An intelligent task planning method based on large model and operation optimization

Abstract

The application belongs to the technical field of operation optimization, and discloses an intelligent task planning method based on a large model and operation optimization, which comprises the following steps: obtaining user task requirements and system available resources; based on the user task requirements, generating multiple initial planning schemes by using a large model, and obtaining corresponding resource requirement lists; based on the system available resources and the resource requirement lists, obtaining optimal resource allocation strategies, subtask scheduling plans and corresponding performance indicators under each initial planning scheme; based on the obtained performance indicators, obtaining comprehensive scores of each initial planning scheme; selecting an initial planning scheme with the highest comprehensive score as a final task planning scheme, and outputting corresponding resource allocation strategies and subtask scheduling plans; the method integrates a large model and operation optimization, and realizes efficient task planning and optimal resource allocation.

Description

Technical Field

[0001] This invention relates to the field of operations research and optimization technology, and more specifically, to an intelligent task planning method based on large models and operations research and optimization. Background Technology

[0002] In today's rapidly evolving digital age, intelligent task planning is playing an increasingly important role across various industries, such as project management, resource scheduling, supply chain optimization, and robot path planning. However, with the increasing complexity of tasks, the dynamic changes in available resources, and the growing diversity of user needs, large-scale models are gradually being used for task planning.

[0003] Existing methods in intelligent task planning typically suffer from the following problems: First, large models struggle to fully understand the complexity and multidimensionality of user task requirements, especially the user's potential expectations for innovative solutions, resulting in insufficient diversity in the generated initial solutions or an inability to meet the user's deeper needs. For example, a logistics company wants to optimize the route planning of its unmanned delivery robots, requiring innovative strategies that are more forward-looking and self-learning, distinct from traditional path optimization algorithms. However, because the expression is too vague, large models struggle to accurately understand the user's intent, potentially failing to identify the user's implicit need for novelty or breakthroughs. Second, when effectively combining the cognitive reasoning capabilities of large models with the precise computational capabilities of operations research optimization methods, a systematic process is often lacking to ensure the effectiveness of both. Collaborative work makes it difficult to generate solutions that both meet user intent and are optimal in resource allocation and task scheduling. For example, large models predict order volume and delivery demand trends in various regions by analyzing historical data and user order information. However, when combined with operations research optimization methods, there is no systematic process to translate the prediction results into specific vehicle scheduling, route planning, and warehouse allocation schemes. This leads to unreasonable vehicle allocation, high empty load rates on some routes, and severe capacity shortages in other areas, increasing costs and affecting delivery timeliness. Thirdly, when faced with multiple potential planning schemes, there is a lack of a comprehensive, objective evaluation mechanism that considers multiple indicators, making it difficult to select the optimal solution from numerous options. In view of this, this invention proposes an intelligent task planning method based on large models and operations research optimization to solve the above problems. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of existing technologies and achieve the above objectives, this invention provides the following technical solution: an intelligent task planning method based on large models and operations research optimization, comprising:

[0005] S1. Obtain user task requirements and available system resources;

[0006] S2. Based on user task requirements, generate multiple initial planning schemes using a large model and obtain the corresponding resource requirement list;

[0007] S3. Based on the system's available resources and resource requirements list, obtain the preferred resource allocation strategy, subtask scheduling plan, and corresponding performance indicators for each initial planning scheme;

[0008] S4. Based on the obtained performance indicators, obtain the comprehensive score of each initial planning scheme; select the initial planning scheme with the highest comprehensive score as the final task planning scheme, and output the resource allocation strategy and subtask scheduling plan corresponding to the final task planning scheme.

[0009] Furthermore, user task requirements include the task objectives to be achieved, constraints, and a description of the innovativeness of the solution.

[0010] Furthermore, the methods for generating multiple initial planning schemes include:

[0011] The task objectives are decoupled into three decision-making dimensions: process advancement, resource strategy, and technical architecture.

[0012] The innovativeness of user task requirements is evaluated based on preset innovativeness evaluation dimensions to obtain an innovativeness coefficient;

[0013] The characteristics and threshold ranges of innovation gradient levels are preset, and an innovation coefficient-innovation gradient level mapping table is constructed to convert the innovation coefficient into an innovation gradient level.

[0014] Input the task objectives, constraints, decision dimensions, and innovation gradient levels into a pre-set large model to generate multiple initial planning schemes.

[0015] Furthermore, the methods for obtaining the innovation coefficient include:

[0016] Each dimension of innovation assessment is assigned a corresponding weight based on the domain of the user's task requirements;

[0017] Obtain keywords related to each dimension of innovation assessment, establish the correspondence between keywords and score ranges, and form a semantic mapping table;

[0018] Extract keywords related to innovation from the description of the innovation of the solution; match the extracted keywords with the semantic mapping table to obtain the score corresponding to each innovation evaluation dimension; if there are multiple related descriptions for the same innovation evaluation dimension, calculate the average score of the multiple related descriptions as the final score of the innovation evaluation dimension.

[0019] The innovation coefficient is calculated by weighting and summing the scores of each dimension of innovation assessment.

[0020] Furthermore, the preferred resource allocation strategies, subtask scheduling plans, and methods for obtaining corresponding performance metrics include:

[0021] The user task requirements are broken down into independent subtasks, and the resource requirements, time consumption, and dependencies between different subtasks are obtained for each subtask. Based on the dependencies between subtasks, an initial scheduling sequence for the subtasks is generated using topological sorting.

[0022] Based on the total available resources of the system and the resource requirements of each subtask, resource total constraints are constructed; based on the time consumption data of the subtasks, time constraints are constructed; based on the dependencies between different subtasks, task dependency constraints are constructed.

[0023] Using the shortest total completion time, lowest total cost, highest average resource utilization, and fewest critical resources as multi-objective optimization objectives, and based on resource constraints, time constraints, and task dependency constraints, operations research optimization methods are used to optimize the initial scheduling sequence of sub-tasks, resulting in a sub-task scheduling plan and resource allocation strategy. Based on the sub-task scheduling plan and resource allocation strategy, the performance indicators of the initial planning scheme are calculated, including total completion time, total cost, average resource utilization, and number of critical resources used.

[0024] Furthermore, during the scheduling optimization of the initial scheduling sequence of subtasks, if multiple subtasks compete for the same resource at the same time, the number of available resources in the current system and the total demand of each subtask for that resource are obtained, and the demand ratio of the total demand to the number of available resources is calculated.

[0025] A preset ratio threshold is used. If the demand ratio is less than or equal to the preset ratio threshold, resources are allocated from high to low based on the comprehensive influence index of the subtasks. If the demand ratio is greater than the preset ratio threshold, all subtasks competing for the same type of resource within the current time slice or decision cycle are considered as an auction market, with each subtask being a bidder in the market. The negative impact of the delay caused by the subtask's failure to obtain resources on the task objective is obtained, resulting in a cost factor. Based on its own comprehensive influence index and cost factor, the subtask calculates its comprehensive bid in the resource auction, and resources are allocated from high to low based on the size of the subtask's comprehensive bid.

[0026] Furthermore, the methods for obtaining the comprehensive impact index of sub-tasks include:

[0027] Obtain the direct contribution of each subtask to the task objective; construct a task dependency graph based on the dependencies between subtasks, and assign a corresponding dependency weight to each directed edge in the task dependency graph to generate a dependency matrix; calculate the betweenness centrality of each subtask and the sum of the dependency weights of all directed edges received by the node corresponding to each subtask to obtain the weighted in-degree of the subtask; evaluate the influence of the subtasks using fuzzy logic based on the direct contribution, betweenness centrality, and weighted in-degree to obtain the comprehensive influence index of the subtasks.

[0028] Furthermore, methods for evaluating the influence of sub-tasks using fuzzy logic include:

[0029] The direct contribution, betweenness centrality, and weighted in-degree of a subtask are fuzzified using a predefined fuzzy membership function to obtain the fuzzy membership degrees corresponding to these parameters. A pre-defined inference rule base is then used to perform inference operations on the fuzzified fuzzy membership degrees of the subtask to generate the comprehensive fuzzy membership degree of the subtask. Finally, the comprehensive fuzzy membership degree is defuzzified to obtain the comprehensive influence index of the subtask.

[0030] Furthermore, the methods for obtaining the comprehensive score include:

[0031] The performance indicators are normalized, and Pareto dominance analysis is performed on the performance indicators of all normalized initial planning schemes to identify and eliminate Pareto-dominated initial planning schemes, generating a candidate scheme set. Based on the candidate scheme set, an ideal optimal solution composed of the optimal values ​​of each performance indicator and an ideal worst solution composed of the worst values ​​of each performance indicator are constructed. Combining the preset weight coefficients of each performance indicator, the weighted optimal solution distance and the weighted worst solution distance of the initial planning schemes in the candidate scheme set to the ideal optimal solution and the weighted worst solution distance are calculated. The relative proximity of the initial planning schemes in the candidate scheme set is calculated based on the weighted optimal solution distance and the weighted worst solution distance. The relative proximity is the comprehensive score of the initial planning scheme.

[0032] The technical effects and advantages of this invention are as follows:

[0033] This invention, at the level of understanding user needs, refines user task requirements into task objectives, constraints, and descriptions of solution innovation, and constructs an innovation evaluation system. It transforms users' potential expectations for solution innovation into quantifiable indicators, and inputs these into a large model using three decision dimensions decoupled from task objectives, generating diverse initial planning schemes that meet deep-seated needs. Regarding scheme optimization, it establishes a system process that coordinates the large model with operations research optimization. Operations research optimization optimizes sub-task scheduling and resource allocation based on resource constraints and multi-objective optimization, handling resource competition through a comprehensive influence index and auction mechanism to ensure rational resource allocation. In the scheme evaluation stage, based on performance indicators, it constructs a scientific and objective comprehensive scoring system through Pareto dominance analysis and calculation of relative proximity to the ideal solution, selecting the globally optimal scheme. Furthermore, the comprehensive influence evaluation of sub-tasks and the setting of multi-objective optimization enable this method to flexibly adapt to complex scenarios in multiple fields, effectively addressing issues such as high task complexity, dynamic resource changes, and diverse user needs, significantly improving the intelligence level of task planning and resource allocation efficiency, and possessing broad application value. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of an intelligent task planning method based on large model and operations research optimization according to the present invention;

[0035] Figure 2 This is a schematic diagram of resource allocation according to the present invention. Detailed Implementation

[0036] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] Example 1

[0038] Please see Figure 1 As shown in this embodiment, an intelligent task planning method based on large models and operations research optimization includes:

[0039] S1. Obtain user task requirements and available system resources;

[0040] S2. Based on user task requirements, generate multiple initial planning schemes using a large model and obtain the corresponding resource requirement list;

[0041] S3. Based on the system's available resources and resource requirements list, obtain the preferred resource allocation strategy, subtask scheduling plan, and corresponding performance indicators for each initial planning scheme;

[0042] S4. Based on the obtained performance indicators, obtain the comprehensive score of each initial planning scheme; select the initial planning scheme with the highest comprehensive score as the final task planning scheme, and output the resource allocation strategy and subtask scheduling plan corresponding to the final task planning scheme.

[0043] Furthermore, user task requirements include the task objectives to be achieved, constraints, and a description of the innovativeness of the solution. Constraints include the time frame for achieving the task objectives, budget limits, and whether a specific solution or process is restricted. The description of the innovativeness of the solution is a written description or requirement regarding the novelty, uniqueness, or breakthrough of the solution provided by the large model in terms of technology, method, or strategy when the user submits the task requirements. The quantity of available system resources includes the resources that the current system can invest in completing the task, including the number of devices, personnel, and available funds.

[0044] Furthermore, the methods for generating multiple initial planning schemes include:

[0045] The task objectives in user task requirements are decoupled into three decision dimensions: process advancement, resource strategy, and technical architecture.

[0046] Taking software project development as an example, the process can be divided into different methods such as serial development and parallel development. Serial development is to proceed step by step according to the established development process, while parallel development divides the project into different modules, with multiple modules being developed simultaneously and then spliced ​​together to complete the project. Resource strategy refers to the amount and method of resource investment, such as internal development, partial outsourcing, or full outsourcing, as well as the personnel and funds invested. Technical architecture refers to the technology used to implement the software, such as using mature industry technologies, using entirely new technologies, or using a combination of old and new technologies.

[0047] Taking the implementation of an engineering project as an example, process advancement includes the order of implementation or construction of the various sub-projects of the project, which may contain some kind of dependency relationship, that is, the progress of one sub-project depends on the completion of another sub-project; resource strategy represents the number and method of investing resources, such as the number of engineering equipment, the number of construction personnel, and the number of engineering equipment and construction personnel allocated to each sub-project; technical architecture can be divided into the construction schemes adopted to complete the project, such as adopting traditional construction schemes, adopting process innovation construction schemes, or adopting completely new construction schemes;

[0048] Based on the innovative description of the solution in the user's task requirements, the innovativeness of the task requirements is evaluated through five innovativeness assessment dimensions: novelty, breakthrough, practicality, uniqueness, and technical difficulty, and an innovativeness coefficient is obtained.

[0049] The characteristics and threshold ranges of innovation gradient levels are preset, and an innovation coefficient-innovation gradient level mapping table is constructed to transform the innovation coefficient into a specific innovation gradient level.

[0050] The task objectives, constraints, decision dimensions, and innovation gradient levels are submitted as input parameters to a large model (such as GPT or Deep Seek). This large model adopts a generative architecture based on deep learning, and establishes a multi-dimensional mapping relationship between goal setting, constraints, decision dimensions, innovation gradient levels, and planning schemes through unsupervised and supervised mixed training on massive historical project cases, industry knowledge graphs, and technical documents. When the model receives the input information, it first preprocesses the data, using semantic parsing algorithms to extract key elements and perform structured encoding. Subsequently, based on the knowledge system and logical reasoning rules formed during training, the large model generates multiple candidate planning schemes under the three decision dimensions of process advancement, resource strategy, and technical architecture, combined with natural language processing and generative adversarial network (GAN) technology. Each scheme strictly follows the established innovation gradient level standard, and the feasibility and compliance of the scheme are dynamically verified through the constraint verification module. Finally, multiple initial planning schemes that take into account both relevance and practicality are output.

[0051] It should be noted that the innovation gradient levels are divided into three categories, including L1 innovation: mature technology recombination, which utilizes existing, widely validated technologies and achieves the goal through recombination and optimization; L2 innovation: industry-specific innovative technology application, which refers to introducing emerging and innovative technologies within the industry to solve problems; and L3 innovation: cross-domain technology migration, which introduces advanced technologies from other fields into the current field to create entirely new solutions. The threshold ranges for each innovation gradient level are: L1 innovation: [0, 0.4]; L2 innovation: [0.35, 0.75]; L3 innovation: [0.6, 1.0]. If the innovation coefficient belongs to both L1 and L2 or L2 and L3 innovation gradient levels, the solutions corresponding to the two gradient levels will be given respectively to better meet user needs.

[0052] Furthermore, the methods for obtaining the innovation coefficient include:

[0053] Based on the user's needs and their respective fields (such as technology, business, engineering, healthcare, etc.), assign corresponding weights to each dimension of innovation evaluation, with the total weight of each dimension being 1.

[0054] Keywords related to each dimension of innovation assessment are obtained. By combining industry terminology and semantic features, a correspondence between keywords and scoring ranges (0-1) is established to form a semantic mapping table. This semantic mapping table clarifies the scoring criteria for different keywords under the corresponding dimensions, providing a quantitative basis for subsequent text scoring.

[0055] The text describing the innovation of the solution is segmented into words and sentences to remove redundant information that is irrelevant to the innovation assessment; natural language processing tools (such as NLP algorithms) or manual annotation are used to accurately identify and extract keywords and key descriptions related to innovation in the text, laying the foundation for dimension scoring;

[0056] The extracted keywords and key descriptions are matched against a semantic mapping table to obtain the score for each innovation evaluation dimension, with a score range of 0-1. If there are multiple related descriptions for the same dimension, the average score of the multiple descriptions is calculated as the final score for that dimension. For example, if a technical solution has two descriptions in the novelty dimension, and the scores after matching with the mapping table are 0.8 and 0.9 respectively, then the score for that dimension is the average of the two, 0.85.

[0057] Based on the preset weights of each innovation evaluation dimension and the corresponding dimension scores, the innovation coefficient of the user task requirement (with a value of 0-1) is calculated by weighted summation, thereby realizing the quantitative evaluation of the text innovation.

[0058] It should be noted that the resource requirement list is generated from the large model using prompt words, and only includes the resources needed to implement the initial planning scheme generated by the large model, without including the quantity of the required resources; the prompt words indicate the types of resources required for the scheme.

[0059] Furthermore, the preferred resource allocation strategy, subtask scheduling plan, and methods for obtaining the corresponding performance metrics include:

[0060] Starting from user needs, a Work Breakdown Structure (WBS) is used to hierarchically break down user task requirements into independent, executable subtasks. Each subtask contains several steps to achieve the task, preventing over-segmentation that leads to task fragmentation (such as breaking down writing a report into meaningless tasks like opening a document and entering a title). For example, if the user goal is to design an app, it can be broken down into five sub-goals: product research, product planning, product development, product testing, and product deployment. Each sub-goal contains several steps to achieve the goal, without further breaking down the implementation steps.

[0061] Use large models or expert systems to obtain resource requirements and time consumption data for each subtask, as well as the dependencies between different subtasks;

[0062] Based on the available resources of the system, a total resource constraint is established to ensure that the total number of resources allocated to subtasks does not exceed the total capacity of the available resources of the system.

[0063] Based on the time consumption data of subtasks, time constraints are constructed; the time constraints are used to ensure that each subtask is fully executed and that the completion time of all subtasks does not exceed the time constraints in the user task requirements.

[0064] Based on the dependencies between different subtasks, task dependency constraints are constructed; these constraints are used to ensure that the execution order of subtasks conforms to logic.

[0065] Based on the dependencies between subtasks, an initial scheduling sequence for subtasks is generated using topological sorting.

[0066] With the optimization objectives of minimizing total completion time, total cost, average resource utilization, and the number of critical resources used, and based on resource constraints, time constraints, and task dependency constraints, operations research optimization methods (such as particle swarm optimization and ant colony optimization) are used to optimize the initial scheduling sequence of subtasks, resulting in a subtask scheduling plan and resource allocation strategy. Based on the subtask scheduling plan and resource allocation strategy, the performance indicators of the initial planning scheme are calculated, including total completion time, total cost, average resource utilization, and the number of critical resources used.

[0067] The methods for obtaining performance indicators for the initial planning scheme include:

[0068] Based on the subtask scheduling plan and resource allocation strategy, a scheduling Gantt chart and a resource allocation table are constructed. The scheduling Gantt chart intuitively displays the start and end times of subtasks and the parallel execution intervals. The resource allocation table details the resource allocation of each subtask in each time slice.

[0069] In the scheduling Gantt chart, the end times of all subtasks are traversed. If there are task execution constraints caused by dependencies, the maximum value should be selected as the total completion time of the entire task, provided that the task dependency logic is satisfied. If there are no dependencies among all subtasks, the maximum end time can be directly selected as the total completion time. This indicator reflects the total time spent on the task from start to finish and is one of the key indicators for measuring task execution efficiency.

[0070] Based on the resource allocation table, the resource usage of each subtask in each time slice is calculated, and combined with the pre-set resource cost unit price, the cost of each subtask is accumulated in time sequence, and finally the total cost is obtained by summing them up, which is used to measure the economic input of task execution.

[0071] Iterate through each time slice in the resource allocation table. For each resource type, summarize the amount of resource occupied by all subtasks within that time slice to obtain the actual usage of each resource in each time slice. For each resource, divide the actual usage of the resource within each time slice by its pre-defined total capacity to obtain the utilization rate of the resource in that time slice. For example, if the total CPU capacity is 8 cores, and the actual CPU usage in time slice TS1 is 5 cores, then the CPU utilization rate in time slice TS1 is 62.5%. For each resource, calculate the arithmetic mean of its utilization rate across all time slices, that is, add up the utilization rates of each time slice and divide by the total number of time slices to obtain the average utilization rate of the resource over the entire task cycle. Summarize the average utilization rates of all resources to form a resource utilization data set. Correct the data through weighted averaging, standardization, and other methods to finally obtain the average resource utilization rate used to measure the utilization of all resources.

[0072] In the resource allocation table, each time slice is traversed sequentially according to time. For each resource type, the total demand for that resource by all subtasks within that time slice is compared with the pre-defined available capacity. If the total demand for a resource in a time slice exceeds its available capacity, the resource corresponding to that time slice is determined to be a critical resource, and the time slice and resource type are recorded. If the total demand does not exceed the available capacity, the resource is not a critical resource in that time slice. After traversing all time slices and resource types, the recorded critical resource entries are deduplicated and counted. The final count result is the number of critical resources used. This indicator can accurately locate the links in the task execution process where resource supply and demand are unbalanced and there is tension pressure, providing a key basis for subsequent optimization of resource allocation.

[0073] Furthermore, such as Figure 2 As shown, during the scheduling optimization of the initial scheduling sequence of subtasks, if multiple subtasks compete for the same resource simultaneously, the resource allocation method is as follows:

[0074] Obtain the current number of available resources R and the total resource requirement D for each subtask; calculate the ratio C of the total requirement D to the number of available resources R.

[0075] A preset ratio threshold (between 1.1 and 1.3, reflecting the intensity of competition for resources) is set. If the demand ratio C is less than or equal to the preset ratio threshold, resources are allocated based on the comprehensive influence index of the sub-task.

[0076] If the demand ratio C is greater than the preset ratio threshold, the resource allocation method is as follows:

[0077] Consider the set of all subtasks competing for the same type of resource R within the current time slice (or decision cycle) as an auction market, where each subtask is a bidder in the market;

[0078] The negative impact of a subtask being delayed due to failure to acquire resources on the task objective is obtained, yielding a cost factor; the functional expression for the cost factor is: ;in, Indicates the first The cost factor of acquiring resources for each sub-task; Indicates the first The total floating time of each subtask is obtained from the results of the topological sorting algorithm described above; Indicates the first Estimated time required to complete each sub-task; Indicates the first The direct contribution of each subtask to the task objective;

[0079] Subtasks provide a comprehensive bid in resource auctions based on their own comprehensive influence index and cost factor. ; ; and The preset weighting coefficients, Indicates the first The overall impact index of each sub-task;

[0080] Resources are allocated based on the overall bid of the subtasks from high to low. Each subtask will receive all the resources it needs until the resources are exhausted. If multiple subtasks have the same overall bid, resources will be allocated to the subtask with the higher overall influence index.

[0081] Furthermore, the methods for obtaining the comprehensive influence index information of sub-tasks include:

[0082] By breaking down OKRs or KPIs, we can obtain the direct contribution (0-1) of each subtask to the task objective.

[0083] Based on the dependencies between subtasks, a task dependency graph is constructed. The task dependency graph is a directed acyclic graph (DAG). Nodes in the task dependency graph represent subtasks and carry the comprehensive influence index information of the subtasks. Subtasks with dependencies are connected by directed edges to present the execution order and parallel possibilities between subtasks.

[0084] The large model assigns a dependency weight of 1-10 to each directed edge in the task dependency graph (1 point indicates a weak dependency, and 10 points indicate a strong dependency). A corresponding dependency matrix is ​​generated based on the dependency weights of the directed edges. The elements in the matrix are the dependency weights of the edges in the task dependency graph; an element value of 0 indicates that there is no dependency.

[0085] Based on the dependency matrix, the betweenness centrality (range 0-1) of each subtask is obtained using tools such as Gephi, and the weighted in-degree of each subtask is obtained by summing the weights of all edges received by the node corresponding to each subtask. Based on the direct contribution, betweenness centrality, and weighted in-degree, the influence of the subtask is evaluated by fuzzy logic method to obtain the comprehensive influence index of the subtask.

[0086] Methods for evaluating the impact of sub-tasks using fuzzy logic include:

[0087] Membership functions are defined for direct contribution, betweenness centrality, and weighted in-degree, respectively. These membership functions map a specific numerical index (e.g., a direct contribution of 0.7) to one or more fuzzy membership degrees (between 0 and 1). For example, the direct contribution index is defined as follows:

[0088] Fuzzy set: {low, medium, high}; In the membership function, low is defined as a direct contribution of less than or equal to 0.3, which is 1, and then decreases to 0; medium is defined as a membership of 0 at 0.2, 1 at 0.5, and 0 at 0.8; high is defined as a contribution of greater than or equal to 0.7, which is 0, and then increases to 1.

[0089] The direct contribution, betweenness centrality, and weighted in-degree are fuzzified using a predefined fuzzy membership function to obtain the fuzzy membership degrees corresponding to the direct contribution, betweenness centrality, and weighted in-degree. For example, if the direct contribution of subtask A is 0.8, it may have a membership degree of 0.9 for "high" contribution, 0.1 for "medium" contribution, and 0 for "low" contribution.

[0090] Establish IF-THEN format rules and store them in the inference rule base. IF-THEN format rules are used to describe how different combinations of indicators affect the overall impact of subtasks. These rules are determined by domain experts or through data-driven methods, for example:

[0091] IF (High direct contribution IS) AND (Medium intermediary centrality IS) THEN (Strong overall influence IS)

[0092] IF (Direct Contribution IS Medium) AND (Weighted Influence IS High) THEN (Overall Influence IS Medium);

[0093] IF (low intermediary centrality) THEN (weak overall influence);

[0094] If (direct contribution is low) OR (weighted input is low) then (overall influence is weak);

[0095] Based on a pre-defined inference rule base, the membership degrees of the fuzzified direct contribution, betweenness centrality, and weighted in-degree are inferred and calculated. Specifically, for each rule, the minimum membership degree of the input indicator is taken as the activation strength. For example, in the rule "IF (direct contribution IS high) AND (betweenness centrality IS medium) THEN (overall influence IS strong)", if the membership degree of "high" direct contribution is 0.9 and the membership degree of "medium" betweenness centrality is 0.6, then the activation strength of this rule is 0.6. For all activation rules of the same output term (such as "strong"), the maximum strength is selected as the final membership degree of that term.

[0096] After completing the inference operation, the fuzzy outputs (multiple subsets representing strong, medium, and weak comprehensive influence respectively) generated by all rule inference are merged to generate a comprehensive fuzzy membership degree. A defuzzification process (such as the centroid method, the maximum membership degree averaging method, etc.) is used to convert the aggregated fuzzy output set into a precise value, which is the comprehensive influence index of the subtask. The higher the comprehensive influence index of all subtasks, the stronger the importance and criticality of the subtask in the task process.

[0097] Furthermore, the methods for obtaining the comprehensive score include:

[0098] The weighting coefficients of each performance index are determined by the Delphi method or AHP.

[0099] The total completion time, total cost, average resource utilization rate, and number of key resources are processed using a normalization method to obtain corresponding normalized indicators, thereby ensuring that each indicator is measured under a uniform scale.

[0100] Pareto dominance analysis was performed on the normalized performance indicators:

[0101] If there exists an initial planning scheme A whose performance indicators are all no worse than those of the initial planning scheme B, and at least one indicator is strictly better than that of scheme B, then scheme A dominates scheme B; eliminate all dominated initial planning schemes to generate a candidate scheme set; construct the ideal optimal solution and the ideal worst solution for each performance indicator based on the candidate scheme set. The ideal optimal solution consists of the optimal value of each performance indicator in the candidate scheme set, and the ideal worst solution consists of the worst value of each performance indicator in the candidate scheme set.

[0102] Based on the weighting coefficients of each performance index, the weighted optimal solution distance and the weighted inferior solution distance to the ideal optimal solution are calculated for the initial planning schemes in the candidate scheme set; the relative proximity of the initial planning schemes in the candidate scheme set is calculated based on the weighted optimal solution distance and the weighted inferior solution distance; the relative proximity is the comprehensive score of the initial planning scheme.

[0103] The initial planning scheme with the highest comprehensive score is selected as the final task planning scheme, and the corresponding resource allocation strategy and task scheduling plan are output.

[0104] This invention, at the level of understanding user needs, refines user task requirements into task objectives, constraints, and descriptions of solution innovation, and constructs an innovation evaluation system. It transforms users' potential expectations for solution innovation into quantifiable indicators, and inputs these into a large model using three decision dimensions decoupled from task objectives, generating diverse initial planning schemes that meet deep-seated needs. Regarding scheme optimization, it establishes a system process that coordinates the large model with operations research optimization. Operations research optimization optimizes sub-task scheduling and resource allocation based on resource constraints and multi-objective optimization, handling resource competition through a comprehensive influence index and auction mechanism to ensure rational resource allocation. In the scheme evaluation stage, based on performance indicators, it constructs a scientific and objective comprehensive scoring system through Pareto dominance analysis and calculation of relative proximity to the ideal solution, selecting the globally optimal scheme. Furthermore, the comprehensive influence evaluation of sub-tasks and the setting of multi-objective optimization enable this method to flexibly adapt to complex scenarios in multiple fields, effectively addressing issues such as high task complexity, dynamic resource changes, and diverse user needs, significantly improving the intelligence level of task planning and resource allocation efficiency, and possessing broad application value.

[0105] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0106] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0107] In the description of this invention, it should be understood that the terms "first," "second," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0108] In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0109] In the description of this invention, "several" means one or more, and "a large number" means two or more.

[0110] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0111] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0112] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An intelligent task planning method based on large models and operations research optimization, characterized in that, include: S1. Obtain user task requirements and available system resources; S2. Based on user task requirements, generate multiple initial planning schemes using a large model and obtain the corresponding resource requirement list; S3. Based on the system's available resources and resource requirements list, obtain the preferred resource allocation strategy, subtask scheduling plan, and corresponding performance indicators for each initial planning scheme; S4. Based on the obtained performance indicators, obtain the comprehensive score of each initial planning scheme; Select the initial planning scheme with the highest comprehensive score as the final task planning scheme, and output the resource allocation strategy and subtask scheduling plan corresponding to the final task planning scheme; Preferred resource allocation strategies, subtask scheduling plans, and methods for obtaining corresponding performance metrics include: The user task requirements are broken down into independent subtasks, and the resource requirements, time consumption, and dependencies between different subtasks are obtained for each subtask. Based on the dependencies between subtasks, an initial scheduling sequence for the subtasks is generated using topological sorting. Based on the total available resources of the system and the resource requirements of each subtask, resource total constraints are constructed; based on the time consumption data of the subtasks, time constraints are constructed; based on the dependencies between different subtasks, task dependency constraints are constructed. Using the shortest total completion time, lowest total cost, highest average resource utilization, and fewest critical resources as multi-objective optimization objectives, and based on resource constraints, time constraints, and task dependency constraints, operations research optimization methods are used to optimize the initial scheduling sequence of subtasks, resulting in a subtask scheduling plan and resource allocation strategy. Based on the subtask scheduling plan and resource allocation strategy, the performance indicators of the initial planning scheme are calculated, including total completion time, total cost, average resource utilization, and number of critical resources used. During the scheduling optimization of the initial scheduling sequence of subtasks, if multiple subtasks compete for the same resource at the same time, the number of available resources in the current system and the total demand of each subtask for that resource are obtained, and the demand ratio of the total demand to the number of available resources is calculated. A preset ratio threshold is used. If the demand ratio is less than or equal to the preset ratio threshold, resources are allocated from high to low based on the comprehensive influence index of the subtasks. If the demand ratio is greater than the preset ratio threshold, all subtasks competing for the same type of resource within the current time slice or decision cycle are considered as an auction market, with each subtask being a bidder in the market. The negative impact of the delay caused by the subtask's failure to obtain resources on the task objective is obtained, resulting in a cost factor. Based on its own comprehensive influence index and cost factor, the subtask calculates its comprehensive bid in the resource auction, and resources are allocated from high to low based on the size of the subtask's comprehensive bid. The methods for obtaining the overall impact index of subtasks include: Obtain the direct contribution of each subtask to the task objective; construct a task dependency graph based on the dependencies between subtasks, and assign a corresponding dependency weight to each directed edge in the task dependency graph to generate a dependency matrix; calculate the betweenness centrality of each subtask and the sum of the dependency weights of all directed edges received by the node corresponding to each subtask to obtain the weighted in-degree of the subtask; evaluate the influence of the subtasks based on the direct contribution, betweenness centrality, and weighted in-degree using fuzzy logic to obtain the comprehensive influence index of the subtasks. Methods for evaluating the impact of sub-tasks using fuzzy logic include: The direct contribution, betweenness centrality, and weighted in-degree of a subtask are fuzzified using a predefined fuzzy membership function to obtain the fuzzy membership degrees corresponding to these parameters. A pre-defined inference rule base is then used to perform inference operations on the fuzzified fuzzy membership degrees of the subtask to generate the comprehensive fuzzy membership degree of the subtask. Finally, the comprehensive fuzzy membership degree is defuzzified to obtain the comprehensive influence index of the subtask.

2. The intelligent task planning method based on large model and operations research optimization according to claim 1, characterized in that, User task requirements include the task objectives to be achieved, constraints, and a description of the innovativeness of the solution.

3. The intelligent task planning method based on large model and operations research optimization according to claim 2, characterized in that, Multiple initial planning schemes can be generated in the following ways: The task objectives are decoupled into three decision-making dimensions: process advancement, resource strategy, and technical architecture. The innovativeness of user task requirements is evaluated based on preset innovativeness evaluation dimensions to obtain an innovativeness coefficient; The characteristics and threshold ranges of innovation gradient levels are preset, and an innovation coefficient-innovation gradient level mapping table is constructed to convert the innovation coefficient into an innovation gradient level. Input the task objectives, constraints, decision dimensions, and innovation gradient levels into a pre-set large model to generate multiple initial planning schemes.

4. The intelligent task planning method based on large model and operations research optimization according to claim 3, characterized in that, The innovation coefficient can be obtained in the following ways: Each dimension of innovation assessment is assigned a corresponding weight based on the domain of the user's task requirements; Obtain keywords related to each dimension of innovation assessment, establish the correspondence between keywords and score ranges, and form a semantic mapping table; Extract keywords related to innovation from the description of the innovation of the solution; match the extracted keywords with the semantic mapping table to obtain the score corresponding to each innovation evaluation dimension; if there are multiple related descriptions for the same innovation evaluation dimension, calculate the average score of the multiple related descriptions as the final score of the innovation evaluation dimension. The innovation coefficient is calculated by weighting and summing the scores of each innovation evaluation dimension.

5. The intelligent task planning method based on large model and operations research optimization according to claim 4, characterized in that, The methods for obtaining the overall score include: The performance indicators are normalized, and Pareto dominance analysis is performed on the performance indicators of all normalized initial planning schemes to identify and eliminate Pareto-dominated initial planning schemes, generating a candidate scheme set. Based on the candidate scheme set, an ideal optimal solution composed of the optimal values ​​of each performance indicator and an ideal worst solution composed of the worst values ​​of each performance indicator are constructed. Combining the preset weight coefficients of each performance indicator, the weighted optimal solution distance and the weighted worst solution distance of the initial planning schemes in the candidate scheme set to the ideal optimal solution and the weighted worst solution distance are calculated. The relative proximity of the initial planning schemes in the candidate scheme set is calculated based on the weighted optimal solution distance and the weighted worst solution distance. The relative proximity is the comprehensive score of the initial planning scheme.

Images