Multi-target task and resource intelligent modeling method

Abstract

The invention discloses a multi-target task and resource intelligent modeling method, particularly relates to the field of complex adversarial simulation, is used for solving the problems of dynamic constraint optimization and robustness improvement in multi-dimensional task planning, and aims at realizing multi-dimensional coupling of space-time resource parameters by constructing a three-dimensional hypergraph model, mining a parameter association rule by means of tensor decomposition, and realizing multi-dimensional optimization of the space-time resource parameters. Dynamic constraint quantization is supported, multi-dimensional index priority evaluation is fused in a constraint layered injection stage, hard constraints are recognized, a solution domain is compressed, a hybrid optimization strategy regulates and controls balance between global exploration and local optimization, and after annealing is simulated to jump out of a local extreme value, a multi-target particle swarm algorithm is used for screening a space-time resource equilibrium solution in a trimming solution domain. Digital twinborn verification promotes physical and virtual space interaction data closed loop, a parameter correlation degree matrix is corrected, scheme robustness is enhanced, efficient generation and adaptive optimization of a task planning scheme under complex constraints are realized, and system stability and multi-target cooperation capability under sudden disturbance are improved.

Description

Technical Field

[0001] The present invention relates to the field of complex confrontation simulation, and more specifically, to a multi-objective task and resource intelligent modeling method. Background Art

[0002] In the task planning of modern complex confrontation simulation systems, various execution units need to meet the requirements of multi-dimensional collaborative constraints at the same time. Task planning involves the coordination of parameters in multiple dimensions, such as numerical indicators (such as energy consumption rate), time windows (such as task start and end periods), spatial paths (such as navigation trajectories), resource collections (such as collaborative unit groups), etc. There is a strong coupling relationship between these parameters. For example, the spatial movement path of a unit directly affects its task execution time, and the time arrangement determines the order of resource use. Existing technologies usually adopt a modular and independent processing method to decompose elements such as spatiotemporal parameters and resource allocation into different subsystems for separate calculations, and then check the feasibility of the solution through post-verification.

[0003] Traditional modeling methods have significant defects in dealing with multi-dimensional task planning: first, the decoupling design of spatiotemporal parameters and resource constraints leads to logical contradictions in planning schemes, such as the inability to automatically adapt the associated time window after the spatial path is adjusted; second, there is a lack of cross-dimensional dynamic association mechanisms. When encountering sudden changes in constraints (such as temporary no-travel zones), the subsystems cannot respond in a coordinated manner and need to be completely replanned; third, the combinatorial explosion problem of multi-dimensional parameters makes it difficult for conventional optimization algorithms to generate feasible solutions within a reasonable time. These problems seriously restrict the real-time and reliability of task planning systems in complex adversarial environments, and cannot meet the actual needs of multi-objective collaborative optimization in dynamic scenarios.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a multi-objective task and resource intelligent modeling method, which realizes multi-dimensional coupling of spatiotemporal resource parameters by constructing a three-dimensional hypergraph model, mines parameter association rules with the help of tensor decomposition, supports dynamic constraint quantification, integrates multi-dimensional indicator priority evaluation in the constraint layered injection stage, identifies hard constraints and compresses the solution domain, and uses a hybrid optimization strategy to regulate the balance between global exploration and local optimization. After simulated annealing jumps out of the local extreme value, a multi-objective particle swarm algorithm is used to screen the spatiotemporal resource equilibrium solution in the pruned solution domain. Digital twin verification promotes the closed loop of physical and virtual space interaction data, corrects the parameter correlation matrix and enhances the robustness of the solution, realizes efficient generation and adaptive optimization of task planning schemes under complex constraints, and improves the system stability and multi-objective coordination capability under sudden disturbances to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions: S1. By defining a three-dimensional hypergraph structure of task nodes, spatio-temporal coordinates, and resource identifiers, using tensor decomposition technology to extract the implicit features of spatio-temporal resources, generating a parameter correlation matrix, and realizing a unified mathematical expression of cross-dimensional constraint relationships; S2. Based on a constraint priority evaluation model that fuses multi-dimensional indicators, first inject critical hard safety constraints into the solution space defined by the hypergraph model, and use the constraint propagation algorithm to dynamically prune the conflicting solution branches to form the boundary of the preliminary feasible solution domain; S3. Within the feasible solution domain pruned by the constraint hierarchical injection mechanism, first use the simulated annealing algorithm for global exploration, and then start the multi-objective particle swarm optimization algorithm to screen for spatio-temporal resource equilibrium solutions, and output a set of candidate solutions that meet multi-dimensional constraints; S4. Map the set of candidate solutions generated by the hybrid intelligent optimization engine to the digital twin simulation environment, evaluate the dynamic adaptability through Monte Carlo simulation, correct the parameter correlation matrix according to the feedback results, and iteratively optimize the solutions until the preset threshold is met.

[0007] In a preferred embodiment, step S1 includes the following: By taking task nodes, spatio-temporal coordinates, and resource identifiers as the basic elements of the hypergraph, construct a three-dimensional hypergraph structure, where the hyperedges represent the association relationships of tasks using specific resources at specific spatio-temporal points; transform the hypergraph into a four-dimensional adjacency tensor, including four dimensions of tasks, space, time, and resources, and perform discretization processing on the space and time dimensions; use the CP decomposition technology to reduce the dimension of the adjacency tensor, extract the implicit feature vectors of tasks, space, time, and resources, and ensure the accuracy of the decomposition by optimizing the reconstruction error; based on the implicit feature vectors, calculate the correlation degree between tasks and resources in the spatio-temporal environment, generate a parameter correlation matrix, and realize a unified mathematical expression of cross-dimensional constraint relationships.

[0008] In a preferred embodiment, step S2 includes the following: S2-1. The multi-dimensional indicators include task dependence chaos degree and resource competition concentration degree; by synthesizing the task dependence chaos degree and resource competition concentration degree, generate a priority coefficient for subsequent constraint injection; the task dependence chaos degree quantifies the non-linear complexity of the dependence relationships between tasks. Based on the three-dimensional hypergraph structure, extract the task dependence graph, embed it into a two-dimensional grid, and define the side length of the grid cell as ; at different scales , calculate the minimum number of boxes required to cover the task dependence graph, and through the linear fitting slope of , obtain the fractal dimension , which reflects the chaos degree of the dependence graph; the final task dependence chaos degree is calculated as: , where is the number of edges in the task dependency graph, is the total number of tasks.

[0009] In a preferred embodiment, the resource contention concentration measures the imbalance of resource competition and uses the correlation matrix to extract the task About Resources The intensity of competition , calculate each resource The probability of competition , and then calculate the competitive entropy , the resource contention concentration is defined as: (normalized to [0,1]), is the number of resource types, For the task About Resources The intensity of competition, For the task About Resources The intensity of competition, For resources The competitive entropy of For resources The competitive entropy of Is an index that traverses all resource types, from 1 to ; The priority coefficient is obtained by weighted summation of task dependency chaos and resource contention concentration.

[0010] In a preferred embodiment, S2-2 identifies and injects critical hard safety constraints into the solution space based on sub-priority coefficient values; sorts the constraints according to the priority coefficients, defines the constraints with priority coefficient values higher than a preset threshold as critical constraints, maps these constraints to a three-dimensional hypergraph structure, updates the parameter correlation matrix, and imposes restrictions.

[0011] In a preferred embodiment, S2-3 utilizes the injected constraints to dynamically prune the solution space through a constraint propagation algorithm to generate a preliminary feasible solution domain boundary; using the full set of three-dimensional hypergraph models as the initial solution space, check one by one whether each solution branch satisfies the injected constraints, remove branches that do not satisfy the constraints, and detect potential conflicts in advance through a propagation mechanism. After pruning, the remaining solution branches constitute a preliminary feasible solution domain.

[0012] In a preferred embodiment, step S3 includes the following contents: First, within the boundary of the feasible solution domain pruned by the constraint hierarchical injection mechanism, a simulated annealing algorithm with adaptive temperature scheduling and dynamically adjusted perturbation intensity is used for global exploration. Specifically, the temperature decay rate is adjusted through a non-linear temperature decay mechanism, and the perturbation intensity is dynamically adjusted according to the quality of the solution, so as to balance breadth and depth during the exploration process, ensure effective coverage of the solution space, and avoid falling into local optima. Subsequently, based on the global exploration, a multi-objective particle swarm optimization algorithm with dynamic inertia weight and Pareto sorting mechanism is launched to screen for spatio-temporal resource balanced solutions within the pruned solution domain. Among them, the dynamic inertia weight is adaptively adjusted according to the optimization process to balance the characteristics of global exploration and local convergence. At the same time, the Pareto sorting mechanism is used to perform multi-objective sorting on the candidate solutions to maintain the diversity of the solution set under multi-dimensional constraints, thereby screening out high-quality balanced solutions. Finally, based on the above logical processing process of global exploration and multi-objective optimization screening, a candidate solution set that meets multi-dimensional constraints is output.

[0013] In a preferred embodiment, step S4 includes the following: First, by mapping the candidate solution set generated by the hybrid intelligent optimization engine to the digital twin simulation environment, a virtual platform highly consistent with the physical task execution environment is constructed, and dynamic data of task execution paths and resource allocations are collected in real time on the virtual platform to ensure that the virtual environment can accurately reflect the actual execution state.

[0014] In a preferred embodiment, step S4 further includes the following: Next, Monte Carlo simulation is used to evaluate the dynamic adaptability of each candidate solution under random perturbation scenarios, and a robustness scoring mechanism combining geometric mean and arithmetic mean is adopted to quantify the robustness of the solutions. Among them, the geometric mean is used to balance the influence of extreme values, and the arithmetic mean is used to reflect the overall performance, thereby generating a comprehensive score. Subsequently, based on the feedback results of Monte Carlo simulation, solutions with robustness scores lower than the preset threshold are screened out, their failure modes are analyzed, the correction coefficients of key task-resource pairs are calculated, and the association weights of key task-resource pairs in the parameter correlation matrix are reduced to dynamically update the parameter correlation matrix.

[0015] In a preferred embodiment, step S4 further includes the following: Finally, a new candidate solution set is iteratively generated based on the updated parameter correlation matrix, and the mapping, evaluation, and correction processes are repeated until the robustness of all solutions meets the preset threshold.

[0016] The technical effects and advantages of the multi-objective task and resource intelligent modeling method of the present invention: The present invention constructs a three-dimensional hypergraph model to establish a multi-dimensional coupling representation system for spatio-temporal resource parameters, and mines the implicit association rules of cross-dimensional parameters based on tensor decomposition technology, providing a quantitative basis for dynamic constraint processing; in the constraint hierarchical injection stage, through the fusion of the priority evaluation model of multi-dimensional indicators, intelligent identification of hard constraints and directional compression of the solution space are realized, forming a solution domain boundary that takes into account both safety and feasibility; the hybrid optimization strategy establishes a dynamic balance mechanism between global exploration and local optimization. After breaking through the local extreme value trap by simulated annealing, the multi-objective particle swarm algorithm is used to extract the optimal solution set with balanced spatio-temporal resources within the trimmed solution domain; the digital twin verification link feeds back the interaction data of the physical space and the virtual space to the hypergraph model in a closed loop, driving the dynamic correction of the parameter correlation matrix and the iterative enhancement of the robustness of the solution. Finally, through the synergistic effects of four levels: multi-dimensional parameter coupling analysis, solution space directional compression, multi-objective balanced optimization, and dynamic environment adaptability, the efficient generation of task planning solutions under complex constraints and the continuous adaptive optimization in a dynamic environment are realized, significantly improving the stability of the system under sudden constraint disturbances and the multi-objective collaborative optimization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic flow chart of the multi-objective task and resource intelligent modeling method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] Embodiment 1: Figure 1 The multi-objective task and resource intelligent modeling method of the present invention is given, including: S1. By defining the three-dimensional hypergraph structure of task nodes, spatio-temporal coordinates, and resource identifiers, using tensor decomposition technology to extract the implicit features of spatio-temporal resources, generating a parameter correlation matrix, and realizing the unified mathematical expression of cross-dimensional constraint relationships; S2. Based on the constraint priority evaluation model that fuses multi-dimensional indicators, first inject key hard safety constraints into the solution space defined by the hypergraph model, and use the constraint propagation algorithm to dynamically prune the conflicting solution branches to form a preliminary feasible solution domain boundary; S3. In the feasible solution domain trimmed by the constraint hierarchical injection mechanism, first use the simulated annealing algorithm for global exploration, and then start the multi-objective particle swarm optimization algorithm to screen the balanced solutions of spatio-temporal resources, and output a candidate solution set that satisfies multi-dimensional constraints; S4. Map the candidate solution set generated by the hybrid intelligent optimization engine to the digital twin simulation environment, evaluate the dynamic adaptability through Monte Carlo simulation, correct the parameter correlation matrix according to the feedback results, and iteratively optimize the solution until the preset threshold is met.

[0020] For the sake of understanding, the following example is given: In the large-scale cross-regional drill simulation between the two parties, the command agency of Party A is responsible for coordinating multiple action units to perform reconnaissance, strike, and defense tasks at different times and locations. Party B tests the adaptability and stability of Party A's mission planning by simulating dynamic interference and resource competition behaviors. Party A needs to formulate a detailed planning scheme based on task nodes (such as specific task objectives), spatio-temporal coordinates (such as the start time of reconnaissance and the action window), and resource identifiers (such as unmanned equipment groups and communication bandwidth). However, Party B may take temporary restrictive measures, such as restricting the operation area or reducing resource availability, which significantly increases the uncertainty of the mission execution environment.

[0021] In addition, there are strong coupling dependencies between tasks. For example, the completion of the reconnaissance task directly determines the start time of subsequent actions. At the same time, resource allocation is strictly restricted. Action units need to reuse equipment among multiple tasks and maintain operation efficiency in the resource bottlenecks caused by Party B's interference. Traditional planning methods often perform poorly in dealing with spatio-temporal resource coupling and dynamic constraints. For example, they fail to adjust the subsequent plan affected by reconnaissance delays in a timely manner, or fail to fully consider the post-interference restrictions in resource allocation, resulting in the failure of the plan in actual execution.

[0022] The present invention uniformly represents the coupling relationship of tasks, spatio-temporal, and resources through a three-dimensional hypergraph model, uses tensor decomposition to mine implicit associations, fuses the task dependence chaos degree and resource competition concentration degree to evaluate key constraints, adopts a hybrid optimization strategy to generate candidate solutions, and verifies its robustness under Party B's interference through digital twin simulation, ultimately achieving efficient and stable mission planning for Party A in complex confrontation scenarios.

[0023] In the mission planning of modern complex confrontation simulation systems, there are strong coupling relationships among multi-dimensional parameters such as tasks, time, space, and resources. Traditional methods often independently process each module, resulting in logical contradictions in the planning scheme and being unable to effectively respond to sudden constraint changes. Therefore, it is crucial to construct a mathematical model that can uniformly represent the coupling relationship of multi-dimensional parameters. By defining the three-dimensional hypergraph structure of task nodes, spatio-temporal coordinates, and resource identifiers, these cross-dimensional association relationships can be modeled in a unified manner, providing a structured data basis for subsequent dynamic constraint processing and solution space optimization.

[0024] Step S1 includes the following content: S1-1, Define the three-dimensional hypergraph structure: First, each task is modeled as a node in a hypergraph, and a spatio-temporal coordinate is assigned to each task, which consists of a three-dimensional spatial position and a timestamp. At the same time, a unique identifier is assigned to each type of resource to distinguish different resource types or instances. Hyperedges in the hypergraph are used to represent the association relationships among tasks, spatio-temporal coordinates, and resources, specifically manifested as a triple, which is used to describe that a certain task uses a certain resource at a specific spatio-temporal point.

[0025] S1-2, construct the adjacency tensor of the hypergraph: The hypergraph is represented as a four-dimensional adjacency tensor, which contains four dimensions: tasks, space, time, and resources. For ease of calculation, the space dimension is discretized into multiple grid points, and the time dimension is discretized into multiple time steps. Each element of the tensor is assigned a value according to the existence of the corresponding hyperedge: if there is a hyperedge connecting a specific task, a specific spatio-temporal point, and a certain resource, the element is assigned a value of 1; if there is no such hyperedge, the value is assigned 0.

[0026] S1-3, tensor decomposition to extract latent features: The CP decomposition technique is used to perform dimensionality reduction on the adjacency tensor, and it is approximately decomposed into the outer product sum of multiple factor vectors. Specifically, four factor matrices are generated through decomposition, corresponding to the latent feature vectors of tasks, space, time, and resources respectively. To ensure the accuracy of the decomposition result, the tensor reconstruction error is minimized through an optimization process, so that the decomposition result can accurately reflect the information in the original tensor.

[0027] S1-4: Generate the parameter correlation matrix: Based on the obtained latent feature vectors, the correlation degree between tasks and resources in the spatio-temporal environment is calculated. For each combination of a pair of tasks and resources, the calculation process of the correlation degree is as follows: First, on each latent feature dimension, calculate the product of the task latent feature component and the resource latent feature component; then, sum all the components of the space latent feature vector and the time latent feature vector on this dimension respectively to obtain the cumulative effects of space and time; then, multiply the product of the task and the resource by the cumulative effects of space and time to obtain the contribution value of this dimension; finally, sum the contribution values of all latent feature dimensions to obtain the final correlation degree between the task and the resource. The correlation degree values of all tasks and resources form the parameter correlation matrix, which is used to quantify the association strength between tasks and resources.

[0028] By using tensor decomposition technology to extract latent features from three-dimensional hypergraphs and generate a parameter correlation matrix, a unified mathematical expression of the cross-dimensional constraint relationships among tasks, space-time, and resources is achieved. This process not only reveals the implicit correlation laws among multi-dimensional parameters but also provides a quantitative basis for dynamic constraint handling in task planning, overcoming the logical contradiction problems caused by parameter decoupling in traditional methods. The generation of the parameter correlation matrix provides crucial structural support for subsequent constraint hierarchical injection and solution space pruning, ensuring the feasibility and efficiency of task planning solutions under complex constraints.

[0029] In step S1, by constructing a three-dimensional hypergraph structure composed of task nodes, space-time coordinates, and resource identifiers and using tensor decomposition technology to extract latent features, a parameter correlation matrix is formed, laying the foundation for the mathematical expression of cross-dimensional constraint relationships. Based on this parameter correlation matrix, step S2 designs a constraint priority evaluation model that fuses multi-dimensional indicators, aiming to intelligently identify critical hard safety constraints and dynamically prune the solution space through a constraint propagation algorithm, providing a safe and efficient preliminary feasible solution domain boundary for subsequent task resource optimization scheduling.

[0030] In the priority analysis of task planning, selecting the task dependence chaos degree and resource competition concentration degree as multi-dimensional indicators is based on their unique ability to capture system complexity and uncertainty, ensuring the scientificity and efficiency of task scheduling.

[0031] First, the task dependence chaos degree quantifies the non-linear complexity of the dependence relationships among tasks through methods such as fractal dimensions, and can effectively reflect the potential chaos degree of task execution order. In complex adversarial simulation systems, the strong coupling dependence relationships among tasks often exhibit highly non-linear characteristics, and small local changes may spread through the dependence chain, triggering global execution risks, namely the so-called butterfly effect. Therefore, incorporating the task dependence chaos degree into the priority analysis can identify high-risk task dependence relationships and prioritize their processing to reduce system uncertainty and potential failure probability, thus ensuring the stability and reliability of task planning.

[0032] Second, the resource competition concentration degree evaluates the imbalance of resource competition and reveals bottlenecks and conflict points in resource allocation through indicators such as competition entropy and improved Herfindahl-Hirschman index. In scenarios with limited resources, some key resources may become constraints on system performance due to concentrated competition among multiple tasks. Ignoring the concentration of resource competition may lead to unbalanced resource allocation, thereby affecting the efficiency and fairness of task execution. Taking the resource competition concentration degree as an analysis indicator can prioritize the identification and optimization of the use of these key resources, ensuring the efficiency of resource allocation and the improvement of the overall system performance.

[0033] Combining the above two points, the task dependence chaos degree and the resource contention concentration respectively start from the complexity of task execution and the conflict of resource allocation, forming a dual consideration of the key constraints in task planning. The selection of such multi-dimensional indicators can not only intelligently identify the key tasks and resources involving hard safety constraints in the system, but also improve the adaptability and stability of the system through priority injection in a dynamic environment.

[0034] Step S2 includes the following: S2-1, a constraint priority evaluation model that integrates multi-dimensional indicators: The multi-dimensional indicators include the task dependence chaos degree and the resource contention concentration; by synthesizing the task dependence chaos degree and the resource contention concentration, a priority coefficient is generated for subsequent constraint injection. The task dependence chaos degree quantifies the non-linear complexity of the dependence relationship between tasks. Based on the three-dimensional hypergraph structure in step S1, a task dependence graph (with nodes as tasks and edges as dependence relationships) is extracted and embedded in a two-dimensional grid. The side length of the grid cell is defined as (The initial value can be set as the average spatio-temporal distance between tasks). At different scales , calculate the minimum number of boxes required to cover the task dependence graph , and through the linear fitting slope of , obtain the fractal dimension (range [1, 2]), reflecting the chaos degree of the dependence graph. The final task dependence chaos degree is calculated as: , where is the number of edges in the task dependence graph (total number of dependence relationships), is the total number of tasks (defined by step S1).

[0035] The resource contention concentration measures the imbalance of resource competition. Using the parameter correlation matrix in step S1, extract the competition intensity of task for resource , calculate the competition probability of each resource , and then calculate the competition entropy . The resource contention concentration is defined as: (normalized to [0, 1]), is the number of resource types, is the competition intensity of task for resource , is the competition intensity of task for resource , is the competition entropy of resource , is the competition entropy of resource ​ Is an index that traverses all resource types, from 1 to .

[0036] The priority coefficient is obtained by weighted summation of task dependency chaos and resource contention concentration, and the two indicators are combined to calculate: , the higher the priority coefficient value, the higher the constraint priority. The weight of the task dependency chaos degree ranges from [0,1], reflecting the impact of task dependency complexity on priority; The weight of resource contention concentration ranges from [0,1], reflecting the impact of resource competition imbalance on priority. The weight can be obtained through expert scoring, historical data analysis, or machine learning model training.

[0037] S2-2, injection of key hard safety constraints: Identify and inject critical hard safety constraints into the solution space based on the sub-priority coefficient values. Sort the constraints according to the priority coefficients and define the constraints with priority coefficient values higher than the preset threshold as critical constraints, such as resource over-allocation and time window restrictions. The resource over-allocation constraint requires that resources The allocation does not exceed the capacity (The resource attributes are defined by step S1), the time window constraint requires the task Completed within the time range defined by the space-time coordinates. Map these constraints to the three-dimensional hypergraph structure of step S1, update the parameter association matrix, and impose restrictions. For example, for resource constraints, mark The upper limit ensures ; For time constraints, adjust the time coordinate boundaries of the task. Ensure that high-priority constraints are integrated into the solution space first, providing a basis for subsequent pruning. Parameter explanation: For resources capacity.

[0038] S2-3, dynamic pruning of constraint propagation algorithm: Using the injected constraints, the solution space is dynamically pruned through the constraint propagation algorithm to generate the initial feasible solution domain boundary. Taking the full set of 3D hypergraph models as the initial solution space, each solution branch is checked one by one to see if it satisfies the injected constraints. For example, verify whether the resource allocation exceeds , or whether the task time meets the window limit. For branches that do not meet the constraints (such as ) are removed, and potential conflicts (such as task dependency failure caused by resource over-allocation) are detected in advance through the propagation mechanism. After pruning, the remaining solution branches constitute the preliminary feasible solution domain, which serves as the input of step S3.

[0039] Based on the parameter correlation matrix of step S1, step S2 intelligently identifies and injects key hard safety constraints into the three-dimensional hypergraph model by calculating the task dependency chaos and resource contention concentration, and uses the constraint propagation algorithm to prune the solution space, forming a safe and efficient preliminary feasible solution domain, which provides a reliable boundary for the subsequent task resource optimization scheduling.

[0040] In step S2, by integrating the priority evaluation model of task dependency chaos and resource contention concentration, key hard safety constraints have been injected first, and the hypergraph model solution space defined by step S1 has been pruned using the constraint propagation algorithm to generate a preliminary feasible solution domain boundary. Based on this feasible solution domain boundary, step S3 designs a hybrid intelligent optimization strategy, uses the simulated annealing algorithm for global exploration, and combines the multi-objective particle swarm optimization algorithm to screen the spatiotemporal resource balance solution, and generates a set of candidate solutions that meet multi-dimensional constraints, providing high-quality initial input for the digital twin simulation verification in step S4.

[0041] Step S3 includes the following contents: S3-1, Global Exploration of Simulated Annealing Algorithm: First, an initial solution is randomly selected from the boundary of the feasible solution domain generated in step S2, and the initial temperature, annealing rate and maximum number of iterations are set. The temperature decreases nonlinearly with the increase of the number of iterations. Specifically, the temperature of each step is calculated by multiplying the initial temperature by a coefficient that varies with the number of iterations. The coefficient is obtained by subtracting a certain power of the ratio of the current number of iterations to the maximum number of iterations from 1, where the exponent of the power is a decay parameter greater than 1, so as to ensure that the temperature decreases slowly in the early stage and decreases faster in the later stage to converge quickly. At the same time, the perturbation intensity is dynamically adjusted according to the objective function value of the current solution. The calculation method is: the perturbation intensity is equal to the initial perturbation intensity multiplied by a factor. The factor is obtained by adding 1 to the difference between the objective function value of the current solution and the objective function value of the historical optimal solution divided by the difference between the objective function value of the historical worst solution and the optimal solution, so that the solution with a poor objective function value obtains a larger perturbation amplitude to help it jump out of the range of the local optimal solution. In each iteration, a new solution is generated by perturbation, and the difference between the objective function value of the new solution and the objective function value of the current solution is calculated. If the difference is less than zero, the new solution is directly accepted as the current solution; if the difference is greater than or equal to zero, a probability calculated based on the current temperature and the difference is used to decide whether to accept the new solution. The algorithm terminates when the preset maximum number of iterations is reached or when multiple consecutive iterations do not improve the historical optimal solution.

[0042] S3-2, Screening of multi-objective particle swarm optimization algorithm: Randomly generate a fixed number of particles within the boundary of the feasible solution domain in step S2. Each particle contains information about its position and velocity. The inertia weight is dynamically adjusted as the number of iterations increases. The specific calculation method is as follows: The inertia weight starts from a relatively large initial value and gradually decreases to a relatively small value as the ratio of the square of the number of iterations to the square of the maximum number of iterations. This is to achieve a balance between extensive exploration of the solution space in the early stage and rapid convergence in the later stage. Then, non-dominated sorting is performed on all particles to generate a Pareto front, which is used to evaluate the superiority and inferiority relationships of multiple objectives, such as indicators like spatio-temporal resource balance and constraint satisfaction. A global optimal solution is randomly selected from this Pareto front to ensure the diversity of the candidate solution set. In each iteration step, the velocity update of the particle is calculated as follows: Multiply the inertia weight by the current velocity, and then add the attraction effects of the particle's individual optimal solution and the global optimal solution, which are jointly controlled by the learning factor and a random number, to obtain a new velocity value. Subsequently, the position of the particle is updated by accumulating the current velocity. The algorithm terminates when the preset maximum number of iterations is reached, or when the solution set of the Pareto front converges and no significant changes occur. Finally, a set of candidate solutions that meet the multi-dimensional constraint conditions is output.

[0043] In step S3, a set of candidate solutions that meet multi-dimensional constraints is generated from the feasible solution domain pruned by the constrained hierarchical injection mechanism through the simulated annealing algorithm and the multi-objective particle swarm optimization algorithm. The task of step S4 is to map this set of candidate solutions to the digital twin simulation environment, evaluate the adaptability of each solution under dynamic perturbations using Monte Carlo simulation, and correct the parameter correlation matrix generated in step S1 based on the simulation feedback. Through iterative optimization, ensure that the solutions meet the preset thresholds. This step focuses on enhancing the dynamic stability and robustness of the task planning solution in complex adversarial scenarios through the closed-loop feedback between virtual simulation and the physical environment.

[0044] Step S4 includes the following: S4-1, Digital twin mapping of the candidate solution set: First, use digital twin technology to create a virtual simulation platform that is highly consistent with the actual task execution environment. Based on the three-dimensional hypergraph structure composed of defined task nodes, spatio-temporal coordinates, and resource identifiers, this virtual simulation platform can simulate the dynamic interactions between tasks, changes in spatio-temporal positions, and real-time resource allocation. Next, map the candidate solution set, that is, a set of task planning solutions, one by one into this virtual simulation platform. Each task planning solution generates a corresponding task execution path and resource allocation strategy in the virtual environment. During the virtual execution process, key performance indicators of each task planning solution are collected in real time, such as the time required to complete the task, the total amount of resource consumption, and the number of spatio-temporal conflicts that occur. These collected key performance indicators are recorded as the performance data set of each task planning solution for subsequent analysis and evaluation.

[0045] S4-2, Dynamic Adaptability Evaluation of Monte Carlo Simulation: Evaluate the dynamic adaptability of each task planning solution under random perturbations through the Monte Carlo simulation method. Random perturbations include fluctuations in resource availability or delays in task execution. First, based on historical data, generate multiple sets of random perturbation scenarios, each set of scenarios containing random changes in resource and task parameters. Then, for each task planning solution, run the virtual simulation under each set of random perturbation scenarios and record the performance indicators of the solution under the corresponding scenario, such as task completion rate and resource utilization rate. Subsequently, calculate the dynamic adaptability score of each task planning solution. The specific calculation method is as follows: process the performance indicators of each task planning solution under all perturbation scenarios. First, calculate the ratio of the value of each performance indicator to a preset reference value (i.e., the ideal or benchmark value), and then take the geometric mean of the ratios of all performance indicators to obtain the comprehensive performance score under this perturbation scenario; finally, take the arithmetic mean of the comprehensive performance scores of all perturbation scenarios to obtain the dynamic adaptability score of this task planning solution.

[0046] S4-3, Correction of Parameter Correlation Matrix and Iterative Optimization of Solutions: First, based on the feedback results of Monte Carlo simulation, select the task planning schemes with dynamic adaptability scores lower than the preset threshold, and analyze the failure modes of these schemes in the perturbation scenario to determine the key tasks and resource pairs that lead to failure. Next, calculate the correction coefficient for each key task and resource pair. The method is as follows: for each failure scenario, count the number of failures of this task and resource pair in this scenario, divide it by the total number of failures in this scenario to obtain the failure ratio in this scenario; then, take the average of the failure ratios of all failure scenarios to get the average failure ratio of this task and resource pair; finally, multiply this average failure ratio by a preset learning rate to calculate the correction value of the correlation degree of this task and resource pair. After that, update the parameter correlation matrix. The specific operation is: subtract this correction value from the correlation degree value corresponding to this task and resource pair in the original matrix to reduce its correlation weight. After the correction is completed, based on the updated matrix, re-run the constraint propagation algorithm and the optimization algorithm to generate a new set of candidate solutions, and repeatedly execute the above mapping, evaluation, and correction steps until the dynamic adaptability scores of all solutions reach the preset threshold or the number of iterations reaches the upper limit.

[0047] By mapping the set of candidate solutions generated in step S1 to the digital twin simulation environment and combining Monte Carlo simulation to evaluate the performance of the solutions under random perturbations, step S4 realizes the quantitative analysis of dynamic adaptability, and corrects the parameter correlation matrix through closed-loop feedback, iteratively optimizing the solutions to meet the robustness requirements. This process significantly enhances the stability and adaptability of the task planning scheme in complex adversarial environments, while ensuring the continuous optimization of parameters under multi-dimensional constraints, providing the system with efficient and reliable planning capabilities.

[0048] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0049] It should be noted that the system of the present invention can be deployed on the device itself to achieve embedded applications, or can also run on a PC or other terminals with a user interface, so as to meet various hardware environments and usage requirements.

[0050] Only some exemplary embodiments of the present invention have been described above by way of illustration. Undoubtedly, for those of ordinary skill in the art, without departing from the spirit and scope of the present invention, the described embodiments can be modified in various different ways. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the protection scope of the claims of the present invention.

[0051] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0052] As described above, the above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should be covered by the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. An intelligent modeling method for multi-objective tasks and resources, characterized in that Includes steps: S1. By defining the three-dimensional hypergraph structure of task nodes, spatiotemporal coordinates, and resource identifiers, tensor decomposition technology is used to extract the hidden characteristics of spatiotemporal resources, generate a parameter correlation matrix, and realize a unified mathematical expression of cross-dimensional constraint relationships; S2. Based on the constraint priority evaluation model integrating multi-dimensional indicators, key hard safety constraints are preferentially injected into the solution space defined by the hypergraph model, and the constraint propagation algorithm is used to dynamically prune conflicting solution branches to form a preliminary feasible solution domain boundary; S3. In the feasible solution domain pruned by the constraint hierarchical injection mechanism, the simulated annealing algorithm is first used for global exploration, and then the multi-objective particle swarm optimization algorithm is started to screen the spatiotemporal resource balance solution, and the candidate solution set that meets the multi-dimensional constraints is output; S4. Map the candidate solution set generated by the hybrid intelligent optimization engine to the digital twin simulation environment, evaluate the dynamic adaptability through Monte Carlo simulation, modify the parameter correlation matrix according to the feedback results, and iterate the optimization solution until the preset threshold is met.

2. The multi-objective task and resource intelligent modeling method according to claim 1, wherein Step S1 includes the following contents: A three-dimensional hypergraph structure is constructed by taking task nodes, spatiotemporal coordinates and resource identifiers as the basic elements of the hypergraph, in which hyperedges represent the association between tasks using specific resources at specific spatiotemporal points; the hypergraph is converted into a four-dimensional adjacency tensor, which includes four dimensions: task, space, time and resources, and the space and time dimensions are discretized; the CP decomposition technology is used to reduce the dimension of the adjacency tensor, extract the latent eigenvectors of tasks, space, time and resources, and ensure the accuracy of the decomposition by optimizing the reconstruction error; based on the latent eigenvectors, the association between tasks and resources in the spatiotemporal environment is calculated, and a parameter association matrix is generated to achieve a unified mathematical expression of cross-dimensional constraint relationships.

3. The multi-objective task and resource intelligent modeling method according to claim 2, wherein Step S2 includes the following contents: S2-1, multi-dimensional indicators include task dependency chaos and resource contention concentration; combining task dependency chaos and resource contention concentration to generate priority coefficients for subsequent constraint injection; The task dependence chaos quantifies the non - linear complexity of the dependencies between tasks. Based on a three - dimensional hypergraph structure, a task dependence graph is extracted and embedded into a two - dimensional grid. The side length of the grid cell is defined as ; At different scales , calculate the minimum number of boxes required to cover the task dependence graph. Through the linear fitting slope of , the fractal dimension is obtained, which reflects the chaos degree of the dependence graph. Finally, the task dependence chaos is calculated as: , where is the number of edges of the task dependence graph, is the total number of tasks.

4. The multi-objective task and resource intelligent modeling method according to claim 3 is characterized by: The resource contention concentration measures the imbalance of resource competition and extracts tasks using the correlation matrix for resources the competition intensity , calculate the competition probability of each resource and then calculate the competition entropy . The resource contention concentration is defined as: , , where \(n\) is the number of resource types, \(p_{ij}\) is the competition intensity of task \(i\) for resource \(j\), \(p_{kj}\) is the competition intensity of task \(k\) for resource \(j\), \(H_j\) is the competition entropy of resource \(j\), \(j\) is an index that traverses all resource types, from 1 to \(n\); The priority coefficient is obtained by weighted summation of task dependency chaos and resource contention concentration.

5. The multi-objective task and resource intelligent modeling method according to claim 4 is characterized in that: S2-2, based on the sub-priority coefficient values, identifies and injects critical hard safety constraints into the solution space; sorts the constraints according to the priority coefficients, defines the constraints with priority coefficient values higher than the preset threshold as critical constraints, maps these constraints to the three-dimensional hypergraph structure, updates the parameter correlation matrix, and imposes restrictions.

6. The multi-objective task and resource intelligent modeling method according to claim 5 is characterized by: S2-3, using the injected constraints, dynamically prune the solution space through the constraint propagation algorithm to generate the boundary of the preliminary feasible solution domain; using the full set of three-dimensional hypergraph models as the initial solution space, check whether each solution branch satisfies the injected constraints one by one, remove the branches that do not satisfy the constraints, and detect potential conflicts in advance through the propagation mechanism. After pruning, the remaining solution branches constitute the preliminary feasible solution domain.

7. The multi-objective task and resource intelligent modeling method according to claim 6, wherein Step S3 includes the following contents: First, within the boundary of the feasible solution domain pruned by the constraint hierarchical injection mechanism, a simulated annealing algorithm with adaptive temperature scheduling and dynamic adjustment of perturbation intensity is used for global exploration. Specifically, the temperature decay rate is regulated through a non-linear temperature decay mechanism, and the perturbation intensity is dynamically adjusted according to the quality of the solution, so as to balance breadth and depth during the exploration process, ensure effective coverage of the solution space and avoid falling into local optima. Subsequently, based on the global exploration, a multi-objective particle swarm optimization algorithm with dynamic inertia weight and Pareto sorting mechanism is launched to screen for spatio-temporal resource equilibrium solutions within the pruned solution domain. Among them, the dynamic inertia weight is adaptively adjusted according to the optimization process to balance the characteristics of global exploration and local convergence. At the same time, the Pareto sorting mechanism is used to perform multi-objective sorting on the candidate solutions to maintain the diversity of the solution set under multi-dimensional constraints, thereby screening out high-quality equilibrium solutions. Finally, based on the above logical processing process of global exploration and multi-objective optimization screening, a candidate solution set that meets multi-dimensional constraints is output.

8. The multi-objective task and resource intelligent modeling method according to claim 7, wherein Step S4 includes the following: First, by mapping the candidate solution set generated by the hybrid intelligent optimization engine to the digital twin simulation environment, a virtual platform highly consistent with the physical task execution environment is constructed, and dynamic data on task execution paths and resource allocation are collected in real time on the virtual platform to ensure that the virtual environment can accurately reflect the actual execution status.

9. The multi-objective task and resource intelligent modeling method according to claim 8, wherein Step S4 also includes the following: Next, Monte Carlo simulation is used to evaluate the dynamic adaptability of each candidate solution under random perturbation scenarios, and a robustness scoring mechanism combining geometric mean and arithmetic mean is adopted to quantify the robustness of the solution. Among them, the geometric mean is used to balance the influence of extreme values, and the arithmetic mean is used to reflect the overall performance, thereby generating a comprehensive score. Subsequently, based on the feedback results of the Monte Carlo simulation, solutions with robustness scores lower than the preset threshold are screened out, their failure modes are analyzed, the correction coefficients of key task-resource pairs are calculated, and the parameter correlation matrix is dynamically updated by reducing the correlation weights of key task-resource pairs in the parameter correlation matrix.

10. The multi-objective task and resource intelligent modeling method according to claim 9, wherein Step S4 also includes the following: Finally, a new candidate solution set is iteratively generated based on the updated parameter correlation matrix, and the mapping, evaluation, and correction processes are repeated until the robustness of all solutions meets the preset threshold.

Images