A decision-making method and system based on constraints and preferences

Abstract

This invention discloses a decision-making method and system based on constraints and preferences, relating to the field of system architecture. It includes: a construction module for selecting forward, reverse, or hybrid modeling links according to system architecture design requirements, parsing structured data from the DoDAF model, and extracting capability, operational, and resource entity information; and an extraction module for extracting original architectural constraints from four dimensions: structure, logic, resources, and timing, assigning a unique identifier to each constraint, and establishing a mapping table between constraint identifiers and constraint expressions. This invention can adapt to modeling requirements, hierarchically manage multi-dimensional constraints, standardize topology mounting and quantify decision preferences, quickly eliminate invalid components, avoid constraint conflicts, reduce the search space, and efficiently generate compliant and design-intent-aligned architectural solutions, improving design accuracy and efficiency, reducing costs and iteration cycles, and balancing rigid constraints with practical application needs.

Description

Technical Field

[0001] This invention relates to the field of system architecture technology, specifically to a decision-making method and system based on constraints and preferences. Background Technology

[0002] System architecture design is a core component of complex systems engineering, requiring the modeling and configuration of operational activities, system resources, and interrelationships under multiple capability constraints. The industry widely adopts the Department of Defense Architecture Framework (DoDAF) for standardized architecture description. Existing decision-making methods mainly fall into two categories: one is a manual decision-making method based on DoDAF, relying on expert experience to complete architecture screening and solution selection; the other is an automated method based on heuristic search, which transforms the design into an optimization problem and generates optimal or near-optimal architecture solutions through algorithms.

[0003] Existing related system architecture decision-making patents are mainly divided into two categories: one is based on DoDAF to carry out architecture modeling and manual decision-making, and only achieves constraint management through implicit rules; the other adopts heuristic search for architecture optimization, focusing on multi-objective solution and simple constraint handling.

[0004] Currently, existing technologies lack complete constraints and rely on repeated manual verification. Decision preferences cannot guide the search during the design phase, and there is a lack of a decision-making mechanism that allows constraints and preferences to work together, making it difficult to generate compliant system architecture solutions that align with decision-making intentions.

[0005] To address this, we propose a decision-making method and system based on constraints and preferences. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a decision-making method and system based on constraints and preferences, which can effectively solve the problems of the existing technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions; This invention discloses a decision-making system based on constraints and preferences, comprising: The system comprises the following modules: a construction module, which selects forward, reverse, or hybrid modeling links based on architecture design requirements, parses structured data from the DoDAF model, and extracts capability, operational, and resource entity information; an extraction module, which extracts original architecture constraints from four dimensions: structure, logic, resources, and time sequence, assigns a unique identifier to each constraint, and establishes a mapping table between constraint identifiers and constraint expressions; a topology generation module, which maps capabilities, operational activities, and resource entities to three types of standardized topology nodes, builds interconnected topology edges between nodes, and orients corresponding constraints to nodes or topology edges; a modeling module, which constructs five types of decision preferences: performance efficiency, cost resources, technology selection, resource utilization, and operational satisfaction, and transforms each preference into a corresponding computable mathematical expression; a reduction module, which performs static invalid component elimination based on formal constraint topology, verifies constraint conflicts in real time during the search process, and prunes search branches that violate constraints; and a solution generation module, which uses a heuristic search algorithm and a preference guidance function within the reduced decision space to generate candidate architecture solutions that satisfy all constraints. The output of the construction module is connected to the input of the extraction module via data communication. The output of the extraction module is connected to the input of the topology generation module via data communication. The output of the topology generation module is connected to the input of the reduction module via data communication. The output of the modeling module is connected to the first input of the scheme generation module via data communication. The output of the reduction module is connected to the second input of the scheme generation module via data communication.

[0008] Furthermore, during the runtime phase of the aforementioned building module, the completeness of the synchronous computing architecture design requirements is considered. and resource completeness Based on modeling link adaptability Dynamically select forward, reverse, or hybrid modeling links, and parse structured XML data according to the entity association rules of the DoDAF metamodel DM2 to extract full attribute information of capabilities, operational activities, and resource entities; The modeling link adaptability The calculation formula is: ; in, Demand-oriented weighting As a resource-oriented weight, ; Take the ratio of the clearly defined high-level capacity requirements to the total capacity requirements; Take the ratio of the number of underlying system resource information that has been obtained to the total number of resource types; when When selecting the forward modeling link, when When selecting the reverse modeling link, when Select the hybrid modeling link at that time.

[0009] Furthermore, the extraction module assigns priority weights to constraints in each dimension, calculating constraint priorities based on the severity of the consequences of constraint violations. And establish a mapping table between constraint identifiers and constraint expressions in descending order of priority; The constraint priority ; In the formula: As the influence weight, For trigger time weighting, ; To constrain the impact of violations; To constrain the trigger time of violations.

[0010] Furthermore, the topology generation module calculates the support weights of the topological edges connecting the nodes. And according to constraint priority The constraints are attached to the corresponding nodes or topological edges, and a constraint-topological association matrix is ​​generated. ; The topology edge support weight The calculation formula is: ; In the formula: , , , These are the attribute weights of the four types of topological edges. ; Capacity-activity support strength; Weighting for activity-resource support; This is the normalized value for resource-to-resource transmission bandwidth. This is the normalized value for the activity-activity sequence delay.

[0011] Furthermore, during the modeling module's runtime phase, a unified preference fusion function is constructed. The mathematical expressions of five types of decision preferences—performance efficiency, cost resources, technology selection, resource utilization, and job satisfaction—are weighted and fused to generate a computable preference guidance function. The preference fusion function The calculation formula is: ; In the formula: , , These are the fusion weights expressed as weighted, rule-based, and priority-based, respectively. ; It is a weighted evaluation function for performance efficiency and cost resources; A reward function for rules governing technology selection and resource utilization; This is the priority evaluation function that the task satisfies.

[0012] Furthermore, the efficiency of the reduction module in running the computing components Static invalid component removal is performed based on a validity threshold, and constraint conflict degree is calculated in real time during the search process. ,when Prune search branches that violate constraints. The effectiveness of the components The calculation formula is: ; In the formula: This represents the total number of capability items in the system architecture. For the first The supporting weight of each capability; For the component to the first The degree of support for the item's capabilities; when The component is then determined to be invalid. The degree of constraint conflict The calculation formula is: ; In the formula: This represents the total number of constraints in the system architecture. For the first Priority of item constraints; To constrain violations of the sign, when a constraint is violated... When the constraints are satisfied ; This is the preset conflict threshold.

[0013] Furthermore, the scheme generation module employs an improved A... Heuristic search algorithms are used to construct a comprehensive evaluation function that includes constraint satisfaction and preference scores. Within the reduced decision space, candidate system architectures that satisfy all constraints are generated. The comprehensive evaluation function The calculation formula is: ; In the formula: This represents the cumulative cost of the current architecture. For heuristic functions; The heuristic function The calculation formula is: ; In the formula: The constraint satisfaction of the current solution; The preference score for the current option; , These are the heuristic weights for constraint satisfaction and preference score, respectively. .

[0014] Furthermore, the scheme generation module also includes a Pareto optimization unit, used to perform Pareto optimal selection on the generated multiple candidate schemes and calculate the overall goodness of each scheme. Output ranking of overall excellence Candidate solutions; The overall excellence The calculation formula is: ; In the formula: , These are the goodness weights for constraint satisfaction and preference score, respectively. ; This represents the preset number of output schemes.

[0015] On the other hand, a decision-making method based on constraints and preferences includes: Based on the system architecture design requirements, forward, reverse, or hybrid modeling links are dynamically selected. The structured data of the DoDAF model is parsed to extract full attribute information for capabilities, operational activities, and resource entities. Original architectural constraints are extracted from four dimensions: structure, logic, resources, and time sequence. Unique identifiers are assigned to constraints, and priorities are calculated. A mapping table between constraint identifiers and expressions is established. Capabilities, operational activities, and resource entities are mapped to three types of standardized topology nodes. Node-related topology edges are built, and constraints are oriented according to their priority, generating a constraint-topology association matrix. Five types of decision preferences, including performance efficiency, cost resources, etc., are constructed and transformed into computable mathematical forms. A unified preference guidance function is generated through weighted fusion. Static invalid components are eliminated based on the constraint topology. Constraint conflicts are checked in real time during the search process, and search branches that violate constraints are pruned. An improved A A heuristic search algorithm generates candidate solutions that meet the constraints, and Pareto optimization is used to calculate the overall goodness and select the optimal solution.

[0016] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: This invention combines dynamic adaptation of system architecture modeling links, multi-dimensional constraint quantification and topology mounting, mathematical fusion of decision preferences, and intelligent reduction of the decision space. It can flexibly match modeling methods according to requirements and resource completeness, improving the adaptability and completeness of the architecture design. It can prioritize four types of constraints—structural, logical, resource, and temporal—and bind them to topology nodes and associated edges, achieving real-time detection and accurate verification of constraint conflicts. It transforms five types of decision preferences—performance, cost, and resources—into a unified and computable guiding function, making the decision-making process more aligned with design intent and practical application needs. By eliminating invalid components and pruning constraint conflicts, it effectively reduces the decision search space, lowering the computational cost and complexity of solution generation. Finally, it employs an improved heuristic search combined with Pareto optimization to generate system architecture solutions that satisfy all constraints and fit decision preferences, balancing compliance, rationality, and generation efficiency, providing stable and efficient decision support for system architecture design. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 This is a schematic diagram of the structure of a decision-making system based on constraints and preferences. Figure 2 This is a flowchart illustrating a decision-making method based on constraints and preferences. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] The present invention will be further described below with reference to embodiments.

[0021] Example 1: This embodiment presents a decision-making system based on constraints and preferences, such as... Figure 1 As shown, it includes: The building module is used to select forward, reverse, or hybrid modeling links according to the system architecture design requirements, parse the structured data of the DoDAF model, and extract capability, operational activities, and resource entity information; During the module runtime phase, the completeness of the synchronous computing architecture design requirements is assessed. and resource completeness Based on modeling link adaptability Dynamically select forward, reverse, or hybrid modeling links, and parse structured XML data according to the entity association rules of the DoDAF metamodel DM2 to extract full attribute information of capabilities, operational activities, and resource entities; When parsing DoDAF structured XML data, based on the entity association rules of the DM2 metamodel, the node fields of the three types of entities Capability, OperationalActivity, and Resource in the XML file are located, and the full attribute information such as entity ID, name, attributes, and association relationship is extracted; the capability entity is extracted through the CV-2 view, the operation activity entity is extracted through the OV-5b view, and the resource entity is extracted through the SV-1 and SV-2 views. Modeling Link Adaptability The calculation formula is: ; in, Demand-oriented weighting As a resource-oriented weight, ; Take the ratio of the clearly defined high-level capacity requirements to the total capacity requirements; Take the ratio of the number of underlying system resource information that has been obtained to the total number of resource types; The above formula obtains the fit value by weighted summation of requirement completeness and resource completeness. Based on the value range, it automatically matches the forward, reverse or hybrid modeling links and dynamically adjusts them in combination with the requirement and resource orientation weights. It can accurately select the modeling method to match different core design goals, break through the limitations of the traditional fixed modeling links, and make the architecture modeling more in line with the actual design requirements. when When selecting the forward modeling link, when When selecting the reverse modeling link, when Select hybrid modeling link at the time; When the system architecture design is driven by operational requirements and prioritizes ensuring the completeness and accuracy of capability requirements, Take the larger value in the range of 0.6 to 0.8; when the system architecture design is primarily constrained by existing available resources and prioritizes the feasibility and economy of the solution, Take the larger value in the range of 0.6 to 0.8; The extraction module is used to extract the original architectural constraints from four dimensions: structure, logic, resources, and timing. It assigns a unique identifier to each constraint and establishes a mapping table between constraint identifiers and constraint expressions. The extraction module assigns priority weights to constraints in each dimension, calculating constraint priorities based on the severity of the consequences of constraint violations. And establish a mapping table between constraint identifiers and constraint expressions in descending order of priority; Constraint Priority ; In the formula: As the influence weight, For trigger time weighting, ; To constrain the impact of violations, the value ranges from 0 to 1, with a larger value indicating a more severe impact of violating the constraint on the feasibility of the system architecture. To constrain the violation trigger time, the value ranges from 0 to 1. The smaller the value, the earlier the constraint is triggered during the design process. This formula calculates the priority by weighting the impact of constraint violation and the triggering time as the core dimensions. It dynamically adjusts the weight coefficients according to different design stages to achieve quantitative classification and stage adaptation of constraint priority. This solves the problem of subjective and arbitrary constraint sorting in traditional constraints and makes the priority allocation of constraint verification more scientific and reasonable. in, , Based on dynamic adjustments during the system architecture design phase, the conceptual design phase takes... , Detailed design phase , ; Impact of constraint violation Quantification is performed hierarchically according to constraint type, with structural consistency constraints and logical completeness constraints being selected. Resource boundary constraints Logical sequence constraints Constraint violation trigger time Normalized quantization is performed according to the constraint triggering phase, and constraints triggered in the static reduction phase are taken as follows: Constraints triggered in the initial stage of node expansion are taken Constraints triggered in the later stages of scheme generation are taken ; The topology generation module is used to map capabilities, operational activities, and resource entities into three types of standardized topology nodes, build the topology edges that connect nodes, and attach the corresponding constraints to the nodes or topology edges in a directional manner. The topology generation module calculates the support weights of the topological edges that connect nodes. And according to constraint priority The constraints are attached to the corresponding nodes or topological edges, and a constraint-topological association matrix is ​​generated. ; Topological edge support weight The calculation formula is: ; In the formula: , , , These are the attribute weights of the four types of topological edges. ; Capacity-activity support strength; Weighting for activity-resource support; This is the normalized value for resource-to-resource transmission bandwidth. Normalized value for activity-activity sequence delay; The above formula integrates the weighted calculation of the support strength of four types of topological association attributes. It can dynamically adjust the weight of each attribute to adapt to different core design objectives. The default balanced weight distribution ensures universality and realizes the accurate quantification of the support capacity of the topological edge, providing a reliable numerical basis for constraint mounting and topological association analysis. The weights of the four attribute categories are dynamically adjusted based on the core objectives of the system architecture design, increasing when the design objectives emphasize the matching between capabilities and activities. The value of should be increased when the design objective emphasizes the resource's ability to support the activity. The value of should be increased when the design objective focuses on the communication performance between resources. The value of should be increased when the design objective emphasizes the temporal rationality of work activities. The default value for each of the four attribute categories is 0.25. When the design focuses on the match between capabilities and activities The values ​​are adjusted to 0.35~0.4, and the remaining weights are proportionally reduced to a sum of 1; when the design focuses on the resource support capability for the activity, The value should be adjusted to 0.35~0.4; when the design focuses on inter-resource communication performance, The value should be adjusted to 0.35~0.4; when the design focuses on the rationality of the sequence of work activities, The value should be adjusted to 0.35~0.4; The value is determined by expert scoring: when the capability is necessary and the only support for the execution of the activity, it is 1; when the capability plays an auxiliary role in the execution of the activity but is not necessary, it is 0.3 to 0.7; and when the capability is not directly related to the execution of the activity, it is 0. The value is determined based on the percentage of tasks completed by the resources for the activity. When the resources can independently complete all tasks of the activity, the value is 1; when the resources can only complete part of the tasks of the activity, the value is the corresponding percentage of task completion; and when the resources cannot support the execution of the activity, the value is 0. It is calculated by the ratio of the actual transmission bandwidth to the preset maximum transmission bandwidth, and the value ranges from 0 to 1; It is calculated by subtracting the ratio of the actual timing delay to the preset maximum allowable timing delay from 1, and the value ranges from 0 to 1.

[0022] Constraint-Topological Incidence Matrix The generation process is as follows: First, initialize the dimensions as follows: The zero matrix, where The total number of nodes in the three types of standardized topology; traverse all topologically related edges, if the node With nodes If there is a topologically related edge between two points and both edges and their endpoints satisfy all mounting constraints, then the matrix elements will be... Set to 1; if node With nodes If there is a topologically related edge, but the edge or its two endpoints violate at least one mounting constraint, then the matrix elements will be... Set to -1; if node With nodes If there are no topologically related edges between them, then the matrix elements are preserved. Set to 0; store the generated constraint-topological correlation matrix for real-time constraint conflict detection in subsequent decision-making processes; The modeling module is used to construct five categories of decision preferences: performance efficiency, cost resources, technology selection, resource utilization, and job satisfaction, and to transform each preference into a corresponding computable mathematical expression. During the modeling module's runtime phase, a unified preference fusion function is constructed. The mathematical expressions of five types of decision preferences—performance efficiency, cost resources, technology selection, resource utilization, and job satisfaction—are weighted and fused to generate a computable preference guidance function. Preference fusion function The calculation formula is: ; In the formula: , , These are the fusion weights expressed as weighted, rule-based, and priority-based, respectively. ; It is a weighted evaluation function for performance efficiency and cost resources; A reward function for rules governing technology selection and resource utilization; The priority evaluation function for the task; Preference fusion weight , , The emphasis is determined by design preferences; when the focus is on balancing performance and cost resources... The value ranges from 0.5 to 0.6. , The corresponding value is 0.2 to 0.3; when the focus is on technology selection and resource usage rules, The value ranges from 0.5 to 0.6; when the priority task is met, The value ranges from 0.5 to 0.6; in typical design scenarios, , , Take 1 / 3 of each; The above formula weighted and integrated the three types of preference expressions, unified the calculation method of the five types of decision preferences, transformed multi-dimensional and scattered preferences into a computable guiding function, solved the problem of difficulty in coordinating and quantifying multiple preferences, and enabled decision preferences to effectively guide the generation of solutions. The weighted evaluation function for performance efficiency and cost resources is as follows: ; In the formula: For performance and efficiency weights, As a cost-resource weight, ; This is the normalized performance value of the current solution, ranging from 0 to 1. A larger value indicates better performance. This is the normalized cost-resource value for the current solution, ranging from 0 to 1. A larger value indicates a higher cost. The above formula adopts a weighted calculation method that adds positive performance and subtracts negative cost, balancing the two types of weights to adapt to the design focus, realizing a synergistic quantitative evaluation of the solution's performance and cost, breaking through the one-sidedness of a single-dimensional evaluation, and making the cost-effectiveness assessment of the solution more objective. The rule-based reward function for technology selection and resource utilization is: ; In the formula: Specify the number of preferred resources; For the first The reward factor for each preferred resource ranges from 0 to 0.2. As a resource hit identifier, when the solution contains the first The value is 1 if the resource is preferred, otherwise it is 0. The number of technologies to choose from; For the first A technology maturity reward factor, when the technology maturity level... When the value is 0.1, When 0 is taken, The value is -0.2. As a technology usage identifier, when the solution adopts the first The value is 1 if the technology is specific, otherwise it is 0. It should be noted that when the fit S≥0.7, the overall information completeness of requirements and resources is relatively high, the top-level capability requirements are clear, the bottom-level resource data is clear, and the complete input conditions for top-down forward design are available. At this time, adopting the forward modeling link, starting from capability requirements, decomposing work activities and matching resource entities layer by layer, can ensure the requirement traction and goal consistency of the architecture design, and reduce the rework cost caused by later requirement iterations. When S≤0.3, the overall information completeness is insufficient, the top-level requirements have not yet been finalized, and there is a lack of clear input foundation for forward modeling. If top-down design is forcibly carried out, it is easy to waste design resources due to repeated changes in requirements. At this time, reverse modeling is carried out from the bottom up based on the resource entities that have been mastered. By aggregating resources, the supporting capabilities and scope of activities can be deduced, which can maximize the reuse of existing resources, control the risk of solution implementation, and is more in line with the engineering practice logic in the stage of insufficient information. When 0.3 < S < 0.7, both demand and resources have a certain information basis, but neither is complete. A single modeling path is prone to deviation: pure forward design is prone to deviating from the actual existing resources, and pure reverse design is prone to deviating from the final demand target. Therefore, a hybrid modeling link is adopted. Through bidirectional iteration of top-down decomposition and bottom-up aggregation, alignment and convergence are completed at the operation activity level. This can simultaneously take into account demand compliance and resource availability, and improve the overall rationality of the architecture solution. The technology maturity level is divided into nine levels based on the depth of technology verification and the application environment. The definitions of each level are as follows: Level 1: Only basic principle observation and analysis reports are completed, and the verification work is limited to the scope of theoretical research; Level 2: A clear technical concept and application plan have been formed, and the project is in the solution demonstration stage, but functional verification has not yet been carried out; Level 3: Complete the principle verification of key functions, and the verification is carried out in a laboratory or pure simulation environment; Level 4: Complete functional and performance verification at the component and unit levels, with verification conditions in a standard laboratory environment; Level 5: Complete the integration verification of components and subsystems, and conduct verification in an environment simulating real working conditions; Level 6: Complete the integrated demonstration of the complete system prototype, verifying the environment as a high-fidelity simulation environment; Level 7: Complete the on-site demonstration of the system prototype in a real operating environment, verifying that the scenario is a real-world application environment; Level 8: The corresponding system has completed formal type approval and evaluation, and all performance indicators have passed full-item testing and certification; Level 9: The corresponding technology has passed the actual operation assessment, achieved industrial application, and entered the stage of mass service or large-scale commercial use; The above formula sets reward factors for preferred resources and sets differentiated reward and punishment coefficients based on technology maturity. It realizes rule-based reward calculation through hit identification, transforming technology and resource selection preferences into quantifiable scores, making technology and resource selection more in line with design rules and maturity requirements. The priority evaluation function for the task is: ; In the formula: The number of task priority levels. For the first The priority coefficient of the task level, and ; For the first The satisfaction level of the task is 0-1, with a higher value indicating a higher degree of satisfaction. The above formula sets a differentiation coefficient according to the priority level of the task, and combines the weighted sum of the satisfaction of each level of task to realize the priority-based quantitative evaluation of task satisfaction, giving priority to ensuring the satisfaction of core tasks, so that the task adaptation evaluation is more in line with the task design requirements. The reduction module is used to perform static invalid component elimination based on formal constraint topology, and to check constraint conflicts and prune search branches that violate constraints in real time during the search process; Reduce the efficiency of the module running the computing component Static invalid component removal is performed based on a validity threshold, and constraint conflict degree is calculated in real time during the search process. ,when Prune search branches that violate constraints. Among them, the constraint conflict degree threshold The number of constraints is determined based on the system architecture size and the total number of constraints. For small systems, the total number of constraints is less than 15. The value is 0.5; for medium-sized systems with a total of 15-30 constraints, The value is 1.0; when the total number of constraints in a large system exceeds 30, The value is 1.5; this threshold is the critical value for triggering pruning, and if it is exceeded, the branch is considered invalid. Component effectiveness The calculation formula is: ; In the formula: This represents the total number of capability items in the system architecture. For the first The supporting weight of each capability; For the component to the first The degree of support for the item's capabilities; when The component is then determined to be invalid. This formula is based on a weighted sum of capability support weight and component capability support degree. It determines invalid components by numerical value and divides capability weights according to the importance of the task, so as to achieve accurate quantitative screening of component effectiveness, simplify the decision space and improve the efficiency of solution generation. Constraint Conflict Degree The calculation formula is: ; In the formula: This represents the total number of constraints in the system architecture. For the first Priority of item constraints; To constrain violations of the sign, when a constraint is violated... When the constraints are satisfied ; The preset conflict threshold; The formula uses constraint priority as the weight and combines it with constraint violation indicators to obtain the conflict degree by weighted summation. The validity of the branch is determined by comparing it with a preset threshold, realizing real-time quantitative detection of constraint conflicts, providing an accurate basis for search pruning, and avoiding invalid branches from consuming computing resources. in, The tasks are quantified based on the priority of the corresponding tasks, and the tasks are divided into three levels of importance: core, important, and auxiliary, with corresponding weights of 0.6, 0.3, and 0.1, respectively. The sum of the supporting weights of all capabilities is 1. Based on the resource-activity mapping, if a component can independently support all job activities under this capability, then... If it can only support some operational activities, then This is the ratio of the number of activities supported by this component to the total number of activities under this capacity. If it cannot support any work activities, then... ; The scheme generation module is used to generate candidate system architecture schemes that satisfy all constraints within the reduced decision space by employing a heuristic search algorithm and a joint preference guidance function. The solution generation module uses an improved A Heuristic search algorithms are used to construct a comprehensive evaluation function that includes constraint satisfaction and preference scores. Within the reduced decision space, candidate system architectures that satisfy all constraints are generated. Comprehensive evaluation function The calculation formula is: ; In the formula: This represents the cumulative cost of the current architecture. For heuristic functions; Heuristic function The calculation formula is: ; In the formula: The constraint satisfaction of the current solution; The preference score for the current option; , These are the heuristic weights for constraint satisfaction and preference score, respectively. ; This formula uses constraint satisfaction and preference score as the core weighted calculation, sets heuristic weights to prioritize constraints, strengthens the core position of constraint satisfaction, and takes into account decision preference guidance, so that the heuristic direction is both compliant and in line with the design intent, thereby further improving the accuracy of solution generation. in, The formula is a weighted sum of the normalized cost, job execution time, total power consumption, and total weight. ,in , , , The weights of each cost factor and , , , , These are the cumulative cost, operation execution time, total power consumption, and total weight of the current solution. , , , These are the corresponding preset upper limits; The ratio of the weighted sum of priorities of satisfied constraints to the weighted sum of priorities of all constraints is calculated using the following formula: ,in For the first Priority of item constraints To restrain violations of signage; The result of the preference fusion function calculation; Cumulative cost weight , , , The emphasis is determined based on the design cost; when the emphasis is on cost control... The value is set to 0.4, with the remaining weights allocated proportionally; when the focus is on work efficiency, The value is 0.4; when the focus is on power consumption control, The value is 0.4; when the focus is on weight control, The value is 0.4; in normal scenarios, , , , All values ​​are taken as 0.25; Constraint Satisfaction Heuristic Weights Values ​​range from 0.6 to 0.8, representing the preference score heuristic weight. The value ranges from 0.2 to 0.4 to represent the constraint-first decision-making principle; Improved A The heuristic search algorithm starts with the initial architecture state and integrates the evaluation function. As the basis for prioritizing nodes, each node is selected from the priority queue. The smallest node is expanded. Before expansion, the constraint conflict degree is checked. If there is a constraint conflict, the branch is pruned. Otherwise, the comprehensive evaluation function value of the child node is calculated and added to the priority queue until the target state covering all the required capability nodes and satisfying all constraints is found, and the corresponding system architecture candidate scheme is generated. The solution generation module also includes a Pareto optimization unit, which performs Pareto optimality screening on the generated candidate solutions and calculates the overall goodness of each solution. Output ranking of overall excellence Candidate solutions; Overall excellence The calculation formula is: ; In the formula: , These are the goodness weights for constraint satisfaction and preference score, respectively. ; The preset number of output schemes; The above formula calculates the overall superiority of the scheme by weighted summation of constraint satisfaction and preference score. The weights can be adjusted according to the decision-maker's needs to achieve Pareto optimal screening of multiple candidate schemes and quickly output high-quality and compliant architecture schemes. in, The value is set according to the strictness of the decision-maker's requirement that the system architecture solution must meet all constraints; the higher the value, the higher the requirement for the solution's compliance. The value is set according to the decision-maker's emphasis on the system architecture solution conforming to their own decision-making intentions; the higher the value, the higher the requirement for the solution to meet design preferences. When the solution is required to satisfy 100% of all constraints The value ranges from 0.7 to 0.8. The value ranges from 0.2 to 0.3; when balancing constraint satisfaction and preference fit, , All values ​​are set to 0.5; when prioritizing fit, The value ranges from 0.3 to 0.4. The value ranges from 0.6 to 0.7; The output of the building module is connected to the input of the extraction module via data communication. The output of the extraction module is connected to the input of the topology generation module via data communication. The output of the topology generation module is connected to the input of the reduction module via data communication. The output of the modeling module is connected to the first input of the scheme generation module via data communication. The output of the reduction module is connected to the second input of the scheme generation module via data communication.

[0023] In this embodiment, the construction module selects forward, reverse, or hybrid modeling links according to the system architecture design requirements, parses the structured data of the DoDAF model, and extracts information on capabilities, operational activities, and resource entities. The extraction module simultaneously extracts the original constraints of the architecture from four dimensions: structure, logic, resources, and timing. It assigns a unique identifier to each constraint and establishes a mapping table between constraint identifiers and constraint expressions. The topology generation module maps capabilities, operational activities, and resource entities into three types of standardized topology nodes in real time, builds topology edges connecting nodes, and attaches corresponding constraints to nodes or topology edges. The modeling module runs in the background to construct five types of decision preferences: performance efficiency, cost resources, technology selection, resource usage, and operational satisfaction. It transforms each preference into a corresponding computable mathematical expression. Then, the reduction module performs static invalid component elimination based on the formal constraint topology. During the search process, it verifies constraint conflicts in real time and prunes search branches that violate constraints. Finally, the solution generation module uses a heuristic search algorithm in the reduced decision space, combined with a preference guidance function, to generate candidate system architecture solutions that satisfy all constraints.

[0024] In the above embodiments, the system can flexibly adapt the modeling method according to the system architecture design requirements, accurately extract and hierarchically manage multi-dimensional constraints, standardize the topological mounting of nodes and relationships, and transform various decision preferences into quantifiable calculation basis. It can quickly eliminate invalid components, avoid constraint conflicts, effectively reduce the decision search space, and efficiently generate compliant and design-intent-fit architecture solutions based on heuristic search and optimization screening. This significantly improves the accuracy and efficiency of system architecture design, reduces design costs and iteration cycles, and ensures that the solution meets both rigid constraint requirements and actual decision-making and application needs.

[0025] Referring to the system in the above embodiments, here is an example of an application of this system: Given that the fixed resource is a large search and rescue vessel A with long endurance support capability but slow speed and no dedicated medical module, the candidate resource pool includes: a high-speed, light-load unmanned surface vessel B with the designer's usage preference; a drone C with long endurance and wide-area reconnaissance payload; a search and rescue helicopter D equipped with casualty slingshot and emergency medical modules; a sea-based unmanned mothership E with launch and recovery capabilities but with a special interface protocol; and a fixed-wing patrol aircraft F with a very large search range but extremely high cost per operation. Communication options include satellite links, Beidou short message modules, and VHF radios. The design objective is to output a system architecture scheme that meets the constraints and matches the decision-making preferences.

[0026] Phase 1 involves constructing the system architecture constraints. First, a hybrid modeling pipeline is used to build the DoDAF model. During forward modeling, capability requirements and operational activity models are defined, decomposing six sub-capabilities: wide-area surveillance and search, precise target identification, system-wide collaborative command, environmental situational awareness, on-site emergency medical treatment, and long-distance rapid transport. These correspond to nine operational activities: area scanning, suspected target detection, close-range reconnaissance, identity verification, cross-domain information distribution, task allocation, meteorological and sea condition monitoring, maintaining vital signs of the wounded, and evacuation and transport of the wounded. During reverse modeling, a system interface description model is constructed, incorporating the existing search and rescue vessel A and all candidate resources into the model system.

[0027] After modeling, constraints were extracted from four dimensions. Regarding logical completeness constraints, coverage checks were performed based on the mapping relationship between resources and activities. Initially, when only search and rescue vessel A was configured, the activities of maintaining vital signs of the wounded and evacuating the wounded lacked effective resource support, automatically generating constraints requiring additional resources with medical and rescue capabilities. Regarding structural consistency constraints, compatibility checks were performed by reading the interface attribute fields of each resource. When attempting to establish a connection between UAV C and the unmanned mothership E, the communication protocol and guidance protocol were inconsistent, determining the connection to be illegal and blocking subsequent expansion of the corresponding path. Regarding resource boundary constraints, the operation window was set to no more than 60 minutes, and a total budget upper limit was defined. During the search process, the cost and operation time of selected resources were accumulated in real time; if the threshold was exceeded, the corresponding branch was immediately terminated. Regarding logical sequence constraints, expansion rules were set based on the temporal dependencies of operational activities. Only after the resource allocation of prerequisite functional nodes was completed was the exploration of the decision space of subsequent related nodes allowed, avoiding the generation of invalid solutions that violate operational logic.

[0028] Finally, the extracted constraints are transformed into a constraint graph, mapping capabilities, activities, and resources to three layers of nodes. Connections between nodes are built according to the view mapping relationship, and four types of constraints are attached to the corresponding nodes and edges: structural consistency constraints are attached to resource connection edges, logical completeness constraints are attached to the mapping chain from capability to activity to resource, resource boundary constraints are attached to the resource node set, and logical sequence constraints are attached to the activity sequence edges. A constraint adjacency matrix is ​​generated synchronously to verify the legality of node connections, compliance of resource boundaries, completeness of capability coverage, and rationality of temporal logic in real time during the search process, transforming multi-view relationships into a directly processable topological constraint space.

[0029] Phase Two involves designing preference-based decision-making guidance rules. Five decision preferences are constructed based on the characteristics of maritime search and rescue operations: performance-efficiency preference focuses on the speed of casualty evacuation and transfer, aligning with the golden rescue time requirements; cost-resource preference targets area scanning operations, limiting the large-scale use of high-cost patrol aircraft while ensuring coverage; technology selection preference prioritizes satellite links with high technological maturity for information distribution and operation allocation, ensuring communication reliability in extreme sea conditions; resource usage preference prioritizes the use of unmanned surface vessel B when conducting close-in reconnaissance activities, verifying its collaborative performance in complex sea conditions; and operation fulfillment preference defines maintaining the vital signs of the casualties as an indispensable core operational node, assigning it the highest priority.

[0030] The five preference categories were then transformed into corresponding mathematical expressions. Performance and cost / resource preferences were expressed using a weighted model to construct a comprehensive evaluation function, with response time accounting for 0.8 and cost for 0.2, guiding the algorithm to converge towards solutions with faster response times. Technology selection and resource usage preferences were expressed using a rule-based and reward-based model, assigning positive reward scores to the prioritized unmanned surface vessel B and penalty factors to components with insufficient technological maturity, guiding the algorithm towards preferred resource paths under equal performance conditions. Operation fulfillment preferences were expressed using a priority model, classifying operations into core, important, and auxiliary levels. Only after high-priority operations achieved the required coverage were lower-priority operations allowed for optimization and expansion, ensuring that core rescue operations were prioritized.

[0031] Phase Three involves system architecture decision-making guided by constraints and preferences. First, a decision model coupling constraints and preferences is constructed, transforming system architecture design into a state-space search problem. Following the principle of constraints first, then preferences for optimal selection, each decision step first undergoes constraint verification to determine feasibility, and then the path's merits are evaluated using a preference function. The initial state only includes search and rescue vessel A. Constraint graph verification identifies three capability gaps: a broken target identification capability, a lack of situational awareness capability, and a complete lack of resource support for emergency medical and rapid transport capabilities. The initial architecture cannot close the complete operational flow, thus initiating a heuristic search to fill these capability gaps.

[0032] Before the search begins, a decision space reduction process is implemented. During the static reduction phase, based on structural consistency constraints, outdated communication modules with low technological maturity or incompatible with the existing command environment are permanently removed from the candidate pool, compressing the initial search width. During the search, dynamic pruning is performed. When attempting to introduce a fixed-wing patrol aircraft (F-type) to cover the search and detection operations, its single-operation cost exceeds resource boundary constraints. This path branch is pruned early in the search and no longer expands downwards.

[0033] Finally, A was adopted. The algorithm searches for solutions within the reduced space, gradually expanding nodes from the initial state. First, it expands the wide-area perception layer, introducing UAV C, whose branch cost increases minimally, while simultaneously closing the gap between search and environmental monitoring activities, resulting in the best overall score and making it a priority exploration node. Next, it expands the precision identification layer; UAV B, due to its alignment with resource usage preferences, receives bonus points and stands out among candidate resources with equivalent functionality, being prioritized for inclusion in the solution. Entering the core lifeline completion phase, the attempt to combine UAV C and the unmanned mothership E is blocked due to interface protocol conflicts. Search and rescue helicopter D, as the only resource capable of supporting medical and transport operations, although more expensive, offers the most significant optimization effect on response speed and is included in the solution to fill the core capability gap.

[0034] After all sub-capabilities are fully covered, the satellite link closure cross-domain collaborative connection is automatically completed. The search ends and the final architecture scheme is output, which consists of a large search and rescue vessel A, an unmanned surface vessel B, an unmanned aerial vehicle C, and a search and rescue helicopter D, combined with a satellite link module. This scheme meets all constraints and fully matches various decision preferences, and can be directly used for the construction and deployment of a maritime search and rescue system.

[0035] Example 2: At the implementation level, based on Example 1, this example refers to... Figure 2 A further detailed description of the constraint- and preference-guided decision-making system in Example 1 is provided below: A decision-making method based on constraints and preferences includes: Dynamically select forward, reverse, or hybrid modeling links based on system architecture design requirements, parse DoDAF model structured data, and extract full attribute information of capabilities, operational activities, and resource entities; The original architectural constraints are extracted from four dimensions: structure, logic, resources, and timing. A unique identifier is assigned to each constraint and its priority is calculated. A mapping table between constraint identifiers and expressions is established. Capabilities, operational activities, and resource entities are mapped to three types of standardized topological nodes. Node-connected topological edges are built, and constraints are oriented and mounted according to constraint priority to generate a constraint-topological association matrix. Five categories of decision preferences, including performance efficiency, cost and resources, are constructed and transformed into computable mathematical forms. A unified preference guidance function is generated through weighted fusion. Static invalid components are eliminated based on constraint topology. Constraint conflicts are checked in real time during the search process, and search branches that violate constraints are pruned. Adopting improved A A heuristic search algorithm generates candidate solutions that meet the constraints, and Pareto optimization is used to calculate the overall goodness and select the optimal solution.

[0036] In summary, the system and method in this embodiment combine dynamic adaptation of the architecture modeling link, multi-dimensional constraint quantification and topology mounting, mathematical fusion of decision preferences, and intelligent reduction of the decision space. This allows for flexible matching of modeling methods based on requirements and resource completeness, improving the adaptability and completeness of the architecture design. It prioritizes four types of constraints—structural, logical, resource, and temporal—and binds them to topology nodes and associated edges, enabling real-time detection and accurate verification of constraint conflicts. It transforms five types of decision preferences—performance, cost, and resources—into a unified and computable guiding function, making the decision-making process more aligned with design intent and practical application needs. Invalid component elimination and constraint conflict pruning effectively reduce the decision search space, lowering the computational cost and complexity of solution generation. Finally, an improved heuristic search combined with Pareto optimization is used to generate an architecture solution that satisfies all constraints and aligns with decision preferences, balancing compliance, rationality, and generation efficiency, providing stable and efficient decision support for architecture design.

[0037] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A decision-making system based on constraints and preferences, characterized in that, include: The building module is used to select forward, reverse, or hybrid modeling links according to the system architecture design requirements, parse the structured data of the DoDAF model, and extract capability, operational activities, and resource entity information; The extraction module is used to extract the original architectural constraints from four dimensions: structure, logic, resources, and timing. It assigns a unique identifier to each constraint and establishes a mapping table between constraint identifiers and constraint expressions. The topology generation module is used to map capabilities, operational activities, and resource entities into three types of standardized topology nodes, build the topology edges that connect nodes, and attach the corresponding constraints to the nodes or topology edges in a directional manner. The modeling module is used to construct five categories of decision preferences: performance efficiency, cost resources, technology selection, resource utilization, and job satisfaction, and to transform each preference into a corresponding computable mathematical expression. The reduction module is used to perform static invalid component elimination based on formal constraint topology, and to check constraint conflicts and prune search branches that violate constraints in real time during the search process; The scheme generation module is used to generate candidate system architecture schemes that satisfy all constraints within the reduced decision space by employing a heuristic search algorithm and a joint preference guidance function.

2. The decision-making system based on constraints and preferences according to claim 1, characterized in that, During the runtime phase of the construction module, the completeness of the requirements for the synchronous computing architecture design is assessed. and resource completeness Based on modeling link adaptability Dynamically select forward, reverse, or hybrid modeling links, and parse structured XML data according to the entity association rules of the DoDAF metamodel DM2 to extract full attribute information of capabilities, operational activities, and resource entities; The modeling link adaptability The calculation formula is: ； in, Demand-oriented weighting As a resource-oriented weight, ; Take the ratio of the clearly defined high-level capacity requirements to the total capacity requirements; Take the ratio of the number of underlying system resource information that has been obtained to the total number of resource types; when When selecting the forward modeling link, when When selecting the reverse modeling link, when Select the hybrid modeling link at that time.

3. The decision-making system based on constraints and preferences according to claim 1, characterized in that, The extraction module assigns priority weights to constraints in each dimension and calculates constraint priorities based on the severity of the consequences of constraint violations. And establish a mapping table between constraint identifiers and constraint expressions in descending order of priority; The constraint priority ; In the formula: As the influence weight, For triggering time weights, ; To constrain the impact of violations; To constrain the trigger time of violations.

4. The decision-making system based on constraints and preferences according to claim 1, characterized in that, The topology generation module calculates the support weights of the topological edges connecting the nodes. And according to constraint priority The constraints are attached to the corresponding nodes or topological edges, and a constraint-topological association matrix is ​​generated. ; The topology edge support weight The calculation formula is: ； In the formula: , , , These are the attribute weights of the four types of topological edges. ; Capacity-activity support strength; Weighting for activity-resource support; This is the normalized value for resource-to-resource transmission bandwidth. This is the normalized value for the activity-activity sequence delay.

5. A decision-making system based on constraints and preferences according to claim 1, characterized in that, During the modeling module's runtime phase, a unified preference fusion function is constructed. The mathematical expressions of five types of decision preferences—performance efficiency, cost resources, technology selection, resource utilization, and job satisfaction—are weighted and fused to generate a computable preference guidance function. The preference fusion function The calculation formula is: ； In the formula: , , These are the fusion weights expressed as weighted, rule-based, and priority-based, respectively. ; It is a weighted evaluation function for performance efficiency and cost resources; A reward function for rules governing technology selection and resource utilization; This is the priority evaluation function that the task satisfies.

6. A decision-making system based on constraints and preferences according to claim 1, characterized in that, The reduction module runs the calculation component's effectiveness Static invalid component removal is performed based on a validity threshold, and constraint conflict degree is calculated in real time during the search process. ,when Prune search branches that violate constraints. The effectiveness of the components The calculation formula is: ； In the formula: This represents the total number of capability items in the system architecture. For the first The supporting weight of each capability; For the component to the first The degree of support for the item's capabilities; when The component is then determined to be invalid. The degree of constraint conflict The calculation formula is: ； In the formula: This represents the total number of constraints in the system architecture. For the first Priority of item constraints; To constrain violations of the sign, when a constraint is violated... When the constraints are satisfied ; This is the preset conflict threshold.

7. A decision-making system based on constraints and preferences according to claim 1, characterized in that, The scheme generation module uses an improved A Heuristic search algorithms are used to construct a comprehensive evaluation function that includes constraint satisfaction and preference scores. Within the reduced decision space, candidate system architectures that satisfy all constraints are generated. The comprehensive evaluation function The calculation formula is: ； In the formula: This represents the cumulative cost of the current architecture. For heuristic functions; The heuristic function The calculation formula is: ； In the formula: The constraint satisfaction of the current solution; The preference score for the current option; , These are the heuristic weights for constraint satisfaction and preference score, respectively. .

8. A decision-making system based on constraints and preferences according to claim 7, characterized in that, The scheme generation module also includes a Pareto optimization unit, used to perform Pareto optimal selection on the generated multiple candidate schemes and calculate the overall goodness of each scheme. Output ranking of overall excellence Candidate solutions; The overall excellence The calculation formula is: ； In the formula: , These are the goodness weights for constraint satisfaction and preference score, respectively. ; This represents the preset number of output schemes.

9. A decision-making system based on constraints and preferences according to claim 1, characterized in that, The output of the construction module is connected to the input of the extraction module via data communication. The output of the extraction module is connected to the input of the topology generation module via data communication. The output of the topology generation module is connected to the input of the reduction module via data communication. The output of the modeling module is connected to the first input of the scheme generation module via data communication. The output of the reduction module is connected to the second input of the scheme generation module via data communication.

10. A decision-making method based on constraints and preferences, said method being an implementation method of a decision-making system based on constraints and preferences as described in any one of claims 1-9, characterized in that, include: Dynamically select forward, reverse, or hybrid modeling links based on system architecture design requirements, parse DoDAF model structured data, and extract full attribute information of capabilities, operational activities, and resource entities; The original architectural constraints are extracted from four dimensions: structure, logic, resources, and timing. A unique identifier is assigned to each constraint and its priority is calculated. A mapping table between constraint identifiers and expressions is established. Capabilities, operational activities, and resource entities are mapped to three types of standardized topological nodes. Node-connected topological edges are built, and constraints are oriented and mounted according to constraint priority to generate a constraint-topological association matrix. Five categories of decision preferences, including performance efficiency, cost and resources, are constructed and transformed into computable mathematical forms. A unified preference guidance function is generated through weighted fusion. Static invalid components are eliminated based on constraint topology. Constraint conflicts are checked in real time during the search process, and search branches that violate constraints are pruned. Adopting improved A A heuristic search algorithm generates candidate solutions that meet the constraints, and Pareto optimization is used to calculate the overall goodness and select the optimal solution.

Images