A battlefield division and attraction map-based position prediction method and system for a war game operator

Abstract

The present application relates to a kind of battlefield division and attraction map-based position prediction method and system of chess operator, belong to intelligent war game deduction technical field.The method includes the following steps: S1, using the method based on time perception and battlefield division carries out war game information processing and feature extraction;S2, establish the position prediction model based on attraction map and operator relationship diagram: using the way of attraction map describes the influence of global situation information including topography to chess operator, using the way of operator relationship diagram represents the influence of mutual action between operators, and carries out feature fusion;S3, in combination with time series prediction carries out deduction to combat, to complete the identification of target operator subsequent intention.The present application effectively improves the problem of insufficient feature extraction capability in the prior art in the field of war game, helps to improve the utilization efficiency of subsequent related research on war game information, provides a new idea and reference for data mining in the field of war game.

Description

Technical Field

[0001] This invention belongs to the field of intelligent wargaming technology, particularly the field of wargaming situation estimation and battlefield situation data processing technology, and relates to a method and system for predicting the position of wargaming operators based on battlefield division and attraction maps. Background Technology

[0002] Modern wargames evolved from Prussian war games and are a type of real-time strategy game. As a simplified simulation of war, it allows two or more players to participate. Each player acts as the commander of a faction, directing combat units such as tanks, armored vehicles, and infantry. Using game theory, probability theory, and other scientific methods, players simulate and deduce battles against the opposing faction within the rules of wargames until the game ends. The winner is determined by the types and numbers of remaining combat units and the battle score.

[0003] As a typical real-time strategy (RTS) game, modern wargaming has been widely used in the military and military academies, and related competitions are being held continuously. Its real-time and strategic nature places high demands on commanders, including but not limited to battlefield analysis capabilities, strategy selection and implementation capabilities, and adaptability. Developing strategies and deducing the enemy's intentions have become crucial to victory, making wargaming intention recognition a hot topic of research in this field.

[0004] Intent recognition is a process of interpreting the current situation, identifying enemy intentions, and inferring their tactical plans based on the combat scenario and the strategic knowledge of domain experts. It is also a prerequisite and foundation for intelligent decision-making.

[0005] Currently, related research mainly involves four aspects in terms of research methods: template matching, expert systems, Bayesian networks, and neural networks. Generally speaking, template matching follows the objective laws of intent recognition and is easy to apply; however, building a template library depends on prior knowledge obtained from domain experts, making its objectivity and reliability difficult to verify. While expert systems possess powerful knowledge representation and reasoning capabilities, they lack certain fault tolerance and learning abilities. Bayesian networks have attracted widespread attention due to their powerful probabilistic reasoning capabilities and can be used to solve uncertain reasoning problems such as intent recognition. Yin Weibin et al. used Bayesian networks to simulate the combat intentions of adversaries' surface ships. Ramirez et al. modeled the decision-making process of target units within the framework of Partially Observable Markov Decision Processes (POMDPs). However, Bayesian networks still face difficulties in determining prior probabilities and conditional probabilities for events at each node.

[0006] Neural networks, due to their powerful adaptive and autonomous learning capabilities, have been introduced into the field of intent recognition. Researchers, with the help of classic models such as CNNs and RNNs, as well as attention mechanisms, are able to handle highly complex situations. For example, Zhou and Yao proposed a novel deep neural network for combat intent recognition. This network improves convergence speed and prevents getting trapped in local optima by introducing the Rectified Linear Unit (ReLU) activation function and the Adaptive Moment Estimation (Adam) optimization algorithm. Suyaun et al. designed an air target intent model based on an Adaptive Neural Fuzzy System (ANFIS). This model achieves effective classification of unknown air target intents through the organic integration of fuzzy logic and neural network methods. Given the excellent performance of LSTM in predicting sequences, Zhang and Xue et al. used it to predict enemy operator trajectory information in wargames and achieved good results.

[0007] The existing ideas and methods of wargame intent recognition technology mainly include the following categories:

[0008] (1) By improving traditional path planning methods, we can analyze the path selection tendencies in wargames and thus achieve intent recognition. For example, in "A Path Planning Method for Hexagonal Grid Maps in Wargames Based on AI Algorithms," Zhu Liang et al. proposed a hexagonal grid positioning method and a method for calculating the effective distance between two cells by comparing four-corner grid maps. Then, considering the movement consumption problem in complex wargame maps, they improved traditional path planning methods to find the optimal path in complex wargames. This is used to explore the combat intent of wargame operators.

[0009] (2) Using clustering for trajectory data analysis. For example, You Xiong et al. proposed in "A Trajectory Data Analysis Method Applied to Land Warfare Wargaming" that the trajectory information of combat units is obtained, and the trajectory of each combat unit is clustered using a clustering algorithm to obtain the trajectory clustering information of each combat unit. Then, the hotspot areas of each combat unit are further determined. The combat units are then grouped, and the hotspot areas of each combat unit in the group are weighted and superimposed to form the hotspot area of ​​the group, thereby analyzing the combat intent.

[0010] (3) By using neural network models, the global map information and operator information of wargames are analyzed to achieve the purpose of intent recognition. A typical example is the "Transformer-based wargame target intent recognition method" proposed by Ye Lei et al., which uses feature extraction methods to initially obtain map information and operator features, and uses a convolutional layer to refine these features, and then uses Transformer for prediction. However, it does not fully analyze the battlefield situation, so the refined features cannot fully represent the current situation of the game.

[0011] However, although the above methods each have their own characteristics, their expressive power in the field of wargaming is limited to a certain extent, mainly facing the following problems: 1) Insufficient feature extraction capability: Since wargaming data is unstructured data that changes over time, and wargaming games are a typical scenario with incomplete, imperfect, and incomplete information, the two sides have serious information gaps, which can easily lead to unknown information sequence length, scattered information distribution, and low information utilization. Most machine learning models can only process structured data, and deep learning models have certain requirements for the sequence length and data distribution of data. Most existing technologies based on these methods have insufficient feature extraction capability. 2) Challenges brought by wargaming game features: (1) The relationships between operators in wargaming include whether they can see each other, whether they can be hit, whether they can be guided to fire, and the level of damage caused. Many technologies do not consider this level of information. (2) When different operators move on the map, different terrains have different strategic significance for them, driving them to move and deploy towards "strategic high ground" with certain strategies. This situational information should be fully considered. (3) The mission of the operators differs at different stages of the game, and the strategic significance of operators changes over time depending on the terrain. Many studies have not adequately elaborated on this aspect. (4) The strategic fog has a profound impact on wargames, representing the visibility of some information. Breaking through the fog is crucial to winning the game. However, the uncertainty of the strategic fog limits the application of many modern technologies in wargames.

[0012] In summary, wargaming presents challenges such as complex information types, frequent information changes, fragmented information structures, and low information value density. These limitations restrict the application of various mature technologies in wargaming.

[0013] Throughout modern warfare, most typical battles are phased and possess certain regional characteristics. Throughout history, numerous battles have been won by leveraging favorable timing, terrain, and popular support. This invention does not rely on traditional path planning methods or clustering concepts, but instead focuses on the overall battlefield situation and neural network methods to further analyze the phased characteristics of war and the regional characteristics of the map. It proposes a wargame feature extraction method based on time perception and regional division. The battle process is divided into the initial phase, the engagement phase, and the conclusion phase. Each phase presents vastly different situations and primary operational objectives, significantly influencing commanders' thinking and, consequently, operational intentions. Furthermore, different terrains on the map possess varying strategic significance, and the positions of both sides' combat units change over time. This invention, by dividing the map into regions, can better describe the strategic significance of each operational area, which is crucial for subsequent analysis of operational intentions. From an operator perspective, this invention, while considering the basic characteristics of the operators themselves, focuses on the explicit and implicit relationships between operators, such as attack and rescue, constructing an operator relationship graph to describe the impact of these relationships on wargame intention recognition. By combining the above two parts, convolution is used to extract this information, and then a time series prediction model is used to predict the subsequent trajectory, thereby analyzing the wargame intentions in the current state. Summary of the Invention

[0014] In view of this, the purpose of this invention is to provide a method and system for predicting the position of wargame operators based on battlefield division and attraction maps. This method divides the time phase and battlefield area, enabling information processing and feature extraction. Simultaneously, considering the basic characteristics of the operators themselves, it focuses on the explicit and implicit relationships between operators, such as attack and rescue, constructing an inter-operator relationship graph to describe the impact of these relationships on wargame intent recognition. By combining the above two parts, convolution is used to refine this information, and a temporal prediction model is used to predict subsequent trajectories, thereby analyzing the wargame intent in the current state.

[0015] To achieve the above objectives, the present invention provides the following technical solution:

[0016] A method for predicting the position of wargame operators based on battlefield partitioning and attraction maps, comprising the following steps:

[0017] S1. A time-aware and battlefield segmentation-based method is used for wargame information processing and feature extraction.

[0018] S2. Establish a location prediction model based on attraction map and operator relationship graph: use attraction map to describe the impact of global situation information, including terrain, on wargame operators, use operator relationship graph to represent the impact of interactions between operators, and perform feature fusion to comprehensively consider the main factors that determine the subsequent intentions of operators.

[0019] S3. Combine time-series prediction to simulate combat operations, thereby identifying the subsequent intentions of the target operator.

[0020] In step S1, the game is divided into an initial phase, an engagement phase, and an end phase based on the timeline:

[0021] 1) Initial Phase: In the initial phase, since the combat units of both sides are far apart and outside the enemy's line of sight or range of force, they are less threatened by enemy force. However, both sides are far from the location of the control point and have a low control over the control area. Therefore, the main task in this phase should be to send the main combat units to the vicinity of the main battlefield as quickly as possible, occupy advantageous terrain, and make strategic deployments for the subsequent combat phases. The secondary task is to protect oneself as much as possible during the journey to the main battlefield, while also spying on the enemy's route and other preliminary deployments, and to arrange a small number of troops in a safe area far away from the main battlefield to provide other combat units with a global view as much as possible.

[0022] 2) Engagement Phase: During the engagement phase, since most combat units on both sides are close to each other, they are easily exposed to the enemy's line of sight and range of force. Moreover, combat units will generally not move forward rashly during the engagement phase. Therefore, the main task at this phase becomes how to use the surrounding terrain and other information to better conceal themselves and minimize losses when attacked, while effectively striking key enemy combat units. In addition, at the appropriate time, capture control points to obtain capture points.

[0023] Furthermore, in step S1, during the operation, due to the movement of combat units and situational information such as the field of vision and the range of force strikes, the role represented by each location on the map is not the same. By combining common sense of combat and the stage of combat, the map is mainly divided into six areas: the area of ​​concern, the area of ​​seizure and control, the area of ​​engagement, the safe area, the danger area, and the main activity area.

[0024] Furthermore, in the initial stage of the game, the map is divided into the focus zone, the control zone, the safe zone, and the combat zone, specifically including:

[0025] 1) Area of ​​Interest: Due to the large map size and the fact that control points are generally located near the center, both sides' combat units are distributed in a ring around these control points. Therefore, combat units rarely venture too far into the outer areas. Even if there are high ground areas providing visibility, these areas are too far from the control points for effective observation of the control points and the main battlefield, rendering them practically meaningless. Therefore, this method uses a rectangular area defined by a certain perimeter around the locations of combat units and control points as the area of ​​interest. This helps in analyzing key battlefield information and processing subsequent data.

[0026] 2) Control Zone: Capturing control points greatly increases the score of the capturing side, playing a crucial role in winning the game. Therefore, this method draws a control zone with a certain distance as the radius, centered on the control point; subsequent analysis of the terrain within the control zone will increase its attractiveness to combat units.

[0027] 3) Safe Zone: First, draw the line segments from the main combat units of both sides to the point of control, representing the shortest main routes for both sides' combat units; the distance from one side to the point of control is d. r The distance from the other side to the control point is d. b When d b≤ d r At that time, one side uses d b Using d as the radius and the initial position as the center, draw a sector, approximating its safe zone; on the other side, using d as the radius... b Using the diameter as the radius, draw a circle along the main route of movement for that side, which approximates the safe zone for that side, and vice versa;

[0028] 4) Combat Zone: Within the Zone of Concern, all areas other than the Capture Zone and the Safe Zone are considered combat zones. This indicates that combat units in this area are more likely to be exposed and attacked. The impact of terrain on their movement, concealment, and damage reduction is different from that in the Safe Zone.

[0029] Furthermore, during the combat phase of the match, the map area division has been revised compared to the initial phase. Two new areas have been added: the danger zone and the main activity zone; one area has been removed: the combat zone. Specifically, these include:

[0030] 1) Area of ​​Concern: Similar to the initial stage, a rectangular area is drawn with the outer perimeter of the combat unit and the control point as the area of ​​concern;

[0031] 2) Control Zone: Similar to the initial stage, the control zone is drawn with the control point as the center and a certain distance as the radius;

[0032] 3) Danger Zone: Within the area of ​​concern, from one side's perspective, the intersection of the field of vision and the range of force of the other side's combat units is calculated as the danger zone of one side. This zone indicates that one side's combat units are exposed in the field of vision and are within the range of force of some enemy combat units, thus increasing the level of danger and making them more likely to take appropriate measures to conceal or counterattack.

[0033] 4) Safe Zone: Within the area of ​​interest, excluding the danger zone of one's own side, all other parts are considered the scope of one's own safe zone; it means that within this area, combat units are less threatened and can quickly move to other advantageous positions by marching, providing vision or force strike effect for the entire army;

[0034] 5) Main Activity Zone: The main activity zone is an area independent of the control zone, danger zone, and safe zone, but it is also part of the area of ​​concern. It is calculated based on the movement speed of combat units and is used to characterize the area where the current combat unit will primarily engage in its activities. While analyzing long-term trajectories is helpful for understanding the overall strategic deployment when predicting combat unit trajectories, it is very difficult. More importantly, long-term strategic deployments can change due to mid-term events, so simply analyzing long-term strategic deployments is not very meaningful. Therefore, prioritizing the analysis of short-term trajectories is more beneficial for identifying the enemy's operational intentions at the current moment.

[0035] Further, in step S2, a location prediction model based on an attraction map and an operator relationship graph is established, wherein the attraction map includes:

[0036] 1) Control Zone: Control points are of great significance to both sides, so the attractiveness of locations within the control zone is significantly increased radially from the control point, although the increase in attractiveness decreases with increasing radial distance. Operators within the control zone do not need to travel long distances, but rather tend to conceal themselves and protect themselves while attacking the enemy. Therefore, the main focus is on increasing the attractiveness contribution of jungles and settlements within the control zone to the operator's concealment and damage reduction effects, moderately increasing the vision contribution of high ground, and minimizing the impact of terrain slowdown effects.

[0037] 2) Safe Zone: Operators in the safe zone are less threatened and need to rush to the battlefield. This mainly increases the attraction contribution of plains and vegetated terrain. However, since the slowing effect of terrain such as settlements, jungles, and soft ground is significant, the attraction of these areas to operators is reduced.

[0038] 3) Combat Zone: Within the combat zone, the operator's crouching or movement range is small, and the deceleration effect is reduced or negligible; jungle and settlement areas significantly reduce the observation distance and damage reduction effect, greatly increasing their attractiveness; open areas such as plains easily expose the field of vision, reducing their attractiveness; elevation has a good field of vision, and if the surrounding area has concealment barriers, it can increase the contribution of elevation to the field of vision attractiveness and the contribution of the surrounding area to the concealment attractiveness.

[0039] 4) Main activity area: During the combat phase, the operator's short-term travel range is relatively small, and some terrain features in its main activity area are easier for it to utilize. Therefore, the attraction contribution of the main activity area should be increased appropriately.

[0040] Furthermore, in step S2, the operator relation graph includes:

[0041] The operator relation graph mainly consists of two parts: node information and edge information.

[0042] 1) Node information is integrated with basic information and state information, and finally an attribute vector is formed for each operator to represent the current operator state;

[0043] 2) The information of the edges mainly includes relationships such as visibility, attack, shared vision, and assistance. Different weights are assigned to each to represent the relationship between operators at the current moment, and finally represented in the form of an adjacency matrix.

[0044] At each moment, an attraction map and an operator relationship graph can be extracted. This method extracts the attraction map and operator relationship graph from multiple moments for each operator as a single dataset. Therefore, it can extract a complete dataset from multiple games for use in the third stage of position prediction.

[0045] Furthermore, in step S2, the location prediction model includes:

[0046] 1) Extracting information from the attraction map: Since the attraction map is in a standard matrix form, convolutional operations are used to extract this grid data (this method uses Convolutional Neural Networks (CNN)) to obtain the embedded representation of the attraction map; through model learning, information aggregation is effectively achieved while reducing information loss.

[0047] 2) Extracting information from the operator relation graph: As a graph, the operator relation graph can also be extracted from the data based on convolution operations, but traditional CNNs cannot be used to aggregate its information. Instead, a graph convolutional neural network (GCN) suitable for graph form should be used to obtain the embedded representation of the operator relation graph.

[0048] 3) Position prediction: The information from the two aspects is fused to obtain the overall embedding representation; then, the recurrent neural network (RNN) in deep learning is used to predict the sequence data of the obtained overall embedding representation and analyze the motion trajectory of the target operator in subsequent time steps, so as to achieve the purpose of recognizing the intention of the enemy operator.

[0049] The present invention also provides a wargame operator position prediction system based on battlefield division and attraction map.

[0050] The beneficial effects of this invention are as follows:

[0051] This invention fully considers the characteristics of wargaming game data, such as its complexity, unstructured nature, low information value density, and significant information gaps over time. Combining relevant domain knowledge and deductive logic, it proposes a reasonable information processing and feature extraction method from the perspectives of time perception and battlefield division. This effectively improves the insufficient feature extraction capabilities of existing technologies in wargaming and forms a standardized semi-structured wargaming dataset, which helps improve the efficiency of subsequent research in utilizing wargaming information. Furthermore, in terms of deep information extraction, this invention proposes using an attraction map to describe the impact of global situational information such as terrain on wargaming operators, and using an operator relationship graph to represent the influence of interactions between operators. Feature fusion is also performed to comprehensively consider the main factors determining the subsequent intentions of operators. Finally, the invention uses a time-series prediction approach to simulate combat, thereby identifying the subsequent intentions of target operators. This provides a new approach and benchmark for data mining in the wargaming field and offers theoretical support for wargaming simulation teaching materials in military academies.

[0052] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0053] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0054] Figure 1 This is a framework diagram of the technical solution of the present invention;

[0055] Figure 2 Here is a detailed flowchart of an example;

[0056] Figure 3 Maps for each stage in the embodiment;

[0057] Figure 4 This is a position sequence diagram from the embodiment;

[0058] Figure 5 This is a diagram illustrating the operator relationships in the embodiment.

[0059] Figure 6 This is an example of an attraction map. Detailed Implementation

[0060] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0061] Figure 1As shown in the figure, the wargame operator position prediction method based on battlefield division and attraction map provided by this invention mainly includes the following steps:

[0062] S1. A time-aware and battlefield segmentation-based method is used for wargame information processing and feature extraction.

[0063] S2. Establish a location prediction model based on attraction map and operator relationship graph: use attraction map to describe the impact of global situation information, including terrain, on wargame operators, use operator relationship graph to represent the impact of interactions between operators, and perform feature fusion to comprehensively consider the main factors that determine the subsequent intentions of operators.

[0064] S3. Combine time-series prediction to simulate combat operations, thereby identifying the subsequent intentions of the target operator.

[0065] In this embodiment, the method is applied to position prediction in wargaming simulations. A specific implementation is as follows:

[0066] Figure 2 As shown in the flowchart of the specific implementation example, the process includes:

[0067] (1) Collect multiple game data of the same map and analyze the game data to mine effective information. Taking the game data of the water network rice field map as an example, put the game data into the same map folder. The system traverses the game files one by one, divides the game into stages, hides some data, and mines the legal operator position sequence.

[0068] Figure 3 The map shown in the example illustrates the different stages of the game. Observing the map at each stage reveals that in the initial stage, both sides are relatively far from the control point, so they are busy rushing to the main combat zone to make strategic deployments. During the combat phase, both sides have discovered some or all of the enemy's pieces, and small-scale battles may have already taken place. In the final stage, the control point is occupied by the blue team, and the red team's main forces have been completely eliminated by the blue team, marking the end of the game.

[0069] The reason for hiding the data is that the game data is presented from a god's-eye view. By hiding some of the enemy's data, the real game situation shown in the picture above can be simulated from a data level.

[0070] By recording the operator's movement position, a pattern can be formed as follows: Figure 4 The position sequence is shown. By tracing the entire game, a set of operator position sequences can be obtained.

[0071] (2) Constructing the attraction map and operator relationship matrix

[0072] Based on the stage of the game and the sequence of operator positions, the relationships between operators are extracted, and an operator relationship graph matrix is ​​constructed. Simultaneously, the battlefield spatial state is divided, and combined with the terrain at corresponding battlefield locations, the attractive force of the terrain on the operators is analyzed, constructing an attraction map information.

[0073] like Figure 5 As shown, Figure 5 This is a diagram depicting the operator relationships at the current moment from the perspective of the red side. Dashed lines represent observable relationships, and solid lines represent attack relationships. For example... Figure 5 As shown, the red team's unmanned combat vehicles and tanks can observe two enemy tanks, while the blue team's tanks can observe the red team's tanks and armored vehicles, and can also attack the red team's armored vehicles. Therefore, adjacency matrices can be used to record the relationships between these nodes, along with some node attributes, to describe the operator relationship graph.

[0074] like Figure 6 As shown, Figure 6 This analysis uses the perspective of the red tank to depict the attractiveness map of the current battle situation to it. The darker the color, the greater the attraction of that location. This aligns with objective reality: when a red tank is attacked by a blue tank, it will quickly retaliate and seek cover in the surrounding area. Therefore, the attractiveness of nearby jungles and settlements increases significantly, while the attractiveness of open areas decreases. Through similar calculations, an attractiveness map for the target operator can be obtained at each moment of the entire game, used for subsequent extraction of embedding representations.

[0075] (3) Extract the embedding representations of the attraction map and operator relationship graph.

[0076] Information about the attraction map is extracted using a convolutional network, and information about the operator relation graph is extracted using a graph convolutional network. Embedded representations of the attraction map and the operator relation graph are then obtained, respectively. This process involves processing the aforementioned information.

[0077] (4) Feature fusion and location prediction

[0078] The attraction map embedding representation and the operator relation graph embedding representation of the same operator position sequence are used as a whole. The same time is used as the alignment standard to perform feature fusion of the two embedding representations. The fused features are predicted by a time series prediction model to obtain the subsequent position information of the operator.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting the position of wargame operators based on battlefield division and attraction maps, characterized in that: The method includes the following steps: S1. A time-aware and battlefield segmentation-based method is used for wargame information processing and feature extraction. S2. Establish a location prediction model based on attraction map and operator relationship graph: use attraction map to describe the impact of global situation information, including terrain, on wargame operators, use operator relationship graph to represent the impact of interactions between operators, and perform feature fusion to comprehensively consider the factors that determine the subsequent intentions of operators; S3. Combine time-series prediction to simulate combat operations, thereby identifying the subsequent intentions of target operators; In step S2, the operator relation graph includes: The operator relation graph consists of two parts: node information and edge information. 1) Node information is integrated with basic information and state information, and finally an attribute vector is formed for each operator to represent the current operator state; 2) The information of the edges includes visibility, attack, shared vision, and aid relationship, which are assigned different weights to represent the relationship between operators at the current time, and are finally represented in the form of an adjacency matrix; 3) At each time step, an attraction map and an operator relationship graph can be extracted; for each operator, the attraction map and operator relationship graph at multiple time steps are extracted as a data set, so a complete dataset can be extracted from multiple games for location prediction. In step S2, the location prediction model includes: 1) Extracting information from the attraction map: Since the attraction map is in a standard matrix form, convolution is used. The operation is used to refine this grid data and obtain the embedded representation of the attraction map; through model learning, information aggregation is effectively achieved while reducing information loss. 2) Extracting information from the operator relation graph: The operator relation graph, in the form of a graph, is based on convolution operations. The data is refined, and a graph convolutional neural network (GCN) suitable for graph form is used to obtain the embedded representation of the operator relation graph; 3) Position prediction: The information from the two aspects is fused to obtain the overall embedding representation; then, the recurrent neural network (RNN) in deep learning is used to predict the sequence data of the obtained overall embedding representation and analyze the motion trajectory of the target operator in subsequent time steps, so as to achieve the purpose of recognizing the intention of the enemy operator.

2. The method for predicting the position of wargame operators based on battlefield division and attraction maps according to claim 1, characterized in that: In step S1, during the operation, the map is divided into six areas: the area of ​​interest, the area to be seized, the area of ​​combat, the safe area, the danger area, and the main activity area.

3. The method for predicting the position of wargame operators based on battlefield division and attraction maps according to claim 2, characterized in that: In the initial stage of the game, the map is divided into the focus zone, the control zone, the safe zone, and the combat zone, specifically including: 1) Area of ​​Interest: A rectangular area is drawn with the outer perimeter of the combat unit and the control point as the area of ​​interest. This helps to standardize the analysis of key battlefield information and the processing of subsequent data. 2) Control Zone: Control points increase the score of the controlling party. The control zone is drawn with a certain distance as the center. 3) Safe Zone: First, draw the line segments from the main combat units of both sides to the control point, representing the shortest routes for both sides' combat units; the distance from one side to the control point is d. r The distance from the other side to the control point is d. b d b≤ d r Then one side uses d b Draw a sector with radius d and the initial position as the center, which is its safe zone; on the other side, with radius d... b Using the diameter as the radius, draw a circle along the direction of the movement path to define the safe zone for that side, and vice versa; 4) Combat Zone: Within the Zone of Concern, all areas other than the Capture Zone and the Safe Zone are considered combat zones.

4. The method for predicting the position of wargame operators based on battlefield division and attraction maps according to claim 3, characterized in that: During the combat phase of the match, the map area division has been revised compared to the initial phase. Two new areas have been added: the danger zone and the main activity zone. One area has been removed: the combat zone. Specifically, the following areas have been added: 1) Area of ​​Concern: Similar to the initial stage, a rectangular area is drawn with the outer perimeter of the combat unit and the control point as the area of ​​concern; 2) Control Zone: Similar to the initial stage, the control zone is drawn with the control point as the center and a certain distance as the radius; 3) Danger Zone: Within the area of ​​interest, from one side's perspective, the intersection of the line of sight and range of force of the other side's combat units is calculated as the danger zone of one side. 4) Safe Zone: Within the area of ​​interest, excluding the player's own danger zone, all other parts are considered the player's safe zone; 5) Main Activity Zone: The main activity zone is an area separate from the control zone, danger zone, and safe zone, but it is also part of the area of ​​concern. It is calculated based on the movement speed of combat units.

5. The method for predicting the position of wargame operators based on battlefield division and attraction maps according to claim 4, characterized in that: In step S2, a location prediction model based on an attraction map and an operator relationship graph is established, wherein the attraction map includes: 1) Control Zone: Starting from the control point, increase the attractiveness of the location within the control zone in a radial pattern. The farther the radial distance, the less the attractiveness increases. Increase the attractiveness contribution of jungle and residential areas within the control zone to the operator's concealment and damage reduction effects, and increase the vision contribution of high ground. 2) Safe Zone: Operators within the safe zone are less threatened and need to quickly move to the battlefield, increasing the attractiveness contribution of plains and unvegetated terrain; while due to the significant slowing effect of terrain such as settlements, jungles, and soft ground on operators, the attractiveness of these areas to operators is reduced. 3) Combat Zone: Within the combat zone, the operator crouches or has a small range of movement, and the deceleration effect is reduced or negligible; jungles and settlements significantly reduce the observation distance and damage reduction effect, increasing their attractiveness; open areas such as plains easily expose the field of vision, reducing their attractiveness; elevation has a good field of vision, and if the surrounding area has concealment barriers, it increases the contribution of elevation to the field of vision attractiveness and the contribution of the surrounding area to the concealment attractiveness. 4) Main activity area: During the combat phase, the operator's short-term travel range is small, and some terrain features in its main activity area are easier for it to utilize, thus increasing the attractiveness contribution of the main activity area.

6. A wargame operator position prediction system based on battlefield division and attraction map, employing the method described in any one of claims 1 to 5.

Images