Method for generating attack plan based on instruction-driven war game

Abstract

The application discloses a kind of attack scheme generation methods based on instruction driving's Kriegspiel, comprising: receiving instruction containing "attack" semantic, extracting attack area description, camp identification, task type and expected effect by using large language model, and converting into attack axis coordinate sequence;Call map data query tool to obtain terrain traffic data, enemy defense posture and key terrain node, and calculate the path segment penetration difficulty coefficient;Based on penetration difficulty coefficient and enemy defense posture, a multi-objective optimization algorithm is used to generate a candidate attack axis set, and the optimal and alternative main attack direction is decided in combination with the task type;According to the optimal main attack direction, automatically organize the assault echelon, the reserve echelon and the fire support echelon;The whole process of attack is divided into fire preparation, breakthrough, depth development and consolidation stage, and the echelon strength, target coordinates and conversion trigger condition in each stage are allocated.

Description

Technical Field

[0001] This application relates to the field of wargaming technology, and in particular to a method for generating wargaming attack plans based on instruction-driven methods. Background Technology

[0002] Wargaming is an important tool for military command training and operational plan evaluation. By simulating battlefield environments, troop deployments, and combat operations, it provides decision support for commanders. Offensive operations are one of the core subjects of wargaming, involving complex decisions such as the selection of the main attack direction, troop organization and configuration, and phase division and coordination.

[0003] The existing wargaming simulation systems have the following main shortcomings in their offensive deployment techniques: 1. High professional threshold and low configuration efficiency. Traditional wargaming systems rely on commanders possessing extensive military tactical knowledge and system operation skills for offensive scenario design. Users must manually select the attack starting point, draw the attack axis segment by segment, and configure echelon operators one by one in a graphical interface. Completing a division-level offensive scenario typically takes several hours. During the configuration process, problems such as axis deviation from the optimal path, echelon coordination breakdown, and misaligned fire support timing are prone to occur.

[0004] 2. Lack of instruction-driven end-to-end generation capabilities The existing system does not incorporate large language model technology and cannot parse commands such as "carry out an attack along the river valley into the enemy's defensive depth, with an expected breakthrough of 20 kilometers." Users need to break down tactical intentions into parameters that the system can recognize (coordinates, time, troop numbers), which is cumbersome and the transmission of intentions is prone to distortion.

[0005] 3. The decision on the main direction of attack relies on experience-based judgment. Existing systems often rely on manual plotting or simple matching based on fixed templates to select the main attack direction, lacking quantitative and comprehensive analysis of terrain access data, enemy defense posture, and penetration difficulty. When facing complex terrain and flexible defenses, it is difficult to scientifically evaluate the advantages and disadvantages of multiple candidate axes, easily overlooking the optimal solution. Summary of the Invention

[0006] The purpose of this invention is to solve the above-mentioned problems by providing a method for generating wargaming attack schemes based on instruction-driven methods.

[0007] The technical solution of this application is implemented as follows: This invention provides a method for generating attack plans in wargaming simulations based on instruction-driven methods, applicable to wargaming simulation systems, comprising the following steps: S1. Semantic parsing step: Receive an instruction containing the semantics of "attack", use a large language model to extract the description of the attack area, the attacker's faction identifier, the attack mission type and the expected effect, and convert the description of the attack area into a sequence of attack axis coordinates in a standard grid coordinate system. S2, Situational Awareness Step: Call the map data query tool to obtain the terrain access data, enemy defense situation data and key terrain node information corresponding to the attack axis coordinate sequence, and calculate the breakthrough difficulty coefficient of each path segment; S3. Main attack direction decision-making steps: Based on the penetration difficulty coefficient and enemy defense situation data, a set of candidate attack axes is generated using a multi-objective optimization algorithm. Combining the attack mission type and expected results, the optimal main attack direction and alternative main attack directions are decided from the set of candidate attack axes. S4. Troop formation steps: Based on the axis length of the optimal main attack direction, terrain complexity and enemy defense strength, automatically form assault echelons, reserve echelons and fire support echelons, and generate troop configuration lists and coordination schedules for each echelon. S5. Phase Division Steps: Divide the entire offensive process into the fire preparation phase, breakthrough phase, in-depth development phase, and consolidation phase, and assign corresponding echelon forces, phase target coordinates, and conversion trigger conditions to each phase. S6. Scheme verification steps: Conduct tactical feasibility verification of the entire offensive process, including verification of the timeliness of the assault window, verification of the timing of echelon coordination, and verification of the security of the logistics supply line. If the verification fails, return to S3 to adjust the weight of the main attack direction or return to S4 to adjust the echelon grouping ratio.

[0008] As a further improvement, in step S1, the attack mission types include: frontal breakthrough, flank flanking maneuver, deep penetration, encirclement and annihilation, and key point capture and control; when the instruction lacks an attack mission type, the large language model infers intent based on the terrain feature words in the attack area description and the enemy's defensive posture keywords. The terrain feature words include "open valley", "mountain pass", and "urban street", and the enemy's defensive posture keywords include "fortified position", "mobile defense", and "key point defense".

[0009] As a further improvement, in step S1, the expected effect includes qualitative description and quantitative indicators. The qualitative description is mapped to a preset quantitative target set through a large language model. The quantitative target set includes: the depth of the defense line in kilometers, the proportion of enemy operators annihilated, the coordinates of key nodes seized and controlled, and the area of ​​the bridgehead established. The quantitative indicators extract values ​​and units from the instructions through regular expressions.

[0010] As a further improvement, in step S1, the method for converting the attack area description into an attack axis coordinate sequence includes: converting the starting area and target area described in natural language into center coordinate points respectively; using the least resistance path algorithm to generate a reference axis connecting the starting coordinate point to the target coordinate point; discretizing the reference axis into a tightly packed hexagonal grid; and extracting the sequence of center points of the grid cells along the path as the attack axis coordinate sequence.

[0011] As a further improvement, in step S2, the penetration difficulty coefficient is calculated using the following formula: D_pen =(H_defense × F_obstacle × C_visibility) / (K_mobility × S_surprise) where H_defense is the enemy's defense strength coefficient, F_obstacle is the terrain obstacle coefficient, C_visibility is the visibility exposure coefficient, K_mobility is the attacker's mobility coefficient, and S_surprise is the surprise attack coefficient, which is determined based on the attack initiation time and the enemy's combat readiness assessment.

[0012] As a further improvement, in step S2, the enemy defense posture data is obtained through multi-source information fusion. The multi-source information includes: enemy operator coordinates and types returned by the deployment posture query tool, enemy forward position image recognition results returned by the reconnaissance operator, and enemy communication intensity distribution intercepted by the electronic warfare system. The multi-source information is fused using DS evidence theory to generate an enemy defense posture confidence heatmap.

[0013] As a further improvement, in step S2, the key terrain nodes include passage nodes, commanding heights, ferry bridges, and urban transportation hubs; each key terrain node is assigned a control priority, which is determined by the analytic hierarchy process (AHP). The factors in the judgment matrix include: the degree of control the node has over the attack axis, the density of enemy defensive forces, and the support value for our subsequent maneuvers after control.

[0014] As a further improvement, in step S3, the multi-objective optimization algorithm takes minimizing penetration time, minimizing expected casualties, and maximizing surprise as optimization objectives, and uses the NSGA-III algorithm to generate a Pareto optimal solution set as the candidate attack axis set; each candidate axis in the Pareto optimal solution set carries a three-dimensional target vector, which the user can choose according to strategic preferences.

[0015] As a further improvement, in step S3, when deciding the optimal main attack direction from the set of candidate attack axes, the weight coefficient of the optimization target is dynamically adjusted according to the attack mission type: when the attack mission type is "deep penetration", the breakthrough time weight coefficient is higher than the expected casualty weight coefficient; when the attack mission type is "encirclement and annihilation", the surprise weight coefficient is higher than the breakthrough time weight coefficient.

[0016] As a further improvement, in step S3, the alternative main attack direction is configured with automatic triggering conditions, which include: the breakthrough time of the main attack direction exceeds the expected threshold, the main attack direction encounters a counterattack by the enemy's reserve force, and the main attack direction fails to seize control of key terrain nodes; when any triggering condition is met, the system automatically transfers the center of gravity of the forces to the alternative main attack direction and re-executes S4-S6 to generate the adjusted attack deployment plan.

[0017] As a further improvement, in step S4, each echelon adopts a modular grouping structure. The assault echelon consists of an armored assault module, an infantry accompanying module, and an engineering obstacle breaching module. The reserve echelon consists of a rapid response module and a fire support module. When the loss of a module exceeds a preset ratio during the simulation, the same type of module is automatically called from the reserve echelon for replenishment and reorganization to maintain the integrity of the echelon formation.

[0018] As a further improvement, in step S4, the transition trigger conditions for each stage include: the achievement of the stage target coordinates, the arrival of the stage time limit, or a major counterattack by the enemy; when the transition trigger conditions are met, the system automatically releases the action authority of the next stage echelon, reorganizes the undamaged troops in the current stage echelon into subsequent echelons, and updates the time sequence dependencies in the coordination schedule.

[0019] As a further improvement, in step S4, the coordination schedule of the fire support echelon includes: the suppression firing window during the fire preparation phase, the accompanying firing window during the breakthrough phase, and the interception firing window during the deep development phase; each window is bound to the maneuver node of the assault echelon through a timestamp, and when the assault echelon is delayed for any reason, the corresponding window of the fire support echelon is automatically extended or switched to the backup firing plan.

[0020] As a further improvement, in step S6, the security verification of the logistics supply line includes: based on the current attack axis coordinate sequence, using the shortest path algorithm to calculate multiple alternative supply lines from the rear supply base to the target coordinates at each stage, and evaluating the enemy fire threat level and terrain accessibility of each alternative supply line in real time; when the threat level of the main supply line exceeds the threshold, automatically switching to the backup supply line and updating the ammunition consumption coefficient of each echelon.

[0021] As a further improvement, in step S6, when the verification fails, the feedback adjustment strategy includes: if the assault window timeliness verification fails, return to S3 to increase the penetration time weight coefficient and regenerate the candidate axis; if the echelon coordination timing verification fails, return to S4 to adjust the departure timestamps and maneuver speed parameters of each echelon; if the logistics supply line security verification fails, return to S2 to expand the situational awareness range to the enemy's deep defense area and reassess the penetration difficulty coefficient.

[0022] The advantages or beneficial effects of the above technical solutions include at least the following: By using a large language model to achieve end-to-end parsing of natural language offensive intentions, complex instructions such as "carry out an attack along the river valley into the enemy's defensive depth, with an expected breakthrough of 20 kilometers" can be automatically converted into an executable offensive deployment plan, significantly reducing the professional threshold. By quantifying the difficulty coefficient of penetration and using multi-objective optimization algorithms, candidate attack axes are scientifically evaluated, avoiding the subjectivity of experience-based judgments and improving the optimization of the main attack direction decision. Modular echelon formation and automatic phase transition enable flexible and adaptive force configuration, supporting dynamic reorganization and adjustment when the simulation process deviates from expectations; By employing a triple-validation closed loop and differentiated feedback adjustment strategy, the tactical feasibility of the offensive deployment plan is ensured. When validation fails, the system automatically backtracks to the corresponding stage for optimization, reducing manual intervention. Attached Figure Description

[0023] The accompanying drawings illustrate exemplary embodiments of the present application and, together with the description thereof, serve to explain the principles of the present application. These drawings are included to provide a further understanding of the present application and are incorporated in and constitute a part of this specification.

[0024] Figure 1 A flowchart of the instruction-driven wargaming attack scheme generation method provided by an embodiment of the present invention is shown. Detailed Implementation

[0025] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0026] It should be noted that, where there is no conflict, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] The names of the messages or information exchanged between multiple devices in the embodiments of this application are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0028] Reference Figure 1 This invention provides a method for generating attack plans in wargaming simulations based on instruction-driven methods, applicable to wargaming simulation systems, comprising the following steps: S1. Semantic parsing step: Receive an instruction containing the semantics of "attack", use a large language model to extract the description of the attack area, the attacker's faction identifier, the attack mission type and the expected effect, and convert the description of the attack area into a sequence of attack axis coordinates in a standard grid coordinate system. S2, Situational Awareness Step: Call the map data query tool to obtain the terrain access data, enemy defense situation data and key terrain node information corresponding to the attack axis coordinate sequence, and calculate the breakthrough difficulty coefficient of each path segment; S3. Main attack direction decision-making steps: Based on the penetration difficulty coefficient and enemy defense situation data, a set of candidate attack axes is generated using a multi-objective optimization algorithm. Combining the attack mission type and expected results, the optimal main attack direction and alternative main attack directions are decided from the set of candidate attack axes. S4. Troop formation steps: Based on the axis length of the optimal main attack direction, terrain complexity and enemy defense strength, automatically form assault echelons, reserve echelons and fire support echelons, and generate troop configuration lists and coordination schedules for each echelon. S5. Phase Division Steps: Divide the entire offensive process into the fire preparation phase, breakthrough phase, in-depth development phase, and consolidation phase, and assign corresponding echelon forces, phase target coordinates, and conversion trigger conditions to each phase. S6. Scheme verification steps: Conduct tactical feasibility verification of the entire offensive process, including verification of the timeliness of the assault window, verification of the timing of echelon coordination, and verification of the security of the logistics supply line. If the verification fails, return to S3 to adjust the weight of the main attack direction or return to S4 to adjust the echelon grouping ratio.

[0029] As a further improvement, in step S1, the attack mission types include: frontal breakthrough, flank flanking maneuver, deep penetration, encirclement and annihilation, and key point capture and control; when the instruction lacks an attack mission type, the large language model infers intent based on the terrain feature words in the attack area description and the enemy's defensive posture keywords. The terrain feature words include "open valley", "mountain pass", and "urban street", and the enemy's defensive posture keywords include "fortified position", "mobile defense", and "key point defense".

[0030] As a further improvement, in step S1, the expected effect includes qualitative description and quantitative indicators. The qualitative description is mapped to a preset quantitative target set through a large language model. The quantitative target set includes: the depth of the defense line in kilometers, the proportion of enemy operators annihilated, the coordinates of key nodes seized and controlled, and the area of ​​the bridgehead established. The quantitative indicators extract values ​​and units from the instructions through regular expressions.

[0031] As a further improvement, in step S1, the method for converting the attack area description into an attack axis coordinate sequence includes: converting the starting area and target area described in natural language into center coordinate points respectively; using the least resistance path algorithm to generate a reference axis connecting the starting coordinate point to the target coordinate point; discretizing the reference axis into a tightly packed hexagonal grid; and extracting the sequence of center points of the grid cells along the path as the attack axis coordinate sequence.

[0032] As a further improvement, in step S2, the penetration difficulty coefficient is calculated using the following formula: D_pen =(H_defense × F_obstacle × C_visibility) / (K_mobility × S_surprise) where H_defense is the enemy's defense strength coefficient, F_obstacle is the terrain obstacle coefficient, C_visibility is the visibility exposure coefficient, K_mobility is the attacker's mobility coefficient, and S_surprise is the surprise attack coefficient, which is determined based on the attack initiation time and the enemy's combat readiness assessment.

[0033] As a further improvement, in step S2, the enemy defense posture data is obtained through multi-source information fusion. The multi-source information includes: enemy operator coordinates and types returned by the deployment posture query tool, enemy forward position image recognition results returned by the reconnaissance operator, and enemy communication intensity distribution intercepted by the electronic warfare system. The multi-source information is fused using DS evidence theory to generate an enemy defense posture confidence heatmap.

[0034] As a further improvement, in step S2, the key terrain nodes include passage nodes, commanding heights, ferry bridges, and urban transportation hubs; each key terrain node is assigned a control priority, which is determined by the analytic hierarchy process (AHP). The factors in the judgment matrix include: the degree of control the node has over the attack axis, the density of enemy defensive forces, and the support value for our subsequent maneuvers after control.

[0035] As a further improvement, in step S3, the multi-objective optimization algorithm takes minimizing penetration time, minimizing expected casualties, and maximizing surprise as optimization objectives, and uses the NSGA-III algorithm to generate a Pareto optimal solution set as the candidate attack axis set; each candidate axis in the Pareto optimal solution set carries a three-dimensional target vector, which the user can choose according to strategic preferences.

[0036] As a further improvement, in step S3, when deciding the optimal main attack direction from the set of candidate attack axes, the weight coefficient of the optimization target is dynamically adjusted according to the attack mission type: when the attack mission type is "deep penetration", the breakthrough time weight coefficient is higher than the expected casualty weight coefficient; when the attack mission type is "encirclement and annihilation", the surprise weight coefficient is higher than the breakthrough time weight coefficient.

[0037] As a further improvement, in step S3, the alternative main attack direction is configured with automatic triggering conditions, which include: the breakthrough time of the main attack direction exceeds the expected threshold, the main attack direction encounters a counterattack by the enemy's reserve force, and the main attack direction fails to seize control of key terrain nodes; when any triggering condition is met, the system automatically transfers the center of gravity of the forces to the alternative main attack direction and re-executes S4-S6 to generate the adjusted attack deployment plan.

[0038] As a further improvement, in step S4, each echelon adopts a modular grouping structure. The assault echelon consists of an armored assault module, an infantry accompanying module, and an engineering obstacle breaching module. The reserve echelon consists of a rapid response module and a fire support module. When the loss of a module exceeds a preset ratio during the simulation, the same type of module is automatically called from the reserve echelon for replenishment and reorganization to maintain the integrity of the echelon formation.

[0039] As a further improvement, in step S4, the transition trigger conditions for each stage include: the achievement of the stage target coordinates, the arrival of the stage time limit, or a major counterattack by the enemy; when the transition trigger conditions are met, the system automatically releases the action authority of the next stage echelon, reorganizes the undamaged troops in the current stage echelon into subsequent echelons, and updates the time sequence dependencies in the coordination schedule.

[0040] As a further improvement, in step S4, the coordination schedule of the fire support echelon includes: the suppression firing window during the fire preparation phase, the accompanying firing window during the breakthrough phase, and the interception firing window during the deep development phase; each window is bound to the maneuver node of the assault echelon through a timestamp, and when the assault echelon is delayed for any reason, the corresponding window of the fire support echelon is automatically extended or switched to the backup firing plan.

[0041] As a further improvement, in step S6, the security verification of the logistics supply line includes: based on the current attack axis coordinate sequence, using the shortest path algorithm to calculate multiple alternative supply lines from the rear supply base to the target coordinates at each stage, and evaluating the enemy fire threat level and terrain accessibility of each alternative supply line in real time; when the threat level of the main supply line exceeds the threshold, automatically switching to the backup supply line and updating the ammunition consumption coefficient of each echelon.

[0042] As a further improvement, in step S6, when the verification fails, the feedback adjustment strategy includes: if the assault window timeliness verification fails, return to S3 to increase the penetration time weight coefficient and regenerate the candidate axis; if the echelon coordination timing verification fails, return to S4 to adjust the departure timestamps and maneuver speed parameters of each echelon; if the logistics supply line security verification fails, return to S2 to expand the situational awareness range to the enemy's deep defense area and reassess the penetration difficulty coefficient.

[0043] Example: Scenario Setting: Red vs. Blue confrontation simulation. The user (Red Commander) inputs the command: "Conduct a deep infiltration from area A03 to area A07, utilizing the river valley passage for breakthrough, and expect to break through the enemy's defense depth of 30 kilometers within 48 hours, seizing control of the cross-river bridge." S1, Semantic parsing: The parsing results of the large language model are shown in Table 1.

[0044] Table 1 shows the parsing results of the large language model.

[0045] In one embodiment, assuming the task type is missing, such as when the user only enters "attack from A03 to A07" without specifying the task type, the intent inference needs to be performed as follows: Terrain feature word recognition: The description contains "river valley", which means the terrain is open and suitable for high-speed maneuvering; Enemy defensive posture keywords: The system query indicates that the enemy is configured with "mobile defense"; Intent inference rules: The combination of open river valley terrain and mobile defense suggests a mission type of deep penetration (using speed to penetrate elastic defenses). Mountain pass + fortified position, the mission type can be inferred to be frontal breakthrough (strong attack and hard take). City blocks + key point defense, the mission type can be inferred to be key point capture (street-by-street battle).

[0046] Then, the quantification of the expected results is as follows: In this embodiment, the qualitative description "penetrating the enemy's defensive depth" is input into a large language model mapping to obtain a quantified target set. - 30km depth of breached defenses (using regular expression to extract "30km"); - Percentage of enemy operators eliminated: Not specified (default 20%); - Coordinates of key control nodes: Bridge coordinates to be identified by S2; - Establish the bridgehead area: a 5km radius around the bridge.

[0047] Furthermore, based on the aforementioned quantified target set, the following attack axis coordinate sequence is generated: Step 1: From starting coordinates (3000, 2000) to target coordinates (7000, 5000); Step 2: Generate the reference axis using the minimum resistance path algorithm; Step 3: Discretize the hexagonal close-packed grid (tactical layer, 400m per cell). Step 4: Extract the sequence of center points of the passing grid; Finally, the specific sequence of attack axis coordinates is generated as follows: [(3000,2000), (3400,2200), (3800,2400), (4200,2800), (4600,3200), (5000,3500), (5400,3800), (5800,4200), [6200,4600), (6600,4800), (7000,5000)], a total of 11 nodes, with an axis length of approximately 3600m (straight-line distance) / actual path length of approximately 4500m.

[0048] S2. Situational awareness mainly includes the following steps: First, the terrain access data was queried for the generated attack axis coordinate sequence, as shown in Table 2: Table 2 is a table of terrain access data query data.

[0049] Table 2 shows that different attack axis coordinate sequences correspond to different terrain types, slopes, and traffic capacity coefficients, among other related data.

[0050] Furthermore, it is also necessary to obtain enemy defense posture data and then perform multi-source fusion.

[0051] In one embodiment, the acquired enemy defense posture data is shown in Table 3.

[0052] Table 3 shows the enemy's defensive posture data.

[0053] Furthermore, in one embodiment, DS evidence theory fusion is performed based on the aforementioned enemy defense posture data, and the specific steps are as follows: Evidence 1: m1 (defense front) = 0.85, m1 (unknown) = 0.15; Evidence 2: m2 (obstacle) = 0.70, m2 (unknown) = 0.30; Evidence 3: m3 (command post) = 0.60, m3 (unknown) = 0.40; Fusion results: m(defense front ∩ obstacle) = 0.85 × 0.70 = 0.595; m(defense front line ∩ command post) = 0.85 × 0.60 = 0.510; m(overall defense strength = high) = 0.78 after normalization.

[0054] Then, a confidence heatmap of the enemy's defensive posture can be generated: that is, the area from (4800, 3200) to (5200, 3500) on the front line of defense is a high-confidence red area.

[0055] Furthermore, it is also necessary to identify key terrain nodes: In one embodiment, the identified key terrain nodes are shown in Table 4.

[0056] Table 4 shows the information on key terrain nodes.

[0057] In one embodiment, taking the path segment (4200,2800)→(5000,3500) as an example, the penetration difficulty coefficient is calculated as shown in Table 5 below.

[0058] Table 5 shows the calculated penetration difficulty coefficients.

[0059] Finally, according to D_pen = (H_defense × F_obstacle × C_visibility) / (K_mobility × S_surprise), D_pen = 0.146 is obtained; based on the above D_pen, the following judgment is made: D_pen = 0.146 < 0.5 (preset threshold), the difficulty of penetration of this path segment is relatively low.

[0060] S3. Decision on the main direction of attack: First, the NSGA-III algorithm is used to generate candidate attack axes.

[0061] Based on the situational awareness results above, three candidate axes were generated using the NSGA-III algorithm, as shown in Table 6: Table 6 shows the information for the three candidate axes.

[0062] In a multi-objective optimization problem, the Pareto optimal set is a set of solutions that cannot dominate each other. "Dominate" means that one solution is no worse than another solution in all objectives, and is strictly better in at least one objective.

[0063] Taking the attack axis selection of the present invention as an example, the performance of the three candidate axes α, β, and γ on three targets (penetration time, expected casualties, and surprise) is shown in Table 6 below.

[0064] Determination of Dominance Relationship: Alpha dominates γ: Alpha's breakthrough time (36h) is slightly longer than γ's (30h), but its expected casualties (25%) are significantly lower than γ's (35%), and its surprise factor (0.6) is higher than γ's (0.4). In other words, Alpha is strictly superior to γ ​​in both "expected casualties" and "surprise factor," and only slightly inferior in "breakthrough time." Overall, γ is completely suppressed by Alpha and has no value in preserving it.

[0065] α and β are not mutually dominant: α has a shorter breakthrough time (36h < 48h), but β has a lower expected casualty rate (15% < 25%) and higher surprise (0.8 > 0.6). Each has its advantages and disadvantages and cannot replace the other.

[0066] Pareto Front: The Pareto front is formed by non-dominant solutions {α, β}. This means that if the commander prioritizes speed, α (36-hour breakthrough) is chosen; if the commander prioritizes concealment and low casualties, β (48-hour detour) is chosen. γ is eliminated because it is dominated by α and does not enter the Pareto front. The final decision requires dynamic weighting based on the mission type (depth penetration), further selecting from {α, β}.

[0067] Finally, the weights are dynamically adjusted based on the "in-depth integration task" as follows: Task type = In-depth interleaving, with the following weight configuration: - Penetration time weight: 0.5 (highest, emphasizing speed penetration); - Expected casualty weighting: 0.3 (some casualties are acceptable); - Suddenness weight: 0.2 (minor, the interlude itself is sudden); Overall score: Axis α: 0.5×(1-36 / 48) + 0.3×(1-25 / 35) + 0.2×0.6 = 0.125 +0.086 + 0.12 = 0.331; Axis β: 0.5×(1-48 / 48) + 0.3×(1-15 / 35) + 0.2×0.8 = 0 +0.171 + 0.16 = 0.331.

[0068] Since it's a draw, the priority rule is triggered: the breakthrough time has the highest weight, so axis α (36 hours < 48 hours) is selected.

[0069] The decision results are then shown in Table 7: Table 7 shows the decision outcome information.

[0070] Finally, the triggering conditions for alternative directions are configured based on the decision results: Triggering condition 1: Axis α penetration time > 48 hours (timeout 12 hours); Triggering condition 2: Axis α encounters the Blue Team's reserve force (coordinates 5200, 3500 area); Triggering condition 3: Bridge capture fails (node ​​(5000,3500) still belongs to the blue team); if any condition is met, the team will automatically switch to axis β and the center of gravity of the forces will shift to the flank.

[0071] S4, troop composition: First, the axis parameters are summarized in Table 8: Table 8 shows the parameter information for the axis.

[0072] Based on the above information, the echelon formation scheme and coordination schedule can be generated as shown in Tables 9 and 10. Echelon formation is a combination of several functional modules that divide the attacking forces according to task assignment and deployment order. Each module independently performs a specific task while also coordinating to achieve the overall offensive objective. Furthermore, the modular design has two main advantages: 1. Automatic replenishment of losses: If the armored assault module suffers losses exceeding 40%, the rapid reaction module is automatically called from the reserve echelon to replenish it, maintaining assault capability; 2. Automatic task switching: After a successful breakthrough, the engineering breaching module can be transferred to the reserve echelon to perform road repair tasks. The coordination schedule specifies the timeline for each echelon's "when, where, and what," ensuring that each branch of service appears at the right place at the right time to form a combined combat force. The echelon formation scheme and the coordination schedule together constitute the "skeleton" (formation) and "blood" (sequence) of the offensive deployment plan, neither of which can be omitted.

[0073] Table 9 shows the echelon formation scheme information.

[0074] Table 10 shows the collaboration schedule.

[0075] S5, Phase Division: Generally speaking, the entire offensive process mainly includes the following four stages: Stage 1: Firepower preparation stage (layout); Stage 2: Breakthrough stage (key); Stage 3: In-depth development stage (expansion); and Stage 4: Consolidation stage (conclusion).

[0076] For Phase 1: Firepower Preparation Phase, please refer to Table 11.

[0077] Table 11 contains information on the fire preparation phase.

[0078] Generally speaking, the fire preparation phase involves bombarding the enemy before battle to stun or cripple them, creating conditions for tank assault. The bombardment cannot be too long (wasting ammunition and losing surprise) nor too short (insufficient effect), hence a 45-minute limit is set.

[0079] Phase 2: Breakthrough Phase (See Table 12)

[0080] Table 12 shows information on the breakthrough stage.

[0081] In this phase, tanks lead the charge, infantry follow, and engineers clear mines and breach obstacles, with the objective of creating an opening in the enemy's defenses and capturing bridges. If casualties are too high (exceeding 40%), it indicates that the advance is stalled, and alternative plans are triggered.

[0082] Phase 3: In-depth development phase (see Table 13)

[0083] Table 13 shows information on the stages of in-depth development.

[0084] During this phase, once a breakthrough is achieved, reserves are deployed to expand the gains and advance deeper into enemy territory. If a counterattack by a large enemy force is encountered, a defensive posture may be necessary.

[0085] Phase 4: Consolidation Phase (See Table 14)

[0086] Table 14 shows information for the consolidation phase.

[0087] During this phase, the main task is to establish a firm foothold, build fortifications, deploy firepower, prevent the enemy from counterattacking and reclaiming the gains, and await new orders from higher command.

[0088] The above four stages require coordinated fire timing. Specifically, the assault echelon movement nodes are bound to the fire timing windows as shown in Table 15. Table 15 shows the firepower timing coordination information.

[0089] S6. Solution Verification: First, verify the timeliness of the emergency window: Planned attack time: 36 hours; Estimated arrival time of enemy reinforcements: 42 hours (from the operational reserve assembly area). Judgment: 36 < 42, the emergency window exists, and the timeliness verification is passed.

[0090] If verification fails (e.g., enemy reinforcements arrive 30 hours earlier than expected): Feedback Adjustment: Return to S3, increase the weight of penetration time to 0.7, and regenerate candidate axes; Possible outcome: Choose axis γ (frontal assault, 30 hours, high casualties but feasible window).

[0091] Then, the timing verification of echelon coordination is performed as follows: Check the dependencies between the departure timestamps of each echelon: The assault echelon departs at T+45 minutes, relying on fire preparation effectiveness ≥70%; The deployment of the reserve echelon in T+6 hours depends on the assault echelon seizing and controlling the bridge; If the assault echelon fails to seize control of the bridge by T+4h (delayed to T+6h): Make feedback adjustments: Return to S4 and adjust the departure timestamp of the reserve echelon to T+8h; or adjust the mobility speed parameters of the assault echelon (increase from off-road speed to road speed).

[0092] Finally, the security of the logistics supply line was verified: Main supply route: Rear base (2000, 1000) -- Valley passage -- Bridge (5000, 3500); Alternative supply route 1: Rear base -- mountain trail -- flank passage -- depth; Alternative supply route 2: Rear base -- air supply -- drop point (6000, 4000); Real-time assessment (T+2h, after the assault team seizes control of the bridge): Main supply route: Passes through (4200, 2800), enemy remaining firepower threat level = medium; Alternative Route 1: Mountain accessibility coefficient = 1.2, low but feasible; Alternative Route 2: Weather is good, helicopters are available.

[0093] Judgment: The threat level of the main supply line is below the threshold, so continue using it; furthermore, if enemy air force counterattacks at T+4h, the threat level of the main supply line will be raised to high. The system will automatically switch to alternative route 2 (air resupply) and update the ammunition consumption coefficient for each echelon (air resupply efficiency × 0.8). In other words, if the logistics supply line security verification fails, it will return to S2 to expand the situational awareness range to the enemy's deep defensive area and reassess the penetration difficulty coefficient. This invention achieves end-to-end automatic generation of offensive deployment plans in wargaming simulations through six core steps: natural language intent parsing, situational awareness and penetration difficulty quantification, multi-target optimization of main attack direction decision-making, modular echelon formation, phase division and fire sequence coordination, and triple verification closed-loop feedback. Compared with existing technologies, this invention has significant advantages in command-driven capabilities, scientific decision-making on the main attack direction, flexible and adaptive force formation, automatic phase transition, and real-time verification feedback, effectively supporting wargaming applications for complex offensive operations.

[0094] Those skilled in the art should understand that the above embodiments are merely for illustrative purposes and are not intended to limit the scope of this application. Those skilled in the art can make other changes or modifications based on the above disclosure, and these changes or modifications still fall within the scope of this application.

Claims

1. A method for generating attack plans in wargaming simulations based on instruction-driven methods, applied to a wargaming simulation system, characterized in that: include: S1. Semantic parsing step: Receive an instruction containing the semantics of "attack", use a large language model to extract the description of the attack area, the attacker's faction identifier, the attack mission type and the expected effect, and convert the description of the attack area into a sequence of attack axis coordinates in a standard grid coordinate system. S2, Situational Awareness Step: Call the map data query tool to obtain the terrain access data, enemy defense situation data and key terrain node information corresponding to the attack axis coordinate sequence, and calculate the breakthrough difficulty coefficient of each path segment; S3. Main attack direction decision-making steps: Based on the penetration difficulty coefficient and enemy defense situation data, a set of candidate attack axes is generated using a multi-objective optimization algorithm. Combining the attack mission type and expected results, the optimal main attack direction and alternative main attack directions are decided from the set of candidate attack axes. S4. Troop formation steps: Based on the axis length of the optimal main attack direction, terrain complexity and enemy defense strength, automatically form assault echelons, reserve echelons and fire support echelons, and generate troop configuration lists and coordination schedules for each echelon. S5. Phase Division Steps: Divide the entire offensive process into the fire preparation phase, breakthrough phase, in-depth development phase, and consolidation phase, and assign corresponding echelon forces, phase target coordinates, and conversion trigger conditions to each phase. S6. Scheme verification steps: Conduct tactical feasibility verification of the entire offensive process, including verification of the timeliness of the assault window, verification of the timing of echelon coordination, and verification of the security of the logistics supply line. If the verification fails, return to S3 to adjust the weight of the main attack direction or return to S4 to adjust the echelon grouping ratio.

2. The method for generating wargaming attack plans based on instruction-driven simulation according to claim 1, characterized in that: In step S1, the attack mission types include: frontal breakthrough, flank flanking maneuver, deep penetration, encirclement and annihilation, and key point capture and control; when the instruction lacks an attack mission type, the large language model infers intent based on terrain feature words and enemy defense posture keywords in the attack area description. The terrain feature words include "open valley", "mountain pass", and "urban street", and the enemy defense posture keywords include "fortified position", "mobile defense", and "key point defense".

3. The method for generating wargaming attack plans based on instruction-driven simulation according to claim 1, characterized in that: In step S1, the expected results include qualitative descriptions and quantitative indicators. The qualitative descriptions are mapped to a preset set of quantitative targets through a large language model. The set of quantitative targets includes: the depth of the defense line breached in kilometers, the percentage of enemy operators annihilated, the coordinates of key nodes seized and controlled, and the area of ​​the bridgehead established. The quantitative indicators extract values ​​and units from the instructions through regular expressions.

4. The method for generating wargaming attack plans based on instruction-driven simulation according to claim 1, characterized in that: In step S1, the method for converting the attack area description into an attack axis coordinate sequence includes: converting the starting area and target area described in natural language into center coordinate points respectively; using the least resistance path algorithm to generate a reference axis connecting the starting coordinate point to the target coordinate point; discretizing the reference axis into a tightly packed hexagonal grid; and extracting the sequence of center points of the grid cells along the path as the attack axis coordinate sequence.

5. The method for generating wargaming attack plans based on instruction-driven simulation according to claim 1, characterized in that: In step S2, the penetration difficulty coefficient is calculated using the following formula: D_pen = (H_defense × F_obstacle × C_visibility) / (K_mobility × S_surprise) where, H_defense is the enemy's defense strength coefficient, F_obstacle is the terrain obstacle coefficient, C_visibility is the visibility exposure coefficient, K_mobility is the attacker's mobility coefficient, and S_surprise is the surprise attack coefficient. The surprise attack coefficient is determined based on the attack initiation time and the enemy's combat readiness assessment.

6. The method for generating wargaming attack plans based on instruction-driven simulation according to claim 1, characterized in that: In step S2, the enemy defense posture data is obtained through multi-source information fusion. The multi-source information includes: enemy operator coordinates and types returned by the deployment posture query tool, enemy forward position image recognition results returned by the reconnaissance operator, and enemy communication intensity distribution intercepted by the electronic warfare system. The multi-source information is fused using DS evidence theory to generate an enemy defense posture confidence heatmap.

7. The method for generating wargaming attack plans based on instruction-driven simulation according to claim 1, characterized in that: In step S2, the key terrain nodes include passage nodes, commanding heights, ferry crossings and bridges, and urban transportation hubs; Each key terrain node is assigned a priority for capture and control. The priority is determined by the hierarchical analysis method. The factors in the judgment matrix include: the degree of control of the node over the attack axis, the density of enemy defensive forces, and the support value for our subsequent maneuvers after capture and control.

8. The method for generating wargaming attack plans based on instruction-driven simulation according to claim 1, characterized in that: In step S3, the multi-objective optimization algorithm takes minimizing penetration time, minimizing expected casualties, and maximizing surprise as optimization objectives, and uses the NSGA-III algorithm to generate a Pareto optimal solution set as the candidate attack axis set; each candidate axis in the Pareto optimal solution set carries a three-dimensional target vector, which the user can choose according to strategic preferences.

9. The method for generating wargaming attack plans based on instruction-driven simulation according to claim 1, characterized in that: In step S3, when deciding the optimal main attack direction from the set of candidate attack axes, the weight coefficient of the optimization target is dynamically adjusted according to the attack mission type: when the attack mission type is "deep penetration", the breakthrough time weight coefficient is higher than the expected casualty weight coefficient; when the attack mission type is "encirclement and annihilation", the surprise weight coefficient is higher than the breakthrough time weight coefficient.

10. The method for generating wargaming attack plans based on instruction-driven simulation according to claim 1, characterized in that: In step S3, the alternative main attack direction is configured with automatic triggering conditions, which include: the breakthrough time of the main attack direction exceeds the expected threshold, the main attack direction encounters a counterattack by the enemy's reserve force, and the main attack direction fails to seize control of key terrain nodes; when any triggering condition is met, the system automatically transfers the center of gravity of the forces to the alternative main attack direction and re-executes S4-S6 to generate the adjusted attack deployment plan.

Images