensor, use the existing graphic recognition method (refer to the existing technology) to identify the type characteristics of the interception target. The semantic information of the target type (such as warhead, decoy, drone) is mapped into a low-dimensional dense vector h through the image recognition module of the optoelectronic sensor and then through the Embedding Layer. embed . For example: If the target is classified as a "hypersonic warhead", then h embed may encode its structural characteristics (such as conical head, thermal protection layer texture).

[0093] Step S22: Based on the asynchronously time-series data, use the asynchronous attention mechanism and the Transformer model to obtain the position and speed of the interception target at time t.

[0094] Step S22 specifically includes:

[0095] S221: For the asynchronously time-series data from multi-source sensors adopt the asynchronous attention mechanism to dynamically adjust the weight A of the multi-source sensor data ij ;

[0096]

[0097] w(|t i -t j |) = exp(-γ|t i -t j |)

[0098] w(|t i -t k |) = exp(-γ|t i -t k |)

[0099] where t i 、t j and t k are the sampling times of sensors i, j, and k closest to the current time t, respectively, and are the observed values of sensors i, j, and k at the corresponding sampling times t i 、t j and t k respectively, γ is the time decay parameter used to control the influence of the time difference on the attention weight, and are the priority weights of the data points and respectively, M ij and M ik are the causal masks of the data points and respectively. When t j > t i or t k > t i ,both M ij and M ik are 0, otherwise they are 1; is the query vector of the data point x ti , and are the key vectors of the data points and respectively.

[0100] In the air defense and antimissile scenario, there are differences in the sampling frequencies and data transmission delays of different sensors (such as radar, infrared, and optoelectronic), resulting in the asynchronization of the timestamps of the data input into the system. For example: The long-range radar may update data at a frequency of 10 Hz (time interval 0.1 second), the short-range fire control radar updates at 100 Hz (time interval 0.01 second), and the infrared detector updates at 200 Hz (time interval 0.005 second). The asynchronous attention mechanism solves the problems of data asynchrony (different timestamps, different sampling rates) and heterogeneity (different types of sensor data) by dynamically adjusting the weights of multi-source sensor data. The priority weights are determined according to factors such as the reliability of the sensors and the confidence of the data. For example, the data of high-precision sensors have higher priority weights; the time-dependent causal masks M ij and M ik, which can ensure that only past information is used when processing data, avoid using future information, and guarantee causality.

[0101] S222: Input the weight A ij and asynchronous time series data into the Transformer model to generate the position x t and velocity v t of the interception target at time t;

[0102]

[0103] where v tj is the observed velocity of the j sensor at the corresponding sampling time t j .

[0104] It should be noted that if the time decay weight w(|t i -t j |) or the priority is set improperly, the fusion parameters will deviate from the true values, resulting in the failure of subsequent predictions and decisions.

[0105] S3: Obtain the threat value of the interception target according to the type characteristics, position and velocity of the interception target at time t.

[0106] In this embodiment, step S3 specifically includes:

[0107] S31: Concatenate the position and velocity of the interception target at time t and the type characteristics of the interception target to form a joint feature vector h joint ;

[0108] h joint = Concat(h embed , v t , x t )

[0109] where v t and x t are the position and velocity of the interception target at time t respectively, and h embed is the type characteristic of the interception target, all of which are three-dimensional data.

[0110] S32: Map the joint feature vector to the threat value T u of the interception target u through a multi-layer perceptron (MLP);

[0111] T u = W2·ReLU(W1h joint + b1)+ b2

[0112] where W1 and b1 are the weights of two dimensions used to extract non-linear interaction features, W2 and b2 are the weights of two dimensions for outputting the highlighted threat value T u ,

[0113] Step S2 outputs the fused motion parameters (x t , v t ), while the target type embedding h embed is usually generated by an independent classification module. The two are deeply fused through joint features, and finally the threat value is output by the MLP. This design realizes the dual guarantee of data-driven and physical laws.

[0114] S4: According to the threat value of the intercept target, the position and velocity at time t, use the physical constraint diffusion model to alternatively predict the short-term trajectory, medium-term trajectory or long-term trajectory of the intercept target.

[0115] Specifically, step S4 is

[0116] S41: Compare the threat value of the intercept target and the velocity at time t with the preset interval threshold, and alternatively select short-term prediction, medium-term prediction or long-term prediction, and obtain the corresponding prediction interval Δt based on the selected prediction trajectory length.

[0117] A: Short-term prediction (0 - 5 seconds)

[0118] Time interval: Adopt a smaller time step (for example, Δt short is 0.1 to 0.5 seconds), and update the target state at high frequency.

[0119] Technical implementation: Based on the real-time position, velocity and acceleration of the current target, combined with the dynamic equation of the physical constraint diffusion model (PCD), quickly predict the trajectory of the target in the next few seconds.

[0120] Application scenario: Instant interception decision-making: Provide accurate initial aiming parameters (such as launch angle, velocity) for launching interceptor missiles.

[0121] Dynamic trajectory correction: In the initial stage of the interceptor missile flight, combine real-time data to adjust the guidance command to cope with the sudden maneuver of the target.

[0122] B: Medium-term prediction (5 - 15 seconds)

[0123] Time interval: Adopt a moderate time step (for example, Δt medium is 1 to 2 seconds), and balance the calculation efficiency and prediction span.

[0124] Technical implementation: Synthesize the motion trend of the target (such as the change rate of acceleration), the external environment (such as atmospheric density, wind direction) and its own characteristics (such as mass, aerodynamic shape), and generate a multi-modal trajectory cluster through the diffusion model.

[0125] Application Scenario: Resource Dynamic Allocation: Optimize the allocation strategies of resources such as interceptors and radars according to the predicted trajectory clusters.

[0126] Cooperative Interception Planning: When there is a multi-target saturation attack, allocate the optimal interception paths for different interceptors to avoid resource conflicts.

[0127] C: Long-term Prediction (>15 seconds)

[0128] Time Interval: Adopt a relatively large time step (e.g., Δt long is 5 to 10 seconds) to reduce the computational load.

[0129] Technical Implementation: Focus on the macroscopic motion trends of the targets (such as strategic intentions and ballistic phase transitions), and combine strategic-level environmental data (such as the distribution of enemy launch positions and the influence of the earth's curvature) to predict the possible positions of the targets after a longer period. Strategic Defense Deployment: Adjust the regional defense forces in advance (such as deploying backup interception systems and evacuating important facilities).

[0130] Interception Missile Launch Window Planning: Predict the reentry points of hypersonic targets to provide a basis for pre-decision-making for the multi-layer interception system.

[0131] S42: Take the position x t and velocity v t of the interception target at time t as the initial state, and use the physical constraint diffusion model to predict the trajectory of the interception target according to the corresponding prediction interval Δt.

[0132] The physical constraint diffusion model is:

[0133]

[0134] where x t+Δt is the predicted position of the interception target at the next moment, β t is the diffusion coefficient; ∈ is a noise vector sampled from the standard normal distribution, and f(x t , v t , Δt) is the physical constraint term;

[0135]

[0136] where F aero (x t , v t , Δt) is the aerodynamic force, and C dis the air resistance coefficient, ρ is the atmospheric density, S is the reference area of the intercepted target. High-resolution images are obtained through optoelectronic sensors, and the target shape features (such as conical head, projectile size) are analyzed in combination with the target recognition module (such as convolutional neural network), and the typical reference area is matched in combination with the predefined target type database (such as hypersonic warhead, decoy, etc.).

[0137] F gravity (x t ,Δt)=mg

[0138] Among them, F gravity (x t ,Δt) is the gravitational force, g is the acceleration due to gravity, m is the weight of the intercepted target. Military research institutions will establish a target characteristic database to statistically model the mass of similar targets. By judging the characteristics such as target type and size, the empirical formula or sample data in the database is matched to estimate the mass of the unknown target. For example, for a certain type of small unmanned aerial vehicle, m can be quickly estimated through the "size-mass" empirical model.

[0139] The physical-constrained diffusion model (Physics-Constrained Diffusion, PCD) of this application is updated in position after training. When the physical-constrained diffusion model is trained, it simultaneously satisfies the energy conservation constraint loss function and the momentum conservation loss function;

[0140] The energy conservation constraint loss function L E is:

[0141]

[0142] The momentum conservation loss function L M is:

[0143]

[0144] Among them, N is the number of sampling points in the training process, E t is the predicted total energy at time t, E0 is the total energy at the initial time, p t is the momentum at time t, and p0 is the initial momentum.

[0145] The total energy E of the system = E k +E p , the kinetic energy The potential energy E p =mgh, where g is the acceleration due to gravity, h is the height of the target, and v is the target movement speed.

[0146] In the position update formula of this application, f(x t ,v t ,Δt) integrates physical laws into the diffusion process through physical constraint terms.

[0147] The physical constraint terms include aerodynamic forces and gravitational forces, and the actions of these forces are calculated through F aero and F gravity .

[0148] The energy conservation constraint ensures that the predicted trajectory satisfies the law of energy conservation, which is achieved through the energy conservation loss function L E .

[0149] The momentum conservation constraint ensures that the predicted trajectory satisfies the law of momentum conservation, which is achieved through the momentum conservation loss function L M .

[0150] The PCD model can more accurately predict the movement trajectory of the target and ensure that the prediction result conforms to the actual physical laws.

[0151] As a specific example: when the step S2 outputs the target position x t =(100km, 0.5°) and the speed v t =2km / s, these parameters will be used as the initial conditions of the diffusion model at time t to predict the state at the next moment. After prediction, the predicted values will be used as the parameters at the current moment to continue the prediction until the set prediction period (prediction length) is reached. When the position at the next moment is predicted, the speed at the next moment is obtained according to the change in position.

[0152] S5: Generate multiple interception plans for all interception targets based on the predicted trajectories, threat values of each interception target, and the performance of all interception missiles, and preferentially select the optimal plan in the interception plans for interception operations.

[0153] Step S5: Specifically includes:

[0154] S51: For each interception target, based on the threat value of the interception target and the performance of all types of interception missiles, use the resource allocation function A(I i ) to obtain the allocation probability of each type of interception missile for the current interception target.

[0155]

[0156] P(I i ) = w4R interceptor + w5v interceptor + w6M interceptor

[0157] where, T u is the threat value of the interception target u, P(I i ) is the performance score of the interception missile I i , R interceptor , vinterceptor and M interceptor For the interceptor missile I i The range, speed and maneuverability, and the maneuverability can be quantified by the maximum maneuvering ability. In this embodiment, the maneuverability is scaled to the range of 0-1; w4, w5 and w6 are respectively for R interceptor , v interceptor and M interceptor corresponding weights, and n is the total number of interceptor missile types.

[0158] S52: Take all types of interceptor missiles that meet the set allocation probability threshold for each interception target as feasible solutions, and obtain the launch time and interception position of each feasible solution.

[0159] Specifically, S52 is as follows:

[0160] S521: For each interception target, based on the predicted trajectory of the interception target, fit the movement process of the interception target to a uniformly accelerated motion, and construct a first uniformly accelerated model;

[0161]

[0162] S522: Simplify the interception movement of each type of interceptor missile for the current interception target to a uniformly accelerated motion, and construct a second uniformly accelerated model;

[0163]

[0164] S523: At the interception moment, the interception target and the interceptor missile need to satisfy the following formula, and use the Newton-Raphson iteration method to solve the launch time and interception position of the interceptor missile for the current interception target.

[0165]

[0166] where, t intercept is the interception moment of the interceptor missile, x target (t intercept ) and y target (t intercept ) are the interception position coordinates of the interception target, t x is the flight time, x interceptor (t intercept ) and y interceptor (t intercept ) are the position coordinates of the interceptor missile, a x,target , a y,target , v x,target (0) and v y,target (0) are respectively the accelerations and initial velocities of the interception target in the x and y directions, a x , a y , v x0and v yo are respectively the acceleration and initial velocity of the intercept missile in the x and y directions; t launch is the launch time of the intercept missile; x launch and y launch are the launch positions of the intercept missile, and x target (0) and y target (0) are the initial positions of the intercept target.

[0167] The Newton-Raphson iteration method refers to the prior art and needs to satisfy basic constraints during solution, such as range constraint and the constraint on the launch time t intercept >t launch and so on.

[0168] S53: Combine all the feasible solutions of the intercept targets to obtain the set of initial intercept plans for intercepting all the intercept targets, and exclude the intercept plans that do not meet the set launch requirements.

[0169] Not meeting the set launch requirements means performing a preliminary screening on the initial intercept plans. Those with a total launch quantity greater than the stockpile and exceeding the system launch interval, etc., can be reasonably adjusted and set specifically according to expert experience or the platform capabilities of the launch system.

[0170] S54: For all the intercept plans after exclusion, use an optimization algorithm to find the optimal plan for intercept operation.

[0171] Specifically, it includes: using the simulated annealing method for evaluation to find the optimal plan that minimizes the objective function O;

[0172]

[0173] Among them, C, E, and R are respectively the intercept cost, intercept error, and resource utilization rate of the current intercept plan, and w c , w e and w r are respectively the weights of the intercept cost, intercept error, and resource utilization rate; C max , E max and R ideal are respectively the set maximum intercept cost, intercept error, and resource utilization rate. The acquisition of the resource utilization rate can refer to the prior art. The resource utilization rate is usually the ratio of the actually used resource amount to the total available resource amount, such as fuel, electricity, time, equipment, etc.

[0174] As Figure 2 shown, Figure 2 in (a) is the comparison chart of the trajectory prediction accuracy between this application and the traditional method, Figure 2 in (b) is the comparison chart of the decision-making real-time performance between this application and the traditional method, Figure 2Among them, (c) is the comparison chart of the interception effectiveness of typical scenarios between this application and traditional methods, Figure 2 Among them, (d) is the comparison chart of comprehensive performance indicators between this application and traditional methods, Figure 2 Among them, (e) is the comparison chart of the energy conservation constraint effect between this application and traditional methods, Figure 2 Among them, (f) is the visual comparison chart of the 3D interception effect between this application and traditional methods.

[0175] It can be seen from Figure 2 that the short-term prediction error of this application is reduced by 66% (only 3.1 meters), and the long-term prediction error is reduced by 64% (185 meters); the decision response time is optimized from 450 milliseconds to 120 milliseconds (a 73% improvement), and at the same time, the computing resource occupancy is reduced by 35%; in the three typical scenarios of hypersonic targets, saturation attacks, and maneuvering targets, the interception success rates reach 73%, 66%, and 82% respectively, which are significantly improved compared with traditional systems; through energy conservation constraints, the total energy deviation between the predicted trajectory and the real trajectory is controlled within ±5%; the 3D interception effect intuitively presents an earlier and more accurate interception process, and the comprehensive data verifies the overall leading advantage of this system in the three key indicators of accuracy, speed, and success rate, providing reliable technical support for dealing with hypersonic weapons and cluster attacks.

[0176] Embodiment 2

[0177] As Figure 3 shown, an air defense and antimissile intelligent interception system is provided, including:

[0178] A data acquisition module, which is used to obtain asynchronous acquisition data of multi-source sensors closest to the current moment t for each interception target and splice them into asynchronous time series data. The multi-source sensors include several radar sensors, optoelectronic sensors, and infrared detectors;

[0179] An ACT module, which is used to identify the type characteristics of the interception target based on the collected pictures of the optoelectronic sensor; and based on the asynchronous time series data, use the asynchronous attention mechanism and the Transformer model to obtain the position and speed of the interception target at moment t;

[0180] A threat value acquisition module, which is used to obtain the threat value of the interception target according to the type characteristics of the interception target, the position and speed at moment t;

[0181] A PCD trajectory prediction module, which is used to select and predict the short-term trajectory, medium-term trajectory, or long-term trajectory of the interception target according to the threat value of the interception target, the position and speed at moment t, using the physical constraint diffusion model;

[0182] An interception plan generation module, which is used to generate multiple interception plans for intercepting all interception targets based on the predicted trajectories, threat values of each interception target, and the performance of all interception missiles, and preferentially select the optimal plan among the interception plans for interception operations.

[0183] For a more specific process of the above method, reference can be made to the corresponding content disclosed in the foregoing embodiments, and details will not be elaborated herein.

[0184] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other. For the system disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and reference can be made to the description of the method part for related parts.

[0185] Those skilled in the art can clearly understand that the technology in the embodiments of the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solutions in the embodiments of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments of the present invention.

[0186] The above are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements made without departing from the principle of the present invention should be regarded as within the protection scope of the present invention.

Claims

1. An air defense and antimissile intelligent interception method, characterized in that, Including: For each interception target, obtain the asynchronous acquisition data of multi-source sensors closest to the current moment t, and splice them into asynchronous time series data. The multi-source sensors include several radar sensors, optoelectronic sensors, and infrared detectors; Based on the acquired images of the optoelectronic sensors, identify the type characteristics of the interception target; Based on the asynchronous time series data, use the asynchronous attention mechanism and the Transformer model to obtain the position and velocity of the interception target at moment t; According to the type characteristics, position, and velocity of the interception target at moment t, obtain the threat value of the interception target; According to the threat value, position, and velocity of the interception target at moment t, use the physical constraint diffusion model to alternatively predict the short-term trajectory, medium-term trajectory, or long-term trajectory of the interception target; Based on the predicted trajectories, threat values of each interception target, and the performance of all interception missiles, generate multiple interception plans for intercepting all interception targets, and preferentially select the optimal plan among the interception plans for interception operations.

2. The air defense and antimissile intelligent interception method according to claim 1, characterized in that The step of using the asynchronous attention mechanism and the Transformer model based on the asynchronous time series data to obtain the position and velocity of the interception target at moment t is specifically as follows: For asynchronous time series data from multi-source sensors An asynchronous attention mechanism is adopted to dynamically adjust the weight A of multi-source sensor data ij ; w(|t i -t j |) = exp(-γ|t i -t j |) w(|t i -t k |)=exp(-γ|t i -t k |) where t i 、t j and t k are the sampling times of sensors i, j, and k closest to the current time t, and are the observed values of sensors i, j, and k at the corresponding sampling times t i 、t j and t k respectively, and γ is the time decay parameter, and are the priority weights of data points and respectively, M ij and M ik are the causal masks of data points and respectively. When t j > t i or t k > t i , both M ij and M ik are 0, otherwise they are 1; is the query vector of data point , and are the key vectors of data points and respectively; Input weight A ij and asynchronous time series data into the Transformer model to generate the position x t and velocity v t ; Among them, is the observed speed of the j sensor at the corresponding sampling moment t j .

3. The air defense and antimissile intelligent interception method according to claim 1, wherein The step of obtaining the threat value of the interception target according to the type characteristics, position, and velocity of the interception target at moment t is specifically as follows: Concatenate the position and velocity of the intercept target at time t and the type characteristics of the intercept target to form a joint feature vector h joint ; h joint = Concat(h embed , v t , x t ) where v t and x t are the position and velocity of the intercept target at time t, respectively, and h embed is the type feature of the intercept target, Mapping the joint feature vector to the threat value T of the intercepted target u through a multi-layer perceptron u ; T u = W2·ReLU(W1h joint + b1) + b2 Among them, W1 and b1 are the weights of two dimensions for extracting non-linear interaction features, W2 and b2 are the weights of two dimensions for outputting the highlighted threat value T u of two dimensions, 4. The air defense and antimissile intelligent interception method according to claim 1, characterized in that The step of using the physical constraint diffusion model to alternatively predict the short-term trajectory, medium-term trajectory, or long-term trajectory of the interception target according to the threat value, position, and velocity of the interception target at moment t; specifically, Compare the threat value of the interception target and the velocity at moment t with the preset interval threshold, and alternatively select short-term prediction, medium-term prediction, or long-term prediction, and obtain the corresponding prediction interval Δt based on the selected prediction trajectory length; Take the position x of the intercept target at time t t and the velocity v t as the initial state, and use the physical constraint diffusion model to predict the intercept target trajectory according to the corresponding prediction interval Δt; The physical constraint diffusion model is: where x t+Δt is the predicted position of the intercepted target at the next moment, β t is the diffusion coefficient; ∈ is a noise vector sampled from a standard normal distribution, and f(x t , v t , Δt) is the physical constraint term; Among them, F aero (x t , v t , Δt) is the aerodynamic force, C d is the air resistance coefficient, ρ is the atmospheric density, and S is the reference area of the intercepted target; F gravity (x t , Δt) = mg Among them, F gravity (x t , Δt) is the gravitational force, m is the weight of the intercepted target, and g is the acceleration due to gravity.

5. The air defense and antimissile intelligent interception method according to claim 4, characterized in that The physical constraint diffusion model updates the position after training. When the physical constraint diffusion model is trained, it simultaneously satisfies the energy conservation constraint loss function and the momentum conservation loss function; The energy conservation constraint loss function L E is as follows: The momentum conservation loss function L M is as follows: where N is the number of sampling points during the training process, E t is the total predicted energy at time t, E0 is the total energy at the initial time, p t is the momentum at time t, and p0 is the initial momentum.

6. The anti-aircraft and anti-missile intelligent interception method according to claim 1, wherein The step of generating multiple interception plans for intercepting all interception targets based on the predicted trajectories, threat values of each interception target, and the performance of all interception missiles, and preferentially selecting the optimal plan among the interception plans for interception operations specifically includes: For each interception target, based on the threat value of the interception target and the performance of all types of interception missiles, the resource allocation function A(I i ) is used to obtain the allocation probability of each type of interception missile for the current interception target; P(I i ) = w4R interceptor + w5v interceptor + w6M interceptor Among them, T u is the threat value of the intercepted target u, P(I i ) is the performance score of the interceptor missile I i . R interceptor , v interceptor and M interceptor are the range, speed and maneuverability of the interceptor missile I i ; w4, w5 and w6 are the weights corresponding to R interceptor , v interceptor and M interceptor respectively, and n is the total number of interceptor missile types; Take all types of interception missiles that meet the set allocation probability threshold for each interception target as feasible solutions, and obtain the launch time and interception position of each feasible solution; Combine the feasible solutions of all interception targets to obtain an initial interception plan set for intercepting all interception targets, and exclude the interception plans that do not meet the set launch requirements; For all the intercepted plans after exclusion, use an optimization algorithm to find the optimal plan for interception operations.

7. The anti-aircraft and anti-missile intelligent interception method according to claim 6, characterized in that, The step of taking all types of interception missiles that meet the set allocation probability threshold for each interception target as feasible solutions, and obtaining the launch time and interception position of each feasible solution is specifically as follows: For each interception target, based on the predicted trajectory of the interception target, fit the movement process of the interception target into a uniformly accelerated motion, and construct a first uniformly accelerated model; Simplify the interception motion of each type of interception missile for the current interception target into a uniformly accelerated motion, and construct a second uniformly accelerated model; At the interception moment, the intercepted target and the interceptor missile need to satisfy the following formula, and use the Newton-Raphson iteration method to solve the launch time and interception position of the interceptor missile for the current intercepted target for the following formula. Among them, t intercwpt is the interception moment of the interceptor missile, x target (t intercwpt ) and y target (t intercept ) are the interception position coordinates of the intercepted target, t x is the flight time, x interceptor (t intercept ) and y interceptor (t intercept ) are the position coordinates of the interceptor missile, a x,target , a y,target , v x,target (0) and v y,target (0) are respectively the accelerations and initial velocities of the intercepted target in the x and y directions, a x , a y , v x0 and v yo are respectively the accelerations and initial velocities of the interceptor missile in the x and y directions; t launch is the launch time of the interceptor missile; x launch and y kaunch are the launch positions of the interceptor missile, x target (0) and y target (0) are the initial positions of the intercepted target.

8. The air defense and antimissile intelligent interception method according to claim 6, characterized in that For all the intercepted scenarios after exclusion, use an optimization algorithm to find the optimal scenario for interception operations, specifically including: using the simulated annealing method for evaluation to find the optimal scenario that minimizes the objective function O; Among them, C, E, and R are the interception cost, interception error, and resource utilization rate of the current interception scheme, respectively, and w c , w e , and w r are the weights of the interception cost, the weight of the interception error, and the weight of the resource utilization rate, respectively; C max , E max , and R ideal are the set maximum interception cost, interception error, and resource utilization rate, respectively.

9. The air defense and antimissile intelligent interception method according to claim 1, characterized in that According to the threat value of the intercepted target, the position and velocity at time t, use the physical constraint diffusion model to alternatively predict the short-term trajectory, medium-term trajectory or long-term trajectory of the intercepted target. The prediction length of the short-term trajectory is less than 5 seconds, the prediction length of the medium-term trajectory is 5 - 15 seconds, and the prediction length of the long-term trajectory is greater than 15 seconds.

10. An air defense and antimissile intelligent interception system, characterized in that, Including: A data acquisition module, which is used to obtain the asynchronous acquisition data of multi-source sensors closest to the current time t for each intercepted target and splice them into asynchronous time series data. The multi-source sensors include several radar sensors, optoelectronic sensors and infrared detectors; An ACT module, which is used to identify the type characteristics of the intercepted target based on the acquired pictures of the optoelectronic sensors; Based on the asynchronous time series data, use the asynchronous attention mechanism and the Transformer model to obtain the position and velocity of the intercepted target at time t; A threat value acquisition module, which is used to obtain the threat value of the intercepted target according to the type characteristics, position and velocity of the intercepted target at time t; A PCD trajectory prediction module, which is used to alternatively predict the short-term trajectory, medium-term trajectory or long-term trajectory of the intercepted target according to the threat value, position and velocity of the intercepted target at time t using the physical constraint diffusion model; An interception scenario generation module, which is used to generate multiple interception scenarios for intercepting all intercepted targets based on the predicted trajectories, threat values of each intercepted target and the performance of all interceptor missiles, and select the optimal scenario from the interception scenarios for interception operations.

Images