Multi-unmanned-platform tactical plan optimization method based on pattern mining

Abstract

A multi-unmanned-platform tactical plan optimization method based on pattern mining comprises the steps that after multi-scene tactical data is generated through a simulation platform, utility indexes of the multi-scene tactical data are calculated, and a time sequence database containing position information is constructed; after data structure optimization and pruning filtering processing are carried out on a time sequence database, a task sequence is extracted by adopting depth-first search, unmanned platform mounting is adjusted based on the task sequence, and formation configuration is optimized. According to the method, through time-space-utility multi-dimensional feature fusion, a dynamic double-layer pruning strategy and a closed-loop verification mechanism, extraction and optimization of efficient cooperation rules are realized. The task organization preference refers to a stable cooperation rule formed due to space-time constraint, resource utility maximization demand and tactical target priority difference in the process that multiple unmanned platforms cooperatively execute tasks.

Description

Technical Field

[0001] The present invention relates to a technology in the field of unmanned aerial vehicle control, specifically a multi-unmanned platform tactical planning optimization method based on pattern mining. Background Art

[0002] Multi-unmanned platform collaborative mission systems are widely used in fields such as military reconnaissance and disaster relief, and their task scheduling efficiency directly impacts overall system performance. Traditional methods rely on static parameter settings or single-dimensional analysis, resulting in insufficient dynamic adaptability, fragmented multidimensional features, and low computational efficiency, making them incapable of supporting real-time decision-making in complex battlefield environments. Existing technologies, such as temporal pattern mining and high-utility item set mining, fail to effectively integrate multidimensional features of time, space, and utility, resulting in poor interpretability of task coordination rules and inefficient resource utilization. Summary of the Invention

[0003] To address the shortcomings of existing technologies, such as low interpretability, low robustness, and low mining efficiency, this paper proposes a multi-unmanned platform tactical planning optimization method based on pattern mining. Through the fusion of spatiotemporal-utility multidimensional features, a dynamic two-layer pruning strategy, and a closed-loop verification mechanism, this method extracts and optimizes efficient coordination rules. The task organization preference described here refers to the stable coordination rules formed during the collaborative execution of tasks by multiple unmanned platforms due to spatiotemporal constraints, the need to maximize resource utility, and the differences in tactical objective priorities.

[0004] The present invention is achieved through the following technical solutions:

[0005] The present invention relates to a multi-unmanned platform tactical planning optimization method based on pattern mining. After generating multi-scenario tactical data through a simulation platform, its utility indicators are calculated and a time series database containing position information is constructed. After data structure optimization and pruning and filtering processing are performed on the time series database, a depth-first search is used to extract task sequences. Based on the task sequences, the unmanned platform mounts are adjusted and the formation configuration is optimized.

[0006] Technical Effects

[0007] This invention improves the cross-linked list data structure to store spatiotemporal-utility multidimensional data and, through a dynamic two-layer pruning strategy, generates high-utility multi-unmanned platform mission models. Compared with existing technologies, this invention reduces time complexity by over 30%, significantly improving large-scale data processing capabilities. Combined with a tactical closed-loop verification mechanism, it ensures that the mined rules are both theoretically rigorous and operationally feasible. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 Flowchart of the present invention;

[0009] Figure 2This is a schematic diagram of the cross-link phenotype data structure proposed in the present invention;

[0010] Figure 3 A schematic diagram of the execution relationship between projects;

[0011] Figure 4 and Figure 5 Schematic diagram of the embodiment effect. DETAILED DESCRIPTION

[0012] like Figure 1 As shown, this embodiment relates to a multi-unmanned platform tactical plan optimization method based on pattern mining, including:

[0013] Step 1: Using Xuanyi's Yuanqi software as a high-fidelity simulation platform, generate multi-scenario tactical data, collect the unmanned platform's spatiotemporal trajectory, mission time window, and utility indicators, and perform position discretization preprocessing, including:

[0014] Step 1.1: Use a high-fidelity simulation platform to pre-set the following differentiated tactical scenarios: Concentrated Penetration: Enemy defense radius R = 120 km, formation configuration is a tight diamond formation (spacing ≤ 50 m), and a synchronous penetration strategy (Δt < 5 s); Distributed Area Control: R = 150 km, formation configuration is an echelon configuration (depth 300 m), and an asynchronous penetration strategy (Δt = 30 s); Dynamic Reorganization: R = 100 km, formation configuration is a variable star configuration (spacing ≥ 200 m), and a phased triggering strategy (Δt = 60 s). Enemy parameters are also set: electromagnetic interference intensity J ~ N (0.7, 0.1²) and target maneuvering speed v∈{5, 10, 15} m / s.

[0015] Step 1.2: Use a high-fidelity simulation platform to simulate and collect spatiotemporal data each time to ensure that the database has multi-dimensional characteristics of time dimension, space dimension, and utility dimension.

[0016] The spatiotemporal data include: time data on the order and duration of task execution, location data on the formation configuration and deployment density of unmanned platforms, and data that quantifies the comprehensive effectiveness and cost efficiency of unmanned platforms in the tasks.

[0017] The sequence of priorities is preferably a synchronous penetration with a mission execution sequence of Δt<5s.

[0018] The formation configuration is preferably a staggered configuration or a star formation.

[0019] Step 1.3: Discretize the location data in the collected spatiotemporal data, that is, divide the continuous coordinates into 500m×500m grids and map them into discrete location labels.

[0020] Step 2: Build a time series database containing utility parameters and location information, including:

[0021] Step 2.1: Calculate the utility parameter, specifically: ,in: Indicates the mount performance. Indicates the weight of the mount type; Indicates the effectiveness of the communication network; represents the total cost, ; Represents the weight coefficient, which can be adjusted according to needs.

[0022] Step 2.2: Build a time series database by integrating the time data obtained in step 1.2, the discretized location data obtained in step 1.3, and the utility parameters obtained in step 2.1 to construct an initial utility time series database containing location information.

[0023] Step 3: Construct an improved cross-linked list data structure to store time-space-utility multidimensional data, and filter out inefficient patterns through support pruning (support threshold) and utility pruning (total utility threshold), specifically including:

[0024] Step 3.1: Construct Figure 2 The cross-link table-type data structure shown includes a head node, a linked list node, and a time interval node, which realizes efficient storage of spatiotemporal-utility multidimensional features and supports dynamic pruning.

[0025] The head node represents the first item and the possible relationship, and is represented as a Head node in the figure.

[0026] The items mentioned above represent the objects to be mined, i.e., a certain unmanned platform task, as shown in Table 1.

[0027] Table 1

[0028] The possible relationships mentioned above refer to the execution relationships between projects, which can be divided into the following five types according to the execution time: follows, meets, overlaps, contains, matches, such as Figure 3 shown.

[0029] The linked list node represents the linked item, the relationship between the current item and the previous item, the current utility occupancy (uo) and the remaining utility occupancy (ruo), and is represented as a Linked node in the figure.

[0030] The aforementioned uo represents the utility occupied by the current project X in the current task, and is used to measure the relative contribution of the unmanned platform's mounting resources, communication capabilities, and cost input to the overall task utility within the current task.

[0031] The aforementioned ruo represents the remaining utility occupied by the current project X in the current transaction, and is used to measure the potential contribution of the unmanned platform's unconsumed mounting efficiency, communication capability, and remaining cost to subsequent mission planning within the current mission.

[0032] The time interval node includes a flag bit and the start and end times of the two items. Since items from the same transaction do not need to be expanded, a flag bit transaction is constructed in the time interval node to mark whether the row is from the same transaction, which is represented as a Time interval node in the figure.

[0033] Step 3.2: Through the dual optimization mechanism of support pruning and utility pruning, the algorithm computational complexity is significantly reduced and the pattern mining efficiency is improved. The two pruning methods are performed sequentially.

[0034] The significant reduction in algorithm computational complexity mentioned above refers to the use of a two-layer pruning strategy (support pruning and utility pruning). This method significantly reduces computational complexity under fixed parameters (minimum support α = 0.01, minimum utility threshold β = 30). Figure 4 Experiments show that when the transaction size increases from 100 to 1000, the runtime of our method increases by 600% (10ms → 60ms), while the comparison algorithms ULB-Miner, FEACP, and LCIM see increases of 660%, 900%, and 800%, respectively. Specifically, at a transaction size of 1000, our method reduces runtime by 11.7% compared to ULB-Miner (60ms vs. 68ms), 33.3% compared to FEACP (60ms vs. 90ms), and 25% compared to LCIM (60ms vs. 80ms). This improvement is attributed to the pruning strategy's dynamic elimination of redundant candidate patterns, thus avoiding inefficient computational resource consumption.

[0035] Step 3.2.1: Prune by support and set the minimum support threshold α to avoid invalid expansion of redundant temporal relations.

[0036] Step 3.2.2: Prune by utility to filter low-value candidate patterns and ensure that only high-reward task sequences are retained.

[0037] Step 4: Use a spatiotemporal pattern mining algorithm to extract the task sequence and adjust the optimization rules for the unmanned platform mounting and formation configuration generation, including:

[0038] Step 4.1: After extracting patterns that satisfy support and utility constraints using the spatiotemporal pattern mining algorithm proposed in step 3, improve the interpretability of the patterns through analysis.

[0039] The described mode includes: wide-area reconnaissance → continuous reconnaissance → precision strike.

[0040] The typical pattern of wide-area reconnaissance → persistent reconnaissance → precision strike represents a closed-loop reconnaissance and strike-killing model. Specifically, after wide-area reconnaissance generates a preliminary target distribution, persistent reconnaissance immediately focuses on key areas, frequently updating target dynamics (at intervals of Δt = 30 seconds) to avoid wasting ammunition on false targets. Strike missions are directly embedded in the real-time data of persistent reconnaissance, ensuring that firepower attacks are based on the latest battlefield situation. The rapid relay of wide-area reconnaissance and persistent reconnaissance (Δt = 30 seconds) reduces the strike decision cycle to less than 60 seconds, significantly faster than the reaction time of enemy defense systems.

[0041] Step 4.2: After selecting the high-utility mission modes from step 4.1, adjust the unmanned platform tasks proposed in Table 1, give priority to the high-utility mission modes among the modes, and finally optimize the formation configuration according to the modes.

[0042] After specific actual experiments, under the specific environment setting of anti-threat target interception by maritime formations, the above method was started with the parameters in Table 2 and Table 3, and the experimental data that can be obtained are shown in Table 4. Figure 5 As shown in the figure, the optimized mission plan can reduce the ship damage rate by 18.3%. This method can effectively extract the winning rules from historical engagement data and produce statistically significant improvements in the anti-threat target interception combat effectiveness indicators.

[0043] Table 2

[0044] Table 3

[0045] Table 4

[0046] Compared to existing technologies, this invention achieves a deep integration of spatiotemporal and utility multidimensional features in simulation data through an improved cross-linked list data structure and a dynamic two-layer pruning strategy. This solves the problems of static temporal rules and fragmented multidimensional features in traditional methods. Furthermore, a closed-loop verification mechanism based on Monte Carlo simulation ensures that the mined high-utility patterns are both theoretically sound and environmentally adaptable. Compared to single-dimensional association rule mining or static task planning methods, this invention provides interpretable and highly robust technical support for intelligent scheduling of unmanned platform swarms in complex battlefield environments.

[0047] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principles and purpose of the present invention. The scope of protection of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. All implementation schemes within its scope shall be subject to the constraints of the present invention.

Claims

1. A multi-unmanned platform tactical planning optimization method based on pattern mining, characterized in that: After generating multi-scenario tactical data through the simulation platform, its utility indicators are calculated and a time series database containing location information is constructed. After data structure optimization and pruning and filtering of the time series database, a depth-first search is used to extract the task sequence. Based on the task sequence, the unmanned platform mount is adjusted and the formation configuration is optimized.

2. The multi-unmanned platform tactical planning optimization method based on pattern mining according to claim 1 is characterized in that: The multi-scenario tactical data is generated by a high-fidelity simulation platform, and the spatiotemporal trajectory, mission time window and utility indicators of the unmanned platform are collected and position discretization preprocessing is performed to obtain the multi-scenario tactical data.

3. The multi-unmanned platform tactical planning optimization method based on pattern mining according to claim 2 is characterized in that: The high-fidelity simulation platform simulates and collects time-space data for each operation after presetting differentiated tactical plans. The preset differentiated tactical plans include: concentrated penetration type: enemy defense radius R = 120km, formation configuration is a close formation with a diamond spacing of ≤50m, and a synchronous penetration strategy of Δt <5s; distributed area control type: R = 150km, formation configuration is a tiered configuration with a depth of 300m, and an asynchronous penetration strategy of Δt = 30s; dynamic reorganization type: R = 100km, formation configuration is a variable star with a spacing of ≥200m, and a phased trigger strategy of Δt = 60s is adopted, and enemy parameters are set: electromagnetic interference intensity J ~ N (0.7, 0.1²), target maneuvering speed v∈{5, 10, 15} m / s; The spatiotemporal data includes: time data on the sequence and duration of mission execution, location data on the formation configuration and deployment density of unmanned platforms, and data quantifying the comprehensive effectiveness and cost efficiency of the unmanned platforms in the mission; The aforementioned sequence is the sequence of task execution with a synchronous penetration of Δt < 5s; The formation configuration is a staggered configuration and a star formation.

4. The multi-unmanned platform tactical planning optimization method based on pattern mining according to claim 2 is characterized in that: The position discretization preprocessing refers to discretizing the position data in the collected spatiotemporal data, that is, dividing the continuous coordinates into 500m×500m grids and mapping them into discrete position labels.

5. The multi-unmanned platform tactical planning optimization method based on pattern mining according to claim 1 is characterized in that: The utility indicators are: ,in: Indicates the mount performance. Indicates the weight of the mount type; Indicates the effectiveness of the communication network; represents the total cost, ; Represents the weight coefficient, which can be adjusted according to needs.

6. The multi-unmanned platform tactical planning optimization method based on pattern mining according to claim 1 is characterized in that: The time series database is obtained by integrating time data, discretized location data and utility data.

7. The multi-unmanned platform tactical planning optimization method based on pattern mining according to claim 1 is characterized in that: The data structure optimization is to construct a cross-link table-type data structure including a head node, a linked list node, and a time interval node, so as to achieve efficient storage of spatiotemporal-utility multi-dimensional features and support dynamic pruning; The head node represents the first item and the possible relationships; The project refers to the object to be mined, that is, a certain unmanned platform mission; The possible relationship mentioned refers to the execution relationship between projects; The linked list nodes represent the linked items, the relationship between the current item and the previous item, the current utility occupation (uo) and the remaining utility occupation (ruo); The time interval node includes a marker bit and the start and end time of the two items.

8. The multi-unmanned platform tactical planning optimization method based on pattern mining according to claim 1 is characterized in that: The pruning and filtering process mentioned above refers to: by performing support pruning and utility pruning in sequence, the algorithm computational complexity is significantly reduced and the pattern mining efficiency is improved, wherein: when performing support pruning, a minimum support threshold is set, and when performing utility pruning, low-value candidate patterns are filtered out to ensure that only high-reward task sequences are retained.

9. The multi-unmanned platform tactical planning optimization method based on pattern mining according to claim 1 is characterized in that: The depth-first search refers to: using a spatiotemporal pattern mining algorithm to traverse an improved cross-linked list to extract patterns that meet support and utility constraints; adjusting the unmanned platform tasks based on the screened high-utility task patterns, giving priority to the high-utility task patterns in the patterns, and finally optimizing the formation configuration based on the patterns.

Images