A hierarchical identification method for target tactical intentions in an air combat simulation environment

Abstract

This invention discloses a hierarchical identification method for target tactical intentions in an air combat simulation environment. First, it extracts target trajectory feature information and relative enemy / friendly situation information, whereby the target trajectory feature information includes instantaneous and cumulative feature information. Then, based on four cases of target trajectory deviation angle and target trajectory inclination angle, it finds target trajectory decomposition feature points and decomposes the target trajectory into a series of continuous maneuvering elements. Next, it designs target maneuvering element state sets, tactical maneuvering state sets, and tactical intention state sets for beyond-visual-range air combat. Finally, it designs a cascaded support vector machine multi-classification algorithm architecture for hierarchical identification of target tactical intentions, achieving hierarchical identification of target tactical intentions. This invention effectively overcomes the convergence and timeliness challenges when the feature information dimension and the number of training samples are large, achieving multi-dimensional online accurate identification of target tactical intentions.

Description

Technical Field

[0001] This invention belongs to the field of flight technology, specifically relating to a method for hierarchical identification of target tactical intentions. Background Technology

[0002] Air combat games and simulation systems both utilize computer simulation to meticulously and realistically simulate the entire combat process of fighter jets. To effectively enhance the realism and ease of use of the user experience, simulation design needs to be based on actual air combat. Accurately identifying target tactical intentions in the highly dynamic and intensely contested environment of real air combat is a crucial foundation for achieving deep situational awareness and intelligent decision-making. It is also key to improving the human-computer interaction and intelligence level of combat games and simulation systems, and is of great significance to the realism and credibility of simulation design.

[0003] Air combat target tactical intent identification is a data mining and online reasoning process. It extracts key feature information from the target's motion state acquired by airborne sensors and perceives the current target behavior and relative situation as quickly and accurately as possible to gradually uncover the target's potential tactical intent. Accurate online identification of air combat target tactical intent can create tactical advantages for our aircraft, improve air combat decision-making efficiency, and thus gain air combat superiority.

[0004] Existing research on this problem generally categorizes target intentions from a mission perspective into types such as surveillance, reconnaissance, jamming, cover, and attack. However, in actual air combat, on the one hand, it is difficult to effectively identify these target intentions solely through target track information; on the other hand, identifying more specific adversarial intentions from a tactical and operational perspective is more practically significant. Current identification methods mainly focus on the application of theories such as expert systems and machine learning. While expert systems possess strong knowledge representation and reasoning capabilities, they require the construction of a complete rule system, resulting in low learning and fault tolerance capabilities, making it difficult to encompass the complex evolution of air combat. Bayesian networks, as a machine learning method with strong causal probabilistic reasoning performance, adapt to changes in the air combat situation through dynamic updates of network parameters. However, their network models are relatively complex, and the determination of network parameters such as prior probabilities and conditional probabilities also relies on expert experience, resulting in a high degree of subjectivity. Some literature also uses neural networks and other methods to transform intent recognition into a pattern recognition problem of data classification. However, the shortcomings of this approach are that the implicit mapping relationship makes the recognition results difficult to interpret and makes it impossible to intuitively analyze the composition of tactical intent. Directly learning the target's tactical intent from high-dimensional data also consumes a lot of time and computational costs, thereby reducing the convergence of the algorithm. In addition, the introduction of key temporal features in the trajectory information and the enhancement of the model's generalization ability are also aspects that need to be improved in this type of method. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides a hierarchical identification method for target tactical intentions in an air combat simulation environment. First, it extracts target trajectory feature information and relative enemy / friendly situation information, where the target trajectory feature information includes instantaneous and cumulative feature information. Then, based on four cases of target trajectory deviation angle and target trajectory inclination angle, it finds target trajectory decomposition feature points and decomposes the target trajectory into a series of continuous maneuvering elements. Next, it designs target maneuvering element state sets, tactical maneuvering state sets, and tactical intention state sets for beyond-visual-range air combat. Finally, it designs a cascaded support vector machine multi-classification algorithm architecture for hierarchical identification of target tactical intentions, achieving hierarchical identification of target tactical intentions. This invention effectively overcomes the convergence and timeliness challenges when the feature information dimension and the number of training samples are large, achieving multi-dimensional online accurate identification of target tactical intentions.

[0006] The technical solution adopted by this invention to solve its technical problem includes the following steps:

[0007] Step 1: Extract target trajectory feature information and relative enemy and friendly situation information, wherein the target trajectory feature information includes instantaneous feature information and cumulative feature information;

[0008] Instantaneous feature information includes the track inclination angle θ t , Track inclination rate Rate of change of track deflection and height change rate Cumulative feature information includes the change in track inclination C θ Change in track deviation C φ ;

[0009] Relative situation information includes the target azimuth q. f , enter angle q j and relative distance change rate

[0010] Step 2: Based on the four situations of changing target track deflection angle, cumulative change of target track deflection angle exceeding 90°, change of sign of target track inclination angle, and change of target track inclination angle trend, find the target track decomposition feature points and decompose the target track into a series of continuous maneuvering elements.

[0011] Step 3: Design the target maneuver state set, tactical maneuver state set, and tactical intent state set for beyond-visual-range air combat respectively;

[0012] Step 4: Design a cascaded support vector machine multi-classification algorithm architecture for hierarchical recognition of target tactical intentions, specifically including:

[0013] Step 4-1: Design the optimal classification function for the cascaded support vector machine;

[0014] For a training sample set of size m {(x1,y1),(x2,y2),...,(x m ,y m Let the optimal classification function be} in, Let be the radial basis function kernel, and σ be the width parameter of the radial basis function kernel. b is the optimal solution for the Lagrange coefficients corresponding to the training samples. * For the corresponding optimal classification hyperplane displacement term, 0 ≤ α i ≤C, where C is a constant penalty factor;

[0015] During the training phase of the cascaded support vector machine, each support vector machine is trained separately for training samples of target maneuver elements, tactical maneuvers, and tactical intentions, and the recognition result of each layer corresponds to one of the feature information of the next layer's training samples.

[0016] Step 4-2: Design hierarchical training samples for target tactical intent;

[0017] The training samples for the target maneuver element are the instantaneous feature information θ of the target trajectory. t , To the motor element number P ME The mapping; tactical maneuver training samples are designed based on tactical maneuvers composed of four-segment, three-segment, and two-segment maneuvers, respectively, i.e., P of consecutive four-segment, three-segment, and two-segment maneuvers. ME C φ and C θ To tactical maneuver number P TM The mapping; the tactical intent training sample is the tactical maneuver number P. TM and relative situation information q f q j , To tactical intent number P TI Mapping;

[0018] The sample data is resampled using an undersampling method and normalized using the maximum and minimum values ​​of each feature variable;

[0019] Step 4-3: Design the target tactical intent hierarchical identification logic to achieve target tactical intent hierarchical identification;

[0020] Tactical maneuver identification involves the segmentation and combination of multiple maneuver elements. The priority of identifying four-segment maneuvers is higher than that of three-segment and two-segment maneuvers, respectively. Specifically, tactical maneuver identification is performed first for four-segment maneuvers; if a maneuver belongs to that category, the tactical maneuver number P is directly output. TM Otherwise, it is identified as a non-four-segment maneuver element and a three-segment maneuver element tactical maneuver identification is performed. If it belongs to this category, the tactical maneuver number P is output directly.TM Otherwise, it is identified as a non-three-segment maneuver element and a two-segment maneuver element tactical maneuver identification is performed. If it belongs to this category, the tactical maneuver number P is directly output. TM Otherwise, directly output the corresponding maneuver element number P. ME The three types of tactical maneuvers mentioned above are trained separately using support vector machine models to form relatively independent recognition modules. During the training process, the optimal combination of C and σ model parameters is found through cross-validation.

[0021] Furthermore, the parameters in the target trajectory feature information and the relative enemy-friendly situation information are defined as follows:

[0022] Instantaneous features in target track information include: track inclination angle. Rate of change of track inclination Rate of change of track deflection and height change rate Where △x t =x t (t)-x t (th), △y t =y t (t)-y t (th) and △z t =z t (t)-z t (th) represents the displacement of the target in the three axes of the geographic coordinate system at time t, [x t (t),y t (t),z t [(t)] represents the target's position coordinates in the geographic coordinate system at time t, and h is the simulation step size.

[0023] The cumulative feature information in the target track information includes: track inclination change C θ =|θ t (t end )-θ t (t start )| and the change in track deviation C φ =|φ t (t end )-φ t (t start )|,C θ C φ ∈[0,π / 2], where t start and t end These are the start and end times of the target maneuver element, respectively;

[0024] Relative situation information includes target azimuth. Target entry angle and relative distance change rate Where D x D y D z These are the projections of the relative distance D between the enemy and ourselves onto the three axes of the geographic coordinate system, v. ux v uy v uz and v tx v ty v tz These are the speeds of my machine, v. u and target speed v t Projection along the three axes of the geographic coordinate system.

[0025] Furthermore, the target maneuver state set, tactical maneuver state set, and tactical intent state set are specifically as follows:

[0026] The state set of the maneuver element is S ME =[ME1,ME2,...,ME i ], i = 1, 2, ... 17, representing 17 maneuvering elements respectively: steady level flight, horizontal left turn, horizontal right turn, straight climb, straight descent, vertical concave climb, vertical convex climb, vertical concave descent, vertical convex descent, left turn concave climb, left turn convex climb, left turn concave descent, left turn convex descent, right turn concave climb, right turn convex climb, right turn concave descent, and right turn convex descent;

[0027] The tactical maneuver state set is S TM =[TM1,TM2,...,TM j ],j=1,2,...15, representing 15 tactical maneuvers respectively: steady level flight, horizontal left turn, horizontal right turn, horizontal S-shape, climb S-shape, dive S-shape, rapid climb, tactical dive, break-S, half loop, left interruption, right interruption, left combat turn, right combat turn, and drum maneuver.

[0028] The tactical intent state set is S TI =[TI1,TI2,...,TI k ], k = 1, 2, ... 5, representing five tactical intentions: attack, defense, evasion, transition in, and transition out.

[0029] Furthermore, the relevant parameters for finding the optimal combination of C and σ model parameters through cross-validation are as follows: the target maneuver element recognition model parameters are C=4 and σ=0.25; the tactical maneuver recognition model parameters for four-segment, three-segment, and two-segment maneuver element classes are C=2 and σ=0.47, C=1.4 and σ=0.21, and C=1.6 and σ=0.45, respectively; and the target tactical intent recognition model parameters are C=4 and σ=0.28.

[0030] The beneficial effects of this invention are as follows:

[0031] 1. This invention constructs a hierarchical identification model by introducing target trajectory feature information and relative enemy / friendly situation information step by step through online automatic decomposition and hierarchical progression of target trajectories. This model consists of target maneuvering elements, tactical maneuvers, and tactical intentions. A cascaded support vector machine multi-classification algorithm is specifically designed for this model, decomposing the computational complexity of traditional models through a cascaded architecture.

[0032] 2. This invention effectively overcomes the convergence and timeliness problems when the feature information dimension and the number of training samples are large. At the same time, the recognition results of each layer can support the analysis of the target's tactical intentions, making the output conclusions easy to interpret, thereby realizing multi-dimensional online accurate recognition of the target's tactical intentions. Attached Figure Description

[0033] Figure 1 This is a flowchart of the implementation method of the present invention.

[0034] Figure 2 This is a structural diagram of the cascaded support vector machine multi-classification algorithm for hierarchical recognition of target tactical intentions according to the present invention.

[0035] Figure 3 These are the feature points representing the decomposition of the three-dimensional flight paths of both sides and the target flight path in an air combat scenario according to an embodiment of the present invention.

[0036] Figure 4 This is the target maneuvering element identification result in an embodiment of the present invention.

[0037] Figure 5 This is the target tactical maneuver identification result in an embodiment of the present invention.

[0038] Figure 6 This is the result of target tactical intent recognition in an embodiment of the present invention. Detailed Implementation

[0039] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0040] This invention provides a hierarchical method for identifying target tactical intentions in an air combat simulation environment. It aims to hierarchically decompose target tactical intentions in beyond-visual-range (BVR) air combat, designing a hierarchical identification architecture and a cascaded support vector machine (SVM) algorithm to overcome convergence and timeliness issues when the feature information dimension and training sample size are large. Simultaneously, it utilizes the identification results of each layer to support the structural analysis of target tactical intentions, making the output conclusions easy to interpret. This achieves accurate online identification of target tactical intentions in BVR air combat, thereby improving the human-computer interaction and intelligence level of combat games and simulation systems.

[0041] like Figure 1As shown, a method for hierarchical identification of target tactical intentions in an air combat simulation environment includes the following steps:

[0042] Step 1: Extract target trajectory feature information and relative enemy and friendly situation information, wherein the target trajectory feature information includes instantaneous feature information and cumulative feature information;

[0043] Instantaneous feature information includes track inclination angle Rate of change of track inclination Rate of change of track deflection and height change rate Where △x t =x t (t)-x t (th), △y t =y t (t)-y t (th) and △z t =z t (t)-z t (th) represents the displacement of the target in the three axes of the geographic coordinate system at time t, [x t (t),y t (t),z t [(t)] represents the target's position coordinates in the geographic coordinate system at time t, and h is the simulation step size.

[0044] Cumulative feature information includes the change in track inclination C θ =|θ t (t end )-θ t (t start )| and the change in track deviation C φ =|φ t (t end )-φ t (t start )|,C θ C φ ∈[0,π / 2], where t start and t end These are the start and end times of the target maneuver element, respectively;

[0045] Relative situation information includes target azimuth. Target entry angle and relative distance change rate Where D x D y D z These are the projections of the relative distance D between the enemy and ourselves onto the three axes of the geographic coordinate system, v. ux v uy v uz and vtx v ty v tz These are the speeds of my machine, v. u and target speed v t Projection along the three axes of the geographic coordinate system.

[0046] Step 2: Based on the four situations of changing target track deflection angle, cumulative change of target track deflection angle exceeding 90°, change of sign of target track inclination angle, and change of target track inclination angle trend, find the target track decomposition feature points and decompose the target track into a series of continuous maneuvering elements.

[0047] Step 3: Design the target maneuvering element, tactical maneuver, and tactical intent state sets for beyond-visual-range air combat, respectively. The maneuvering element state set is S. ME =[ME1,ME2,...,ME i [i = 1, 2, ... 17], representing 17 maneuvering elements: steady level flight, horizontal left turn, horizontal right turn, straight climb, straight descent, vertical concave climb, vertical convex climb, vertical concave descent, vertical convex descent, left turn concave climb, left turn convex climb, left turn concave descent, left turn convex descent, right turn concave climb, right turn convex climb, right turn concave descent, and right turn convex descent; the tactical maneuvering state set is S. TM =[TM1,TM2,...,TM j ],j=1,2,...15, representing 15 tactical maneuvers including steady level flight, horizontal left turn, horizontal right turn, horizontal S-shape, climb S-shape, dive S-shape, rapid climb, tactical dive, break-S, half-loop, left interruption, right interruption, left combat turn, right combat turn, and roller maneuver; the tactical intent state set is S TI =[TI1,TI2,...,TI k ], k = 1, 2, ... 5, representing five tactical intentions: offense, defense, evasion, transition in, and transition out, respectively;

[0048] Step 4: Design a cascaded support vector machine multi-classification algorithm architecture for hierarchical recognition of target tactical intentions, specifically including:

[0049] Step 4-1: Design the optimal classification function for the cascaded support vector machine;

[0050] For a training sample set of size m {(x1,y1),(x2,y2),...,(x m ,y m Let the optimal classification function be} in, Let be the radial basis function kernel, and σ be the width parameter of the radial basis function kernel. b is the optimal solution for the Lagrange coefficients corresponding to the training samples. * For the corresponding optimal classification hyperplane displacement term, 0 ≤ α i ≤C, where C is a constant penalty factor;

[0051] like Figure 2 As shown, during the training phase of the cascaded support vector machine, each support vector machine is trained separately for training samples of target maneuver elements, tactical maneuvers, and tactical intentions, and the recognition result of each layer corresponds to one of the feature information of the next layer training sample.

[0052] Step 4-2: Design hierarchical training samples for target tactical intent;

[0053] The training samples for the target maneuver element are the instantaneous feature information θ of the target trajectory. t , To the motor element number P ME The mapping; considering the special nature of tactical maneuvers consisting of multiple maneuver elements, the tactical maneuver training samples are designed separately for tactical maneuvers composed of four, three, and two maneuver elements, that is, the P of the most recent consecutive four, three, and two maneuver elements. ME C φ and C θ To tactical maneuver number P TM The mapping; the tactical intent training sample is the tactical maneuver number P. TM and relative situation information q f q j , To tactical intent number P TI Mapping;

[0054] To avoid imbalance in the number of positive and negative samples and inconsistencies in the range of feature variables, an undersampling method is used to resample the sample data and then normalize it. Where X is the true value of a certain feature variable, X min and X max Let X be the minimum and maximum values ​​of X in the training samples, and Y be the normalized value of X.

[0055] Step 4-3: Design the target tactical intent hierarchical identification logic to achieve target tactical intent hierarchical identification;

[0056] Tactical maneuver identification involves the segmentation and combination of multiple maneuver elements. The priority of identifying four-segment maneuvers is higher than that of three-segment and two-segment maneuvers, respectively. Specifically, tactical maneuver identification is performed first for four-segment maneuvers; if a maneuver belongs to that category, the tactical maneuver number P is directly output. TM Otherwise, it is identified as a non-four-segment maneuver element and a three-segment maneuver element tactical maneuver identification is performed. If it belongs to this category, the tactical maneuver number P is output directly.TM Otherwise, it is identified as a non-three-segment maneuver element and a two-segment maneuver element tactical maneuver identification is performed. If it belongs to this category, the tactical maneuver number P is directly output. TM Otherwise, directly output the corresponding maneuver element number P. ME The three types of tactical maneuvers mentioned above are trained separately using support vector machine models, forming relatively independent recognition modules. In addition, to improve the training efficiency and recognition accuracy of the models, the optimal combination of C and σ model parameters is found through cross-validation.

[0057] The optimal combination of C and σ model parameters found through cross-validation is as follows: Target maneuver element recognition model parameters are C=4, σ=0.25; Tactical maneuver recognition model parameters for four-segment, three-segment, and two-segment maneuver element classes are C=2, σ=0.47, C=1.4, σ=0.21, and C=1.6, σ=0.45, respectively; Target tactical intent recognition model parameters are C=4, σ=0.28. Specific implementation examples:

[0059] First, a large number of target tracks with different initial track angles, speeds, altitudes, and process control variables are generated based on the flight simulation program. By calculating the instantaneous feature information of the target tracks and the corresponding maneuvering element types, 2551 sets of data are randomly selected as training sets and 1310 sets of data are selected as test sets. The test results show that the maneuvering element recognition layer in this invention can achieve an average recognition accuracy of 94.21% for various maneuvering elements.

[0060] Based on the target trajectory information and maneuver element identification results generated by the flight simulation program, the P values ​​of four consecutive maneuver elements are calculated. ME and its C φ and C θ Then, 1279, 1316, and 950 training datasets for four-segment, three-segment, and two-segment maneuver metaclasses were randomly selected and generated, respectively, along with 610, 657, and 473 test datasets. The tests showed that the tactical maneuver recognition layer in this invention can achieve average recognition accuracies of 92.87%, 93.37%, and 95.85% for four-segment, three-segment, and two-segment maneuver metaclasses, respectively.

[0061] Based on the beyond-visual-range air combat simulation program, a large amount of combat flight trajectory data of both sides is generated, and corresponding P values ​​are identified, calculated, and given. TM q f q j , and P TI Information such as these was collected, and then 3142 sets of data were randomly selected as the training set and 950 sets of data were selected as the test set. The test results showed that the tactical intent recognition layer in this invention can achieve an average target tactical intent recognition accuracy of 93.24%.

[0062] To comprehensively verify the hierarchical recognition model and algorithm in this invention, a simulation test was conducted on the hierarchical and coupled online recognition effect of target maneuvering elements, tactical maneuvers, and tactical intentions using a segment of beyond-visual-range air combat flight trajectories between opposing forces. The three-dimensional trajectories of both sides and the target trajectory decomposed feature points are shown below. Figure 3 As shown in the figure, the online automatic trajectory decomposition strategy proposed in this invention generates 34 trajectory decomposition feature points, dividing the target trajectory into 35 maneuvering elements. Combining the flight trajectories, it can be seen that both sides roughly experienced an air combat process from our aircraft pursuing the target, to mutual offensive confrontation, and then to the start of the second round of offensive confrontation. This process encompasses a complete beyond-visual-range air combat cycle.

[0063] Based on the target tactical intent hierarchical recognition model and algorithm designed according to the present invention, and combined with the recognition models of each layer obtained through offline training in the above experiments, the target tactical intent hierarchical recognition model and algorithm are used to identify the target tactical intent hierarchically. Figure 3 The target maneuvering elements, tactical maneuvers, and tactical intentions are identified online in a hierarchical manner, and the identification results are as follows: Figures 4 to 6 As shown. Comparison Figure 3 It can be seen that the types and timing of target maneuver elements, tactical maneuvers, and tactical intentions are accurately decomposed and identified, which is consistent with reality, further verifying the feasibility and effectiveness of the target tactical intention hierarchical identification method designed in this invention.

Claims

1. A method for target tactical intent hierarchical recognition in an air combat simulation environment, characterized in that, The method comprises the following steps: Step 1: extracting target track feature information and enemy-ally relative situation information, wherein the target track feature information comprises instantaneous feature information and cumulative feature information; The instantaneous characteristic information includes a track inclination angle θ t , a track inclination angle change rate , a track deviation angle change rate , and a height change rate The cumulative characteristic information includes a track inclination angle change amount C θ , a track deviation angle change amount C φ ; friendly situation information includes target azimuth q f , angle of entry q j and relative distance rate of change Step 2: finding target track decomposition feature points according to four conditions of target track deflection angle change trend change, target track deflection angle change accumulation exceeding 90°, target track inclination angle positive-negative change and target track inclination angle change trend change, and decomposing the target track into a series of continuous maneuver elements; Step 3: respectively designing target maneuver element state set, tactical maneuver state set and tactical intention state set in over-the-horizon air combat; Step 4: designing a cascade support vector machine multi-classification algorithm architecture for target tactical intention hierarchical recognition, and realizing target tactical intention hierarchical recognition, specifically comprising: Step 4-1: designing an optimal cascade support vector machine classification function; For a training sample set of size m {(x1,y1),(x2,y2),...,(x m ,y m Let the optimal classification function be} in, Let be the radial basis function kernel, and σ be the width parameter of the radial basis function kernel. b is the optimal solution for the Lagrange coefficients corresponding to the training samples. * For the corresponding optimal classification hyperplane displacement term, 0 ≤ α i ≤C, where C is a constant penalty factor; In the training stage of the cascade support vector machine, each support vector machine respectively performs single model training on the training samples of the target maneuver element, the tactical maneuver and the tactical intention, and each layer of recognition result corresponds to one of the feature information of the next layer of training samples; Step 4-2: designing target tactical intention hierarchical training samples; Target maneuver element training sample is target track instantaneous characteristic information θ t , Mapping of P to maneuver element number ME ; tactical maneuver training sample is designed according to tactical maneuvers composed of four-stage, three-stage and two-stage maneuver elements, that is, P ME , C φ and C θ to tactical maneuver number P TM mapping; tactical intent training sample is tactical maneuver number P TM and relative situation information q f , q j , to tactical intent number P TI mapping; The sample data is resampled by using an under-sampling method and is normalized by using the maximum value and the minimum value of each feature variable; Step 4-3: designing target tactical intention hierarchical recognition logic to realize target tactical intention hierarchical recognition; The tactical maneuver identification is segmentation and combination of multi-stage maneuver elements, wherein the priority of tactical maneuver identification of four-stage maneuver elements is higher than that of three-stage and two-stage, that is, first, the tactical maneuver identification of four-stage maneuver elements is performed, if it belongs to this class, the tactical maneuver number P is directly output TM , otherwise, it is identified as non-four-stage maneuver elements and the tactical maneuver identification of three-stage maneuver elements is performed, if it belongs to this class, the tactical maneuver number P is directly output TM , otherwise, it is identified as non-three-stage maneuver elements and the tactical maneuver identification of two-stage maneuver elements is performed, if it belongs to this class, the tactical maneuver number P is directly output TM , otherwise, the corresponding maneuver element number P is directly output ME ; three classes of tactical maneuvers are trained by separate support vector machine models, and form relatively independent identification modules; in the training process, the optimal C and sigma model parameter combination is found by cross-validation method.

2. The method of claim 1, wherein the method further comprises: determining a target tactical intent of the target aircraft based on the target aircraft's flight path and the target aircraft's flight path's change rate. The parameters in the target track feature information and the enemy-ally relative situation information are defined as follows: The instantaneous characteristic information in the target track information includes: track inclination Track inclination rate of change Track deflection rate of change And height rate of change Wherein Δx t = x t (t) - x t (t - h), Δy t = y t (t) - y t (t - h), and Δz t = z t (t) - z t (t - h) are the displacements of the target in the three-axis directions of the geographic coordinate system at time t, [x t (t), y t (t), z t (t)] is the position coordinate of the target in the geographic coordinate system at time t, and h is the simulation step length, The cumulative characteristic information in the target track information includes: a track inclination change amount C θ = |θ t (t end )-θ t (t start )| and a track deflection change amount C φ = |φ t (t end )-φ t (t start )| , C θ , C φ ∈ [0, π / 2], wherein t start and t end are the start and end time of the target maneuver element respectively; Relative situation information includes target azimuth. Target entry angle and relative distance change rate Where D x D y D z These are the projections of the relative distance D between the enemy and ourselves onto the three axes of the geographic coordinate system, v. ux v uy v uz and v tx v ty v tz These are the speeds of my machine, v. u and target velocity v t Projection along the three axes of the geographic coordinate system.

3. The method of claim 1, wherein, The target maneuver element state set, the tactical maneuver state set and the tactical intention state set are specifically as follows: The set of maneuver element states is S ME = [ME1, ME2,..., ME i ], i = 1, 2,... 17, respectively, representing the 17 maneuver elements of steady level flight, horizontal left turn, horizontal right turn, straight climb, straight descent, concave climb in the vertical plane, convex climb in the vertical plane, concave descent in the vertical plane, convex descent in the vertical plane, left turn concave climb, left turn convex climb, left turn concave descent, left turn convex descent, right turn concave climb, right turn convex climb, right turn concave descent, right turn convex descent. The set of tactical maneuver states is S TM = [TM1, TM2,..., TM15], j = 1, 2,... 15, respectively, represent steady level flight, horizontal left circle, horizontal right circle, horizontal S, jump S, dive S, snap-up, tactical dive, break-S, half Cuban eight, left break, right break, left combat turn, right combat turn, barrel roll maneuver 15 kinds of tactical maneuvers. j , j = 1, 2,... 15, respectively, represent steady level flight, horizontal left circle, horizontal right circle, horizontal S, jump S, dive S, snap-up, tactical dive, break-S, half Cuban eight, left break, right break, left combat turn, right combat turn, barrel roll maneuver 15 kinds of tactical maneuvers. The tactical intent state set is S TI =[TI1,TI2,...,TI k ], k = 1, 2, ... 5, representing five tactical intentions: attack, defense, evasion, transition in, and transition out.

4. The method of claim 1, wherein, The related parameters for finding the optimal C and σ model parameter combination by the cross-validation method are as follows: the target maneuver element recognition model parameter is C=4 and σ=0.25; the tactical maneuver recognition model parameters of the four-segment, three-segment and two-segment maneuver element classes are C=2 and σ=0.47, C=1.4 and σ=0.21, and C=1.6 and σ=0.45, respectively; and the target tactical intention recognition model parameter is C=4 and σ=0.28.

Images