Graph model-based simulation of attack posture recognition method for a group of opponents

Abstract

The application discloses a kind of simulation combat group attack situation identification method based on graph model, can according to the moving track of each unit in group, identify the attack situation of group to its strike target;Specific steps are as follows: S1, the trajectory of each unit is fitted using Bezier curve, the tangent direction at the end of fitting curve is regarded as moving direction, and a kind of quantization priority degree algorithm is designed, the identification of attack target and attack direction is realized;S2, the attack, angle, distance and homology relationship between each unit in group are established, which is abstracted as point edge topological structure;S3, the graph element structure of several typical attack situations is defined, and Ullmann algorithm is used for matching, to realize the identification of specific situation.The application realizes the novel, efficient and intuitive identification of simulation combat group attack situation, improves the efficiency of command personnel to perceive battlefield situation.

Description

Technical Field

[0001] This invention belongs to the field of data visualization technology, and in particular relates to a method for identifying the offensive posture of simulated combat groups based on graph models. Background Technology

[0002] Battlefield simulation scenarios are virtual environments created to mimic the situational evolution of relevant objects in a real battlefield. Commanders frequently use these scenarios to analyze the battlefield situation. To better formulate deployments based on the battlefield situation, commanders often need to analyze the offensive posture of group units against targets within the simulation scenario. However, in simulation scenarios simulating combat against other targets, commanders often manually observe the movement of enemy groups to identify their offensive posture. But manual situational awareness is not only extremely inefficient but also lacks accuracy.

[0003] On the other hand, there is currently no complete and efficient method for automatically identifying the offensive posture of group units in battlefield simulation scenarios, thus failing to assist commanders in efficiently understanding the battlefield situation. Furthermore, existing graph model methods, often used for relationship analysis, are mostly applied to recommendation systems, social networks, and search systems, and are rarely used in the military battlefield. Therefore, there is an urgent need for a graph model-based algorithm for identifying the offensive posture of simulated combat groups. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying simulated combat group offensive postures based on graph models, so as to automatically identify group offensive postures on the battlefield, use graph models to construct reasonable relationships between various elements on the battlefield, improve the efficiency and accuracy of posture identification, and solve the problems of low efficiency and low accuracy of posture identification in traditional battlefields, which mostly rely on manual methods.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is a method for identifying the simulated attack posture of a combat group based on a graph model, comprising the following steps:

[0006] S1. Identify attack targets of group units based on trajectory prediction;

[0007] S2. Abstract the battlefield data and construct an abstract graphical model of the battlefield simulation scenario based on the abstracted battlefield data;

[0008] S3. Define the corresponding primitive structure based on the simulated battlefield offensive posture, match the defined primitives based on the subgraph isomorphism algorithm, and obtain the corresponding offensive posture based on the subgraph matching result.

[0009] Furthermore, S1 specifically includes the following steps:

[0010] S11. Use Bézier curves to fit the trajectory data of enemy group units in the simulated battle over a period of time, and use the tangent direction at the end of the fitted curve as the predicted direction of the group trajectory in future time slices.

[0011] S12. Identify the bearing points of our targets; the bearing points of our targets include the front, flanks and rear.

[0012] S13. Design a quantification priority algorithm and identify enemy targets attacking our side based on trajectory prediction results.

[0013] Furthermore, the specific process of S11 is as follows: based on the number N of group units in the simulated combat group over a period of time [T]... start T end Movement trajectory data within ]{[D 1start D 1start+1 D 1start+2 , ..., D 1end ], [D 2start D 2start+1 D 2start+2 , ..., D 2end ], [D 3start D 3start+1 D 3start+2 , ..., D 3end ],...,[D Nstart D Nstart+1 D Nstart+2 , ..., D Nend ]}, thus obtaining the mean trajectory data D1~D for each group. N For each group's movement trajectory data, a Bézier curve is fitted to obtain the fitted curve B1 to B2 for each group. N In [T] start T end Trajectory within time {[D] 1Fitstart D 1Fitstart+1 D 1Fitstart+2 , ..., D 1Fitend ], [D 2Fitstart D 2Fitstart+1 D 2Fitstart+2 , ..., D 2Fitend ], [D 3Fitstart D 3Fitstart+1 D 3Fitstart+2 , ..., D 3Fitend ],...,[D NFitstart D NFitstart+1 D NFitstart+2 , ..., D NFitend]}, where if the number of time slices is n, then its order is n-1; Bézier curve B i,n The calculation method for (t) is as follows:

[0014]

[0015]

[0016] Where P(t) represents the coordinates of any point on the fitted curve obtained by interpolation based on t∈[0,1]; t represents an intermediate parameter in the range [0,1]. For the combination formula, P i Representing the i-th control point of the Bézier curve, the trajectory data of a fitted population unit curve B1 is obtained [D]. 1Fitstart D 1Fitstart+1 D 1Fitstart+2 , ..., D 1Fitend ], select the last two fitting points [D 1Fitend-1 D 1Fitend The direction of the movement and the point [D] 1Fitend Draw a tangent line to B1, and use the direction of this tangent line as the predicted trajectory direction of the fitted curve B1:

[0017]

[0018]

[0019]

[0020] In the above formula, a, b, and c are parameters for predicting the direction ax + by + c = 0, used to predict the movement direction of five enemy groups. These are the last two fitting points [D] 1Fit58 D 1Fit59 The coordinates of ].

[0021] Furthermore, the specific process of S12 is as follows: First, calculate the latitude and longitude range of our target and determine a rectangular range. Second, take the center points of the four sides of the rectangle and mark them as the four azimuth points of our target. The azimuth points mainly include three types: front, flank, and rear. Among them, for our static target, the direction facing the enemy is taken as the front azimuth point, and for our dynamic target, the direction of its movement trend is taken as the front azimuth point. After marking the front azimuth points according to the combat action type of our different group targets, the other two sides are marked as flank azimuth points, and the opposite sides are marked as rear azimuth points.

[0022] Furthermore, the specific process of S13 is as follows:

[0023] S131. Let the enemy cluster currently being analyzed be A. For azimuth point P, calculate the distance d between points P and A, and the angle θ between the direction of the line connecting AP and the direction of the predicted ray. The priority of azimuth point P to A is:

[0024]

[0025] Where k1 and k2 are coefficients. Since θ and d have different units of measurement, empirical parameters m and n are introduced to make them dimensionless.

[0026] S132. For analysis object A, calculate the priority of each location {P1, P2, P3, P4} relative to A, denoted as {Ω1, Ω2, Ω3, Ω4}; the minimum value Ω among {Ω1, Ω2, Ω3, Ω4}. k :

[0027] Ω k =min{Ω1,Ω2,Ω3,Ω4}

[0028] Its corresponding location P k As the target of A's attack, the attack target is identified; the priority of the four directional points of each of the three friendly targets relative to the five enemy groups is determined, and finally the attack directional points of each enemy group are determined.

[0029] Furthermore, the process of abstracting battlefield data in S2 is as follows: First, draw the circumscribed rectangle corresponding to our target, take the corresponding side of the circumscribed rectangle as the front, its adjacent side as the flank, and the opposite side as the back, and take the geographical coordinates of the midpoint of each side as the azimuth point of the corresponding direction; the position of our target is represented by its 4 azimuth points, which are abstracted as 4 nodes in a heterogeneous network, while the enemy group is abstracted as a node in a heterogeneous network, and the key positions of the enemy group and our target are abstracted in the evolution of the battlefield environment to obtain key nodes; finally, a two-dimensional matrix is ​​used to store the point-edge relationship, and an array is used to record the category of each node; among them, edges need to be connected between related nodes.

[0030] Furthermore, in S2, the graph model is constructed through the following four types of node relationships:

[0031] Offensive Relationship: Establish a relationship based on a node in a certain direction of a friendly target that the enemy group nodes want to attack, thereby constructing an offensive relationship between the group target and the offensive target, and establishing an enemy-friendly relationship connection.

[0032] Angular Relationship: Based on the calibration results of our target's azimuth nodes, an angular relationship can be established between the four azimuth nodes of our target, and a connection edge can be established between our own azimuth nodes.

[0033] Distance Relationship: When the distance between different enemy group nodes in a certain time slice in the past is less than the distance relationship threshold, the group constitutes a distance relationship, and an enemy self-relationship edge is established;

[0034] Homologous relationship: When the distance between different enemy group nodes in a certain time slice in the past is less than the homologous relationship threshold, the group constitutes a homologous relationship and establishes an enemy self-relationship edge.

[0035] Furthermore, the distance relationship threshold is a time period T < 60s and a distance D < 200km; the homology relationship threshold is a time period T < 60s and a distance D < 10km.

[0036] Furthermore, the process of defining the corresponding primitive structure in S3 involves abstracting the offensive postures on the battlefield, including pincer attack postures, flank attack postures, and multi-point attack postures, into point-edge relationships, as follows:

[0037] The flank attack has 4 friendly azimuth nodes and 1 enemy node. The 4 friendly azimuth nodes are in an angular relationship with each other, and the relationship with the enemy node is an offensive relationship.

[0038] The multi-point attack has 4 friendly directional nodes and 2 enemy nodes. The 4 friendly directional nodes are related by angle, and the friendly nodes are related by attack. The 2 enemy nodes are related by distance.

[0039] The pincer attack has 4 friendly azimuth nodes and 3 enemy nodes. The 4 friendly azimuth nodes are related by angle and by attack with the 2 enemy nodes. The 3 enemy nodes are related by origin and by distance.

[0040] Based on the point-edge relationships of the three offensive postures, the matrix relationship representation of the primitive structure for each offensive posture is defined as follows:

[0041] Flanking point set:

[0042]

[0043] Wing side set:

[0044]

[0045] Multi-point collection:

[0046]

[0047] Multi-point edge set:

[0048]

[0049] Pincer point set:

[0050]

[0051] Pincer edge set:

[0052]

[0053] Furthermore, in S3, the defined attack posture is matched based on the subgraph isomorphism algorithm, and the corresponding attack posture is obtained based on the subgraph matching result as follows:

[0054] S321. During matching, based on the complexity of the primitive structure, primitives attacking from the flanks are identified first, and it is determined whether the enemy nodes satisfy the distance relationship. If they do, it is determined to be a multi-point attack; otherwise, it is a flank attack. When a multi-point attack involves two or more enemy nodes attacking the same point on our side, it is determined whether the distance relationship between each pair of enemy nodes is satisfied. If they do, it is determined to be a multi-point attack. For pincer attacks, if there is a homogeneous relationship in the structural relationship, it has a historical time correlation. When two enemy nodes attack two of our flank nodes or our front and back nodes, forming a pincer attack, it is determined to be a pincer attack.

[0055] S322. Based on the offensive posture obtained in S321, the redundant information and other abnormal results obtained from repeated matching need to be deduplicated and cleaned to obtain the final offensive posture of the enemy group.

[0056] The beneficial effects of this invention are

[0057] The graph model-based simulated combat group offensive posture recognition algorithm described in this invention automatically identifies the offensive posture of battlefield group units based on their trends. It can assist battlefield commanders in efficiently, quickly, clearly, and conveniently grasping the core forces of all parties on the battlefield and the direction of their main offensive postures. It solves the problems of low efficiency, low accuracy, time-consuming and labor-intensive real-time identification of battlefield postures by manual methods. At the same time, by using graph model methods dedicated to solving relationship analysis problems, the task of identifying group unit postures can be completed more accurately and efficiently. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart illustrating the steps of a simulated battle group offensive posture recognition algorithm provided in an embodiment of the present invention.

[0060] Figure 2 A schematic diagram of the fitted trajectories of multiple simulated combat groups;

[0061] Figure 3 A schematic diagram illustrating the trajectory direction prediction of multiple simulated combat group units;

[0062] Figure 4 A schematic diagram showing the azimuth points for identifying the target.

[0063] Figure 5 A schematic diagram of an algorithm for quantifying strike priority;

[0064] Figure 6 A schematic diagram illustrating the offensive relationship between a group of targets and an attacking target;

[0065] Figure 7 A schematic diagram for establishing angular relationships between the four directional nodes of the target;

[0066] Figure 8 A diagram illustrating the distance relationships between group targets;

[0067] Figure 9 A diagram illustrating the homology among group objectives;

[0068] Figure 10 A schematic diagram of an abstract graphical model for battlefield simulation scenarios;

[0069] Figure 11 A schematic diagram of the primitive structure for three typical offensive postures:

[0070] Figure 12 This is a schematic diagram for matching primitive structures in an abstract graph model. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] like Figure 1 This invention provides a method for identifying the simulated attack posture of a combat group based on a graph model. The specific implementation steps are as follows:

[0073] Example 1: Target Identification in Group Attacks

[0074] Based on the trajectory data of enemy group units in the simulated battle over a period of time, Bézier curves are used for fitting. The tangent direction at the end of the fitted curve is taken as the trajectory direction of the group in the future time slice and the trajectory is predicted accordingly. The azimuth points (front, flank, and rear) of our targets are marked, and a quantification priority algorithm is designed to identify enemy attacking our targets based on the trajectory prediction results.

[0075] (1) This simulated battle includes 5 enemy groups and 3 attack targets (friendly targets), with each group containing 4 to 6 group units. First, based on the time period [T] of each group unit... start T end The original movement trajectory data within ]{[D 1start D 1start+1 D 1start+2 , ..., D 1end ], [D 2start D 2start+1 D 2start+2 , ..., D 2end ],...,[D 6start D 6start+1 D 6start+2 , ..., D 6end The time period here can be dynamically selected. The longer the selected time period, the more accurate the subsequent fitting and prediction direction will be. In this embodiment, the time period [0s, 60s] is selected. The mean trajectory data D1 to D5 of each group are further calculated, and Bézier curves are fitted to them. Then, the fitted curves B1 to B5 are sampled at equal intervals to obtain the fitted curve trajectory {[D] of each group in the time period [0s, 60s]. 1Fitstart D 1Fitstart+1 D 1Fitstart+2 , ..., D 1Fitend ], [D 2Fitstart D 2Fitstart+1 D 2Fitstart+2 , ..., D 2Fitend ],...,[D 5Fitstart D 5Fitstart+1 D 5Fitstart+2 , ..., D 5Fitend ]},like Figure 2 As shown. Since time can be dynamically selected, the order of the Bézier curve is determined by the data from the time slices. If the number of time slices is n, then its order is b-1, that is, the order B of the Bézier curve changes with time. i,b (t) is dynamically changing, and its specific calculation method is as follows:

[0076]

[0077]

[0078] In the above formula, P(t) represents the coordinates of any point on the fitted curve obtained by interpolation based on t∈[0,1]; n is the order of the Bézier curve; This is a combination formula, which represents the number of combinations of choosing i elements from n elements; P i Let represent the i-th control point of the Bézier curve, where i ranges from 0 to 59, and t represents an intermediate parameter within the range [0,1]. This yields the trajectory data of a fitted group unit curve B1 [D]. 1Fit1 D 1Fit2 D 1Fit3 , ..., D 1Fit59 ], select the last two fitting points [D 1Fit58 D 1Fit59 The direction of the movement and the point [D] 1Fit59 Draw a tangent line to B1, and use the direction of this tangent line as the predicted trajectory direction of the fitted curve B1. Figure 3 As shown, the specific calculation method is as follows:

[0079]

[0080]

[0081]

[0082] In the above formula, a, b, and c are parameters for predicting the direction ax + by + c = 0, used to predict the movement direction of five enemy groups. These are the last two fitting points [D] 1Fit58 D 1Fit59 The coordinates of ].

[0083] (2) Figure 4 As shown, the azimuth points of three friendly targets are marked: front, flank, and rear. First, the latitude and longitude range of the friendly targets is calculated, defining a rectangular area. Then, the center points of the four sides of the rectangle are used to mark the four azimuth points of the friendly targets. These azimuth points are mainly divided into three categories: front, flank, and rear. Different methods are used to mark the azimuth points according to different combat action types of the friendly targets: for static friendly targets, the direction facing the enemy is used as the front azimuth point; for dynamic friendly targets, the direction of their movement is used as the front azimuth point. After marking the front azimuth points according to the combat action types of different groups of friendly targets, the remaining two sides are marked as flanks, and the opposite side is marked as the rear, thus marking all the azimuth points of the three friendly targets.

[0084] (3) Figure 5 Based on trajectory prediction results, enemy targets attacking our side are identified; an algorithm for quantifying priority is designed, specifically including the following steps:

[0085] S131. Let the enemy cluster currently being analyzed be A. For azimuth point P, calculate the distance d between points P and A, and the angle θ between the direction of the line connecting AP and the direction of the predicted ray. The priority of azimuth point P over A is:

[0086]

[0087] Where k1 and k2 are coefficients. Since θ and d have different units of measurement, we introduce empirical parameters m and n to make them dimensionless. Specifically, n = 360° (unit: degree) and m = 111km (unit: kilometer).

[0088] S132. For analysis object A, calculate the priority of each azimuth point {P1, P2, P3, P4} relative to A {Ω1, Ω2, Ω3, Ω4}. For azimuth point P... i Its priority relative to A Ω i The smaller the value, the more likely it is to be attacked by A. Therefore, find the minimum value Ω in {Ω1,Ω2,Ω3,Ω4}. k :

[0089] Ω k =min{Ω1,Ω2,Ω3,Ω4}

[0090] Its corresponding location P k As the target of A's attack, the attack target is identified. Using this method, the priority of three friendly targets and the four azimuth points of each target relative to five enemy groups are determined, thereby determining the attack azimuth points of each enemy group.

[0091] Example 2: Graph Model Construction

[0092] (1) The battlefield data is abstracted, and the point-edge relationships in the data are extracted to construct a heterogeneous network on the battlefield. First, the circumscribed rectangle corresponding to our target is drawn. According to the combat deployment, the corresponding edge of the circumscribed rectangle is taken as the front, its adjacent edge as the flank, and the opposite edge as the back. The geographical coordinates of the midpoint of each edge are taken as the azimuth points of the corresponding position. The position of our target is represented by its 4 azimuth points, which are further abstracted into 4 nodes in the heterogeneous network. These 4 nodes are divided into front, flank, and back. Thus, the 12 azimuth points of 3 our targets are abstracted into 12 nodes in the heterogeneous network, while the enemy group is abstracted into 1 node in the heterogeneous network. Thus, the 5 enemy groups are abstracted into 5 nodes in the heterogeneous network. In the evolution of the battlefield environment, the key positions of the enemy group targets and our targets are abstracted to obtain key nodes, such as homogeneous nodes representing homogeneous relationships. In addition, edges need to be connected between related nodes. Finally, a two-dimensional matrix is ​​used to store the point-edge relationships, and an array is used to record the category of each node.

[0093] (2) Based on the semantic information in the battlefield simulation scenario, the graph model is constructed mainly through the following four types of node relationships: attack relationship, angle relationship, distance relationship, and homology relationship:

[0094] Offensive relationship: such as Figure 6 As shown, in the attack target identification step, the three friendly targets to be attacked by the five enemy groups are identified. That is, the relationship is established based on the location node of the friendly target to be attacked by the enemy group node, thereby constructing the attack relationship between the group target and the attack target, which belongs to the enemy-friendly relationship, and then establishing the connection edge.

[0095] Angular relationship: such as Figure 7 As shown, based on the calibration results of the three azimuth nodes of our target, an angular relationship can be established between the four azimuth nodes (front, flank, and back) of each target, which is a relationship within our own territory, and then a connection can be established.

[0096] Distance relationship: such as Figure 8 As shown, when two or more nodes of five enemy groups are less than a certain threshold in a certain time slice in the past, taking air warfare as an example (duration 10-20 minutes), the recommended time period T < 60 seconds and distance D < 200 km, these groups are considered to have formed a distance relationship, which belongs to the enemy's own relationship, and then a connection is established.

[0097] Homology: such as Figure 9 As shown, the definition of a homologous relationship is similar to that of a distance relationship, except that the threshold set for homologous relationship nodes is much smaller than that set for distance relationship nodes. They are considered to originate from the same source and disperse from a single homologous node. If two or more nodes of five enemy groups have a distance between them that is less than a certain threshold in a past time slice, and homologous nodes exist, taking air combat as an example (often 10-20 minutes), the recommended time slice T < 60 seconds and distance D < 10 km are considered to originate from the same source. In this case, these groups are considered to have a homologous relationship, belonging to the enemy's own relationship, and thus a connection is established.

[0098] Based on the four types of node relationships defined above, an abstract graph model for a battlefield simulation scenario is constructed, yielding various relationships between five enemy groups and three friendly targets, such as... Figure 10 As shown.

[0099] Example 3: Defining and Identifying Graphic Elements

[0100] Based on the specific meaning of each enemy group's offensive posture (pincer, flank, multi-point), the corresponding primitive structure is defined. Then, the defined primitives are matched using the Ullmann subgraph matching algorithm. The corresponding posture is obtained based on the subgraph matching results, thus completing the posture identification.

[0101] (1) It is necessary to define corresponding primitive structures based on the specific meaning of each offensive posture of the enemy, such as Figure 11 As shown. The main offensive postures on the battlefield include pincer attack, flank attack, and multi-point attack. These can be abstracted into point-edge relationships, and their specific meanings and characteristics are as follows:

[0102] The flank attack has 4 friendly azimuth nodes and 1 enemy node. The 4 friendly azimuth nodes are in an angular relationship with each other, and the relationship with the enemy node is an offensive relationship.

[0103] The multi-point attack has 4 friendly directional nodes and 2 enemy nodes. The 4 friendly directional nodes are related by angle, and the friendly nodes are related by attack. The 2 enemy nodes are related by distance.

[0104] The pincer attack has 4 friendly directional nodes and 3 enemy nodes. The 4 friendly directional nodes are related by angle and by attack with the 2 enemy nodes. The 3 enemy nodes are related by origin and by distance.

[0105] Based on the point-edge relationships of the three offensive postures, the matrix relationship representation of the primitive structure for each offensive posture is defined as follows:

[0106] Flanking point set:

[0107]

[0108] Wing side set:

[0109]

[0110] Multi-point collection:

[0111]

[0112] Multi-point edge set:

[0113] Pincer point set:

[0114]

[0115] Pincer edge set:

[0116]

[0117] The above content will serve as the basis for identifying specific structures in the heterogeneous network established by various relationships between 5 enemy groups and 3 friendly targets.

[0118] (2) Use the Ullmann subgraph matching algorithm to match the defined primitives. During matching, based on the complexity of the primitive structure, primitives of flank attacks are identified first. Since flank attacks and multi-point attacks are similar in terms of primitive structure, it is necessary to further determine whether the enemy nodes meet the distance relationship. That is, for air combat lasting 10-20 minutes, if the time period T < 60 seconds and the distance D < 200 km, it is judged as a multi-point attack; otherwise, it is a flank attack. Furthermore, for multi-point attacks, there may be two or more enemy nodes attacking the same point on our side. In this case, it is necessary to determine whether the distance relationship between each pair of enemy nodes is met. If it is met, it is judged as a multi-point attack. For pincer attacks, if there is a homogeneous relationship in the structural relationship, that is, for air combat lasting 10-20 minutes, if the time period T < 60 seconds and the distance D < 10 km, it is considered to have a historical time correlation. If two enemy nodes attack two of our flank nodes or our front and rear nodes, forming a pincer attack, it is judged as a pincer attack.

[0119] Based on the above, we obtain the offensive posture of 5 enemy groups against 3 friendly targets. Finally, based on the matching results, we need to deduplicate and clean up redundant information and other abnormal results to obtain the final offensive posture of the enemy groups. Figure 12 As shown.

[0120] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0121] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A method for identifying simulated combat group offensive postures based on graph models, characterized in that, Includes the following steps: S1. Identify attack targets of group units based on trajectory prediction; S2. Abstract the battlefield data and construct an abstract graphical model of the battlefield simulation scenario based on the abstracted battlefield data; S3. Define the corresponding primitive structure according to the simulated battlefield offensive situation, match the defined primitives based on the subgraph isomorphism algorithm, and obtain the corresponding offensive situation based on the subgraph matching results. S1 specifically includes the following steps: S11. Use Bézier curves to fit the trajectory data of enemy group units in the simulated battle over a period of time, and use the tangent direction at the end of the fitted curve as the predicted direction of the group trajectory in future time slices. S12. Determine the bearing points of our target; the bearing points of our target include the front, flank, and rear. S13. Design a quantification priority algorithm and identify enemy attacking targets based on trajectory prediction results; The specific process of S11 is as follows: based on the number of group units N in the simulated combat group over a period of time [T]... start T end Movement trajectory data within ]{[D 1start D 1start+1 D 1start+2 , ..., D 1end ], [D 2start D 2start+1 D 2start+2 , ..., D 2end ], [D 3start D 3start+1 D 3start+2 , ..., D 3end ],...,[D Nstart D Nstart+1 D Nstart+2 , ..., D Nend ]}, thus obtaining the mean trajectory data D1~D for each group. N For each group's movement trajectory data, a Bézier curve is fitted to obtain the fitted curves B1~B1 for each group. N In [T] start T end Trajectory within time {[D] 1Fitstart D 1Fitstart+1 D 1Fitstart+2 , ..., D 1Fitend ], [D 2Fitstart D 2Fitstart+1 D 2Fitstart+2 , ..., D 2Fitend ], [D 3Fitstart D 3Fitstart+1 D 3Fitstart+2 , ..., D 3Fitend ],...,[D NFitstart D NFitstart+1 D NFitstart+2 , ..., D NFitend ]}, where the number of time slices is Then its order is Bézier curve The calculation method is as follows: in, Indicates the fitted curve based on The coordinates of any point obtained through interpolation; This represents an intermediate parameter with a value range of [0,1]. For the combination formula, The first Bézier curve represents the second curve. Using control points, the trajectory data of a fitted population unit curve B1 is obtained [D] 1Fitstart D 1Fitstart+1 D 1Fitstart+2 , ..., D 1Fitend ], select the last two fitting points [D 1Fitend-1 D 1Fitend The direction of the movement trend and passing through point [D] 1Fitend Draw a tangent line to B1, and use the direction of this tangent line as the predicted trajectory direction of the fitted curve B1: In the above formula, a, b, and c are parameters for predicting the direction ax + by + c = 0, used to predict the movement direction of five enemy groups. These are the last two fitting points [D] 1Fit58 D 1Fit59 The coordinates of ].

2. The method for identifying simulated combat group offensive postures based on graph models according to claim 1, characterized in that, The specific process of S12 is as follows: First, calculate the latitude and longitude range of our target and determine a rectangular range. Then, take the center points of the four sides of the rectangle and mark them as the four azimuth points of our target. The azimuth points include three types: front, flank, and rear. For our static target, the direction facing the enemy is taken as the front azimuth point. For our dynamic target, the direction of its movement trend is taken as the front azimuth point. After marking the front azimuth points according to the combat action type of our different groups of targets, the other two sides are marked as flank azimuth points, and the opposite sides are marked as rear azimuth points.

3. The method for identifying simulated combat group offensive postures based on graph models according to claim 1, characterized in that, The specific process of S13 is as follows: S131. Let the enemy cluster currently being analyzed be A. For azimuth point P, calculate the distance d between points P and A, and the angle θ between the direction of the line connecting AP and the direction of the predicted ray. The priority of azimuth point P to A is: in , For the coefficient, since and Since they have different units of measurement, empirical parameters are introduced. To make it dimensionless; S132. For analysis object A, calculate the location of each position. Priority relative to A ; minimum value : Its corresponding location As the target of A's attack, the attack target is identified; the priority of the four directional points of each of the three friendly targets relative to the five enemy groups is determined, and finally the attack directional points of each enemy group are determined.

4. The method for identifying simulated combat group offensive postures based on graph models according to claim 1, characterized in that, The process of abstracting battlefield data in S2 is as follows: First, draw the circumscribed rectangle corresponding to our target. Take the corresponding side of the circumscribed rectangle as the front, its adjacent side as the flank, and the opposite side as the back. Take the geographical coordinates of the midpoint of each side as the azimuth point of the corresponding direction. The position of our target is represented by its four azimuth points, which are abstracted as four nodes in a heterogeneous network. The enemy group is abstracted as a node in a heterogeneous network. In the evolution of the battlefield environment, the key positions of the enemy group and our target are abstracted to obtain key nodes. Finally, a two-dimensional matrix is ​​used to store the point-edge relationship, and an array is used to record the category of each node. Among them, the nodes that are related need to be connected by edges.

5. The method for identifying simulated combat group offensive postures based on graph models according to claim 1, characterized in that, The graph model in S2 is constructed using the following four types of node relationships: Offensive Relationship: Establish a relationship based on a node in a certain direction of a friendly target that the enemy group nodes want to attack, thereby constructing an offensive relationship between the group target and the offensive target, and establishing an enemy-friendly relationship connection. Angular Relationship: Based on the calibration results of our target's azimuth nodes, an angular relationship can be established between the four azimuth nodes of our target, and a connection edge can be established between our own azimuth nodes. Distance Relationship: When the distance between different enemy group nodes in a certain time slice in the past is less than the distance relationship threshold, the group constitutes a distance relationship, and an enemy self-relationship edge is established; Homologous relationship: When the distance between different enemy group nodes in a certain time slice in the past is less than the homologous relationship threshold, the group constitutes a homologous relationship and establishes an enemy self-relationship edge.

6. The method for identifying simulated combat group offensive postures based on graph models according to claim 5, characterized in that, The distance relationship threshold is when the time period T < 60s and the distance D < 200km; the homology relationship threshold is when the time period T < 60s and the distance D < 10km.

7. The method for identifying simulated combat group offensive postures based on graph models according to claim 1, characterized in that, The process of defining the corresponding primitive structure in S3 is to abstract the offensive postures on the battlefield, including pincer attack postures, flank attack postures, and multi-point attack postures, into point-edge relationships, as follows: The flank attack has 4 friendly azimuth nodes and 1 enemy node. The azimuth nodes of the 4 friendly targets are related by angle, and the relationship between them and the enemy node is one of attack. The multi-point attack has 4 friendly azimuth nodes and 2 enemy nodes. The azimuth nodes of the 4 friendly targets are related by angle, and the azimuth nodes of the 4 friendly targets are related by attack. The azimuth nodes of the 2 enemy nodes are related by distance. The pincer attack has 4 friendly azimuth nodes and 3 enemy nodes. The azimuth points of the 4 friendly targets are related by angle and by attack with the 2 enemy nodes. Among them, the 3 enemy nodes are related by origin and by distance. Based on the point-edge relationships of the three offensive postures, the matrix relationship representation of the primitive structure for each offensive posture is defined as follows: Flanking point set: Wing side set: Multi-point collection: Multi-point edge set: Pincer point set: Pincer edge set: 。 8. The method for identifying simulated combat group offensive postures based on graph models according to claim 1, characterized in that, In S3, the defined attack posture is matched based on the subgraph isomorphism algorithm, and the corresponding attack posture is obtained based on the subgraph matching result as follows: S321. During matching, based on the complexity of the primitive structure, primitives attacking from the flanks are identified first, and it is determined whether the enemy nodes satisfy the distance relationship. If they do, it is determined to be a multi-point attack; otherwise, it is a flank attack. When a multi-point attack involves two or more enemy nodes attacking the same point on our side, it is determined whether the distance relationship between each pair of enemy nodes is satisfied. If they do, it is determined to be a multi-point attack. For pincer attacks, if there is a homogeneous relationship in the structural relationship, it has a historical time correlation. When two enemy nodes attack two of our flank nodes or our front and back nodes, forming a pincer attack, it is determined to be a pincer attack. S322. Based on the offensive posture obtained in S321, the redundant information and other abnormal results obtained from repeated matching need to be deduplicated and cleaned to obtain the final offensive posture of the enemy group.

Images