Computing task weaving method and related device

Abstract

The invention belongs to the field of electric power automation, and discloses a calculation task weaving method and a related device, which are based on a data association graph, acquire a target node corresponding to a calculation target, find out a calculation task node associated with the target node, and construct a task directed acyclic graph according to the calculation task node, thereby realizing automatic construction of a workflow. The dependency relationship between tasks and resource requirements are fully represented, and local optimum and resource mismatching are avoided. And then obtaining a task key path and associated node floating time, allocating resource priorities to associated nodes based on the key path, quantifying the floating time to guide resource allocation, realizing efficient resource utilization and risk avoidance, and ensuring a key calculation task. And meanwhile, the Pareto optimal leading edge of the associated node is generated during multi-target optimization scheduling, and a group of scheduling schemes are calculated through balancing among multiple targets, so that the flexibility and adaptability of services are greatly improved, different weaving schemes can be output according to different optimization requirements, and efficient scheduling of complex workflows is realized.

Description

Technical Field

[0001] This invention belongs to the field of power automation and relates to a computational task weaving method and related apparatus. Background Technology

[0002] In power grid dispatch automation systems, data computation or business data flow processes are typically organized into workflows consisting of a series of interrelated computational tasks. For example, a typical AI (Artificial Intelligence) model training process may include multiple stages such as data acquisition, data cleaning, feature engineering, model training, model evaluation, and model deployment. Each stage consists of one or more computational tasks, and there are strict sequential orders and data dependencies between them. With the popularization of big data and AI technologies, the scale and complexity of these workflows have increased dramatically, leading to a corresponding increase in the demand for computing resources. Therefore, how to efficiently and economically schedule these complex workflows, maximizing resource utilization and controlling costs while meeting business timeliness requirements, has become a key technical challenge.

[0003] Currently, task orchestration and automated scheduling for workflows are typically managed through workflow management systems. These systems are based on static directed acyclic graphs (DAGs) pre-programmed by developers. This approach lacks flexibility and cannot adapt to dynamic changes in data, resource availability, or business needs. Computational resource requirements rely on manual estimation, and task scheduling uses queue-based fair scheduling or capacity scheduling. This approach cannot automatically optimize resource allocation, and configuring, managing, and optimizing workflows usually requires significant human effort.

[0004] Furthermore, the core design of existing workflow management systems is task orchestration, not resource optimization. They know that task A must execute after task B, but they typically don't know what tasks A and B require or how much computing resources they need. Their scheduling strategies are usually simple "first-come, first-served" or priority-based queues; as long as dependencies are met and there are available execution slots, tasks are submitted. This inevitably leads to resource contention and misallocation. A non-critical task with low computational load and insensitive time might, by chance, preempt a valuable high-performance GPU resource, while a task that truly needs that resource and is on the critical path has to wait in the queue. This "first-come, first-served" rather than "on-demand allocation" model directly results in unnecessary waiting for critical tasks and an increase in the overall project timeline. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a computational task weaving method and related apparatus.

[0006] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a computational task weaving method, comprising: obtaining a computational target and corresponding nodes in a data association graph to obtain a target node; and obtaining all computational task nodes associated with the target node in the data association graph to obtain associated nodes; wherein the nodes in the data association graph include data nodes, computational task nodes, and computational resource nodes; constructing a directed acyclic graph of tasks based on the data association graph and the associated nodes; obtaining the critical path of the tasks and the float time of the associated nodes based on the task directed acyclic graph, allocating resource priorities to the associated nodes based on the critical path of the tasks, and performing multi-objective optimization scheduling of the associated nodes based on the critical path of the tasks, float time, and resource priorities, combined with the data association graph, to obtain the Pareto optimal front of the associated nodes; and obtaining a computational task weaving scheme based on the optimization requirements and the Pareto optimal front of the associated nodes.

[0007] Optionally, the data association graph is constructed using a graph construction method. This method includes: establishing a graph schema for the data association graph; extracting data node entities, data attribute entities, and data lineage relationships from business database tables using a large language model, extracting computing task node entities, computing task attribute entities, and business process relationships from process documents, and extracting computing resource node entities and computing resource attribute entities based on computing resource information to obtain extracted information; and constructing the data association graph based on the extracted information and the graph schema. The computing resource attribute entities include available capacity, unit cost, and availability at each time point.

[0008] Optionally, the step of extracting computational task node entities, computational task attribute entities, and business process relationships from the process document includes: using a large language model to extract task node entities, computational task attribute entities, and business process relationships from the process document in a zero-shot manner; the step of constructing a data association graph based on the extracted information and graph pattern includes: using a large language model to obtain the semantic definition of business process relationships and mapping the semantic definition to the relationship type of the graph pattern, and mapping data lineage relationships to the relationship type of the graph pattern to obtain mapping information; and constructing a data association graph based on the graph pattern according to the data node entities, data attribute entities, computational task node entities, computational task attribute entities, computational resource node entities, computational resource attribute entities, and mapping information.

[0009] Optionally, the graph construction method further includes: obtaining candidate node pairs by acquiring node pairs whose attribute vector embedding similarity is greater than a preset threshold in the data association graph; and using a large language model to perform several predictions of the same entity based on the attribute entity and neighbor node pairs of the candidate node pairs, and when the prediction results in all prediction results are higher than a preset proportion of the candidate node pairs being the same entity, the candidate node pairs are fused.

[0010] Optionally, obtaining the floating time of the task critical path and associated nodes based on the task directed acyclic graph includes: obtaining the floating time of the task critical path and associated nodes based on the task directed acyclic graph using the critical path method or a deep reinforcement learning algorithm.

[0011] Optionally, allocating resource priorities to associated nodes based on the task critical path includes: assigning the priority of associated nodes belonging to the task critical path as a first resource priority, and assigning the remaining associated nodes as a second resource priority; wherein, when a resource conflict occurs between the computing task corresponding to the associated node with the first resource priority and the computing task corresponding to the associated node with the second resource priority, the computing task corresponding to the associated node with the first resource priority is executed first.

[0012] Optionally, the multi-objective optimization scheduling of associated nodes based on the task critical path, float time, and resource priority, combined with the data association graph, includes: dividing associated nodes with continuous execution relationships in the task critical path into a task group, and treating the remaining associated nodes as separate task groups, resulting in several task groups; performing intra-group multi-objective optimization scheduling of each task group based on float time and resource priority, combined with the information of resource nodes in the data association graph, and using a non-dominated sorting genetic algorithm with an elitist strategy to perform inter-group multi-objective optimization scheduling of each task group; wherein, the information of the calculated resource nodes includes available capacity, unit cost, and available time.

[0013] In a second aspect, the present invention provides a computational task weaving system, comprising: an association analysis module, configured to obtain a computational objective and obtain a target node by identifying the corresponding node of the computational objective in a data association graph; and to obtain all computational task nodes associated with the target node in the data association graph to obtain associated nodes; wherein the nodes of the data association graph include data nodes, computational task nodes, and computational resource nodes; a graph construction module, configured to construct a directed acyclic graph of tasks based on the data association graph and the associated nodes; a multi-objective optimization module, configured to obtain the critical path of the task and the float time of the associated nodes based on the task directed acyclic graph, allocate resource priorities to the associated nodes based on the critical path of the task, and perform multi-objective optimization scheduling of the associated nodes based on the critical path of the task, float time, and resource priorities, combined with the data association graph, to obtain the Pareto optimal front of the associated nodes; and an output module, configured to obtain a computational task weaving scheme based on the optimization requirements and the Pareto optimal front of the associated nodes.

[0014] In a third aspect, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described computational task weaving method.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described computational task weaving method.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention presents a computational task weaving method based on a data association graph. It obtains the target node by identifying the node corresponding to the computational objective in the data association graph, and then obtains all computational task nodes associated with the target node in the data association graph to obtain associated nodes. Based on these associated nodes, a directed acyclic graph (DAG) of tasks is constructed, enabling automated workflow construction and fully characterizing the dependencies and resource requirements between computational tasks, thus avoiding local optima and resource misallocation caused by information fragmentation. Next, the critical path of the task and the float time of associated nodes are obtained from the DAG, and resource priorities are allocated to associated nodes based on the critical path. By precisely quantifying the float time of each computational task and using this information to guide resource allocation, efficient resource utilization and risk avoidance are achieved. Resources can be used in a staggered manner, prioritizing limited high-value resources for critical computational tasks, thereby significantly improving resource utilization without extending the overall project duration. Meanwhile, during multi-objective optimization scheduling, Pareto optimal fronts of associated nodes are generated. By weighing multiple objectives, a set of scheduling schemes is calculated, which greatly improves the flexibility and adaptability of the business. Different weaving schemes can be output according to different optimization needs, achieving efficient scheduling of complex workflows. Attached Figure Description

[0017] Figure 1 This is a flowchart of the computational task weaving method according to an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the data association graph construction principle of an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram of the decision generation task for the scheduling and operation auxiliary decision-making service in an embodiment of the present invention.

[0020] Figure 4 This is a block diagram of the computational task weaving system according to an embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] The present invention will now be described in further detail with reference to the accompanying drawings: See Figure 1 In one embodiment of the present invention, a computational task weaving method is provided to significantly improve the execution efficiency and resource utilization of complex workflows, which can be effectively applied to power grid dispatch automation operations.

[0024] Specifically, the computational task weaving method of the present invention includes the following steps: S1: Obtain the computation target and the corresponding node of the computation target in the data association graph to obtain the target node, and obtain all computation task nodes associated with the target node in the data association graph to obtain the associated nodes; wherein, the nodes of the data association graph include data nodes, computation task nodes and computation resource nodes.

[0025] S2: Based on the data association graph, construct a directed acyclic graph of tasks according to the associated nodes.

[0026] S3: Obtain the critical path of the task and the floating time of the associated nodes based on the directed acyclic graph of the task, allocate resource priorities to the associated nodes based on the critical path of the task, and perform multi-objective optimization scheduling of the associated nodes based on the critical path of the task, the floating time and resource priorities, combined with the data association graph, to obtain the Pareto optimal frontier of the associated nodes.

[0027] S4: Based on the optimization requirements and the Pareto optimal frontier of the associated nodes, obtain the computational task weaving scheme.

[0028] This invention presents a computational task weaving method based on a data association graph. It obtains the target node by identifying the node corresponding to the computational objective in the data association graph, and then obtains all computational task nodes associated with the target node in the data association graph to obtain associated nodes. Based on these associated nodes, a directed acyclic graph (DAG) of tasks is constructed, enabling automated workflow construction and fully characterizing the dependencies and resource requirements between computational tasks, thus avoiding local optima and resource misallocation caused by information fragmentation. Next, the critical path of the task and the float time of associated nodes are obtained from the DAG, and resource priorities are allocated to associated nodes based on the critical path. By precisely quantifying the float time of each computational task and using this information to guide resource allocation, efficient resource utilization and risk avoidance are achieved. Resources can be used in a staggered manner, prioritizing limited high-value resources for critical computational tasks, thereby significantly improving resource utilization without extending the overall project duration. Meanwhile, during multi-objective optimization scheduling, Pareto optimal fronts of associated nodes are generated. By weighing multiple objectives, a set of scheduling schemes is calculated, which greatly improves the flexibility and adaptability of the business. Different weaving schemes can be output according to different optimization needs, achieving efficient scheduling of complex workflows.

[0029] In one possible implementation, the data association graph is constructed using a graph construction method. This graph construction method includes: establishing a graph schema for the data association graph; extracting data node entities, data attribute entities, and data lineage relationships from business database tables using a large language model, extracting computing task node entities, computing task attribute entities, and business process relationships from process documents, and extracting computing resource node entities and computing resource attribute entities based on computing resource information to obtain extracted information; and constructing the data association graph based on the extracted information and the graph schema. The computing resource attribute entities include available capacity, unit cost, and availability at each time point.

[0030] Interpretive data association graphs store various data assets (such as database tables and files), computational tasks (such as ETL jobs and AI training scripts), and computational resources (such as database connection pools, GPU cards, and storage systems) in the form of nodes and edges. They also include relationships such as has Lineage To, is Input For, and requires Resource, which constitute a complete trajectory of data flow and resource requirements.

[0031] Interpretively, the first step is to establish a data association graph schema, which defines the node types, attributes, and relationship types of the data association graph. This provides a target structure for subsequent information extraction, ensuring that all extracted knowledge can be systematically integrated into the data association graph. For text-based data, data can be extracted first using a large language model, which then summarizes the graph schema. Graph schemas are merged when they contain data lineage relationships designed from business database tables; otherwise, the graph schema can be expanded.

[0032] In one possible implementation, the step of extracting computational task node entities, computational task attribute entities, and business process relationships from the process document includes: using a large language model to extract task node entities, computational task attribute entities, and business process relationships from the process document in a zero-shot manner; the step of constructing a data association graph based on the extracted information and graph pattern includes: using a large language model to obtain the semantic definition of the business process relationship and mapping the semantic definition to the relationship type of the graph pattern, and mapping the data lineage relationship to the relationship type of the graph pattern to obtain mapping information; and constructing a data association graph based on the graph pattern according to the data node entities, data attribute entities, computational task node entities, computational task attribute entities, computational resource node entities, computational resource attribute entities, and mapping information.

[0033] For example, see Figure 2 Taking the power grid control field as an example, the data sources of the data association map are divided into two parts. The first is the data lineage, which comes from the business database table structure of tools such as data asset planning, data maintenance / collection, and data processing. The second is the business process, which includes the data types, business links, data flow, time-series flow, and business and data flow between upper and lower level dispatchers in the power grid control operation, spot market operation, and big data platform processing. These can be extracted from industry standards, technical books, and software design documents.

[0034] For example, the core business logic, rules, and context of power grid operation exist in a large number of unstructured and semi-structured documents, such as industry standards, technical books, operating specifications, and software design documents. To make this tacit knowledge explicit and integrate it into the knowledge graph, this invention employs a three-stage framework of "extraction-definition-normalization" based on a Large Language Model (LLM) for extraction, specifically including the following steps: Step 1) Extraction: LLM performs open information extraction from document content in a zero-shot manner. It is not limited by predefined patterns and freely identifies all potential entity and relation triples in the text. For example, from a sentence in the design document, “The day-ahead market clearing process depends on the grid topology model and load forecasting data,” LLM might extract the original triples: ['day-ahead market clearing process', 'depends on', 'grid topology model'] and ['day-ahead market clearing process', 'depends on', 'load forecasting data'].

[0035] Step 2) Definition: For the informal relationships and entities extracted in Step 1), prompt the LLM to generate natural language definitions and categorize them. For example, the LLM might categorize "day-ahead market clearing process" as a "Business Process" entity, and define the "dependency" relationship as "indicating that the execution of a process or step requires another data or model as input".

[0036] 3) Normalization: LLM uses these temporary, natural language-defined relations to map onto predefined graph patterns. For example, by comparing semantic definitions, LLM maps various expressions such as "depends on" and "input is" to the unique "CONSUMES" relation in the pattern.

[0037] In one possible implementation, the graph construction method further includes: obtaining candidate node pairs by acquiring node pairs whose attribute vector embedding similarity is greater than a preset threshold in the data association graph; and using a large language model to perform several predictions of the same entity based on the attribute entity and neighbor node pairs of the candidate node pairs, and when the prediction results in all prediction results are higher than a preset proportion of the candidate node pairs being the same entity, the candidate node pairs are fused.

[0038] Interpretive knowledge from both "data lineage" and "business processes" inevitably leads to entity redundancy and ambiguity. For example, the "load forecast data" in the business process design document and the `load_fact` table recorded in the data lineage database actually refer to the same business concept. To construct a single, consistent data association graph, this invention introduces a multi-stage entity alignment framework: First, candidate node pairs are generated. Using vector embedding similarity of entity names, descriptions, and other attributes, candidate node pairs that may refer to the same entity are quickly filtered out. Then, the LLM performs logical reasoning on each candidate entity pair. The reasoning process first focuses on the entity's intrinsic attributes (such as table column names and document definitions), and then extends to their relational context in the graph (i.e., their neighboring nodes). For example, if the "load forecast data" node and the load_fact node are both inputs to the "day-ahead market clearing" process in the graph, the probability that they are the same entity increases significantly. Finally, for each candidate entity pair, the LLM is called multiple times for judgment, with the order of the information in the prompts randomly shuffled each time. When the vast majority of judgments are consistent, a merging operation is performed, effectively solving the illusion problem and position bias problem of the LLM.

[0039] In one possible implementation, the first step in workflow optimization is to accurately represent the workflow itself. This invention automates the process of transforming computational goals into a directed acyclic graph (DAG) of tasks through a data-driven approach. A task DAG is a mathematical model representing a workflow, where nodes represent computational tasks and directed edges represent dependencies between tasks (e.g., computational task A can only begin after computational task B is completed).

[0040] The specific steps include: 1) Receiving the target: The user submits a computation target, such as "Calculate the load forecast data for the State Grid operating area in January 2025" or "Complete the retraining of model Alpha". In the data association graph, this target corresponds to a final entity node, such as Report:load_Forecast or Model:Alpha_v2. 2) Reverse graph traversal: Starting from the target node, a reverse traversal is performed in the data association graph. It recursively searches for all direct or indirect predecessor data entities and computation task nodes that generate these entities along relationships such as has, lineage, and To, which represent data sources. 3) Constructing the task DAG: All computation task nodes identified during the traversal constitute the set of computation task nodes in the task DAG. The dependencies between computation tasks (i.e., a task can only start after another task has been completed) constitute the set of directed edges in the task DAG. Each task node is also associated with its expected execution time and the type and quantity of resources required, which can be based on historical execution data or estimated in advance. In this way, a complete, end-to-end workflow model can be automatically constructed, laying the foundation for subsequent optimization.

[0041] In one possible implementation, obtaining the floating time of the task critical path and associated nodes based on the task directed acyclic graph includes: obtaining the floating time of the task critical path and associated nodes using the critical path method or a deep reinforcement learning algorithm based on the task directed acyclic graph.

[0042] Interpretively, after the task DAG is constructed, the critical path method is used for optimization analysis. The critical path method is used to determine the scheduling flexibility of various tasks in a project and to identify the task sequence that determines the total project duration. This invention innovatively applies it to the data processing workflow.

[0043] Specifically, the analysis process using the critical path method includes the following steps: Step 1: Workflow modeling.

[0044] The workflow is modeled as a directed acyclic graph G=(T, D), where, Let D be the set of computational tasks, and let D be the set of dependencies between tasks. Depends on If the completion of this process is achieved, then there exists a directed edge ( , )in D, each computational task Each has an execution duration. .

[0045] Step 2: Forward propagation calculation.

[0046] Calculate the earliest start time for each task starting from the beginning of the graph. ) and earliest completion time ( For initial missions without any preceding missions. ,That For any other task ,That It is all its direct precursor missions of Maximum value:

[0047] Any computational task of For Including its own duration:

[0048] The shortest total execution time of the entire workflow (i.e. This equals the largest EF value among all computational tasks. .

[0049] Step 3: Backpropagation calculation.

[0050] Calculate the latest start time for each task in reverse order. ) and latest completion time ( For tasks that end without any follow-up tasks. ,That For any other task ,That It is all its direct successor tasks of Minimum value:

[0051] Any computational task of For Subtract its own duration:

[0052] Step 4: Calculate the float time and critical path.

[0053] Floating time ( This indicates the duration by which a computational task can be delayed without affecting the overall project completion time. or .

[0054] The critical path is the path consisting of all tasks with zero float time. The critical path is the longest path in the entire workflow, and its total duration is the shortest completion time for the project.

[0055] For example, deep reinforcement learning (DRL) algorithms can be used instead of the critical path method. DRL can learn an optimal strategy for mapping computational tasks to resources in a given state through interaction with the environment, making it more suitable for highly dynamic and uncertain environments.

[0056] In one possible implementation, allocating resource priorities to associated nodes based on the task critical path includes: assigning the priority of associated nodes belonging to the task critical path as a first resource priority, and assigning the remaining associated nodes as a second resource priority; wherein, when a resource conflict occurs between the computing task corresponding to the associated node with the first resource priority and the computing task corresponding to the associated node with the second resource priority, the computing task corresponding to the associated node with the first resource priority is executed first.

[0057] Explaining the concept, resource balancing aims to resolve resource conflicts (i.e., demand exceeding supply). The core idea is to balance resource demand by adjusting the start and end dates of tasks without changing the overall project duration. The specific method is as follows: First, assign the highest priority to computational tasks on the critical path, ensuring they receive the necessary resources first, as these tasks are the bottleneck of the entire workflow, and any delay will increase the total duration. Then, use float time to resolve conflicts. When resource conflicts occur (e.g., a critical computational task and a non-critical computational task simultaneously require the same GPU), the execution of the non-critical computational task is delayed, utilizing its float time to stagger peak times and thus allocate limited resources to the critical computational task.

[0058] In one possible implementation, the multi-objective optimization scheduling of associated nodes based on the task critical path, float time, and resource priority, combined with the data association graph, includes: dividing associated nodes with continuous execution relationships in the task critical path into a task group, and treating the remaining associated nodes as separate task groups, resulting in several task groups; performing intra-group multi-objective optimization scheduling on each task group based on float time and resource priority, combined with the information of resource nodes calculated in the data association graph; and performing inter-group multi-objective optimization scheduling on each task group using a non-dominated sorting genetic algorithm with an elitist strategy; wherein, the information of the calculated resource nodes includes available capacity, unit cost, and available time.

[0059] Interpretively, in the data association graph, computing resources (such as CPUs, GPUs, and specific servers) are modeled as computing resource nodes, whose attributes include available capacity, unit cost, and available time. By combining real-time data from the monitoring system, the resource availability of computing resource nodes at any point in time can be determined.

[0060] Interpretive scheduling decisions require balancing multiple conflicting objectives, such as execution time or resource consumption. This invention achieves multi-objective optimization scheduling through a hybrid grouping optimization algorithm based on heuristics and evolutionary algorithms. Before scheduling, the topology of the task DAG is analyzed, and heuristic rules are used to divide the tasks into several task groups. Specifically, the task DAG is identified, and consecutive computational tasks on the critical path are grouped into the same task group, while the remaining computational tasks are grouped separately. After grouping, the original task DAG is abstracted into simpler macro-tasks. Intra-group scheduling: Local optimization is performed within each task group. Inter-group scheduling: Each task group is treated as a macro-task, and an evolutionary algorithm (such as the non-dominated sorting genetic algorithm NSGA-II with an elitist strategy) is used to optimize the resource allocation and execution order of these macro-tasks. Through multi-objective optimization scheduling, a Pareto optimal front is generated, which is a set of scheduling schemes, where each scheme is Pareto optimal, meaning that it is impossible to improve any objective without sacrificing at least one objective. For example, one solution might be the fastest to execute but with the highest resource consumption, while another might have the lowest resource consumption but take the longest. Users can flexibly choose based on their current business priorities (e.g., whether they are in a hurry or want to save resources). Through the method of this invention, the management of computing tasks is transformed from a static, sequential execution mode to a dynamic, intelligent weaving mode that considers multiple constraints and performs global optimization, greatly improving the efficiency of complex data processing workflows.

[0061] In one possible implementation, see Figure 3 The present invention provides a computational task weaving method for decision generation tasks in scheduling operation auxiliary decision-making services. First, alarm information is received, then the alarm classification is determined, and then different computational tasks are selected and executed according to the classification identifier, such as obtaining the historical maximum load of the power grid, judging the equipment over-limit situation, or diverting over-limit equipment types, and finally generating a disposal decision.

[0062] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0063] See Figure 4 In another embodiment of the present invention, a computational task weaving system is provided, which can be used to implement the above-mentioned computational task weaving method. Specifically, the computational task weaving system includes an association analysis module, a graph construction module, a multi-objective optimization module, and an output module.

[0064] The system comprises the following modules: **Association Analysis:** The association analysis module obtains the computational objective and its corresponding node in the data association graph (resulting in the target node), and obtains all computational task nodes associated with the target node in the data association graph (resulting in the associated nodes). The nodes in the data association graph include data nodes, computational task nodes, and computational resource nodes. **Graph Construction:** The graph construction module constructs a directed acyclic graph (DAG) based on the data association graph and associated nodes. **Multi-Objective Optimization:** The multi-objective optimization module obtains the critical path of the task and the float time of the associated nodes from the DAG, allocates resource priorities to the associated nodes based on the critical path, and performs multi-objective optimization scheduling of the associated nodes based on the critical path, float time, and resource priorities, combined with the data association graph, to obtain the Pareto optimal front of the associated nodes. **Output:** The output module obtains a computational task weaving scheme based on the optimization requirements and the Pareto optimal front of the associated nodes.

[0065] In one possible implementation, the data association graph is constructed using a graph construction method. This graph construction method includes: establishing a graph schema for the data association graph; extracting data node entities, data attribute entities, and data lineage relationships from business database tables using a large language model, extracting computing task node entities, computing task attribute entities, and business process relationships from process documents, and extracting computing resource node entities and computing resource attribute entities based on computing resource information to obtain extracted information; and constructing the data association graph based on the extracted information and the graph schema. The computing resource attribute entities include available capacity, unit cost, and availability at each time point.

[0066] In one possible implementation, the step of extracting computational task node entities, computational task attribute entities, and business process relationships from the process document includes: using a large language model to extract task node entities, computational task attribute entities, and business process relationships from the process document in a zero-shot manner; the step of constructing a data association graph based on the extracted information and graph pattern includes: using a large language model to obtain the semantic definition of the business process relationship and mapping the semantic definition to the relationship type of the graph pattern, and mapping the data lineage relationship to the relationship type of the graph pattern to obtain mapping information; and constructing a data association graph based on the graph pattern according to the data node entities, data attribute entities, computational task node entities, computational task attribute entities, computational resource node entities, computational resource attribute entities, and mapping information.

[0067] In one possible implementation, the graph construction method further includes: obtaining candidate node pairs by acquiring node pairs whose attribute vector embedding similarity is greater than a preset threshold in the data association graph; and using a large language model to perform several predictions of the same entity based on the attribute entity and neighbor node pairs of the candidate node pairs, and when the prediction results in all prediction results are higher than a preset proportion of the candidate node pairs being the same entity, the candidate node pairs are fused.

[0068] In one possible implementation, obtaining the floating time of the task critical path and associated nodes based on the task directed acyclic graph includes: obtaining the floating time of the task critical path and associated nodes using the critical path method or a deep reinforcement learning algorithm based on the task directed acyclic graph.

[0069] In one possible implementation, allocating resource priorities to associated nodes based on the task critical path includes: assigning the priority of associated nodes belonging to the task critical path as a first resource priority, and assigning the remaining associated nodes as a second resource priority; wherein, when a resource conflict occurs between the computing task corresponding to the associated node with the first resource priority and the computing task corresponding to the associated node with the second resource priority, the computing task corresponding to the associated node with the first resource priority is executed first.

[0070] In one possible implementation, the multi-objective optimization scheduling of associated nodes based on the task critical path, float time, and resource priority, combined with the data association graph, includes: dividing associated nodes with continuous execution relationships in the task critical path into a task group, and treating the remaining associated nodes as separate task groups, resulting in several task groups; performing intra-group multi-objective optimization scheduling on each task group based on float time and resource priority, combined with the information of resource nodes calculated in the data association graph; and performing inter-group multi-objective optimization scheduling on each task group using a non-dominated sorting genetic algorithm with an elitist strategy; wherein, the information of the calculated resource nodes includes available capacity, unit cost, and available time.

[0071] All relevant content of each step involved in the aforementioned embodiments of the computational task weaving method can be referenced to the functional description of the corresponding functional module of the computational task weaving system in the embodiments of the present invention, and will not be repeated here.

[0072] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0073] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of a computational task weaving method.

[0074] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the computational task weaving method in the above embodiments.

[0075] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0076] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0077] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0078] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A computational task weaving method, characterized in that, include: The process involves obtaining the computation target and its corresponding node in the data association graph to obtain the target node, and obtaining all computation task nodes associated with the target node in the data association graph to obtain the associated nodes; wherein, the nodes in the data association graph include data nodes, computation task nodes, and computation resource nodes; Based on the data association graph, a directed acyclic graph of tasks is constructed according to the associated nodes; The task critical path and the floating time of associated nodes are obtained from the task directed acyclic graph. Resource priorities are allocated to associated nodes based on the task critical path. Multi-objective optimization scheduling of associated nodes is performed based on the task critical path, floating time and resource priorities, combined with the data association graph, to obtain the Pareto optimal frontier of associated nodes. Based on the optimization requirements and the Pareto optimal frontier of the associated nodes, a computational task weaving scheme is obtained.

2. The computational task weaving method according to claim 1, characterized in that, The data association graph is constructed using a graph construction method, which includes: establishing a graph pattern for the data association graph; extracting data node entities, data attribute entities, and data lineage relationships from business database tables using a large language model, extracting computing task node entities, computing task attribute entities, and business process relationships from process documents, and extracting computing resource node entities and computing resource attribute entities based on computing resource information to obtain extracted information; and constructing the data association graph based on the extracted information and the graph pattern. The computing resource attribute entities include available capacity, unit cost, and availability at each point in time.

3. The computational task weaving method according to claim 2, characterized in that, The extraction of computation task node entities, computation task attribute entities, and business process relationships from the process document includes: A large language model is used to extract task node entities, computation task attribute entities, and business process relationships from process documents in a zero-shot manner. The process of constructing a data association graph based on extracted information and graph patterns includes: using a large language model to obtain the semantic definition of business process relationships and mapping the semantic definition to the relationship type of the graph pattern, as well as mapping data lineage relationships to the relationship type of the graph pattern, to obtain mapping information; and constructing a data association graph based on the graph pattern according to data node entities, data attribute entities, computing task node entities, computing task attribute entities, computing resource node entities, computing resource attribute entities, and the mapping information.

4. The computational task weaving method according to claim 2, characterized in that, The graph construction method further includes: obtaining candidate node pairs by acquiring node pairs whose attribute vector embedding similarity is greater than a preset threshold in the data association graph; and using a large language model to perform several predictions of the same entity based on the attribute entity and neighbor node pairs of the candidate node pairs, and when the prediction results in all prediction results are higher than a preset proportion of the candidate node pairs being the same entity, the candidate node pairs are fused.

5. The computational task weaving method according to claim 1, characterized in that, The floating time for obtaining the critical path and associated nodes of a task based on a directed acyclic graph includes: Based on the directed acyclic graph of the task, the critical path method or deep reinforcement learning algorithm is used to obtain the critical path of the task and the floating time of related nodes.

6. The computational task weaving method according to claim 1, characterized in that, The allocation of resource priorities to associated nodes based on the critical path of the task includes: Assign the priority of associated nodes belonging to the critical path of the task as the first resource priority, and assign the remaining associated nodes as the second resource priority; wherein, when there is a resource conflict between the computing task corresponding to the associated node with the first resource priority and the computing task corresponding to the associated node with the second resource priority, the computing task corresponding to the associated node with the first resource priority shall be executed first.

7. The computational task weaving method according to claim 1, characterized in that, The multi-objective optimization scheduling of associated nodes based on the task's critical path, floating time, and resource priority, combined with the data association graph, includes: The task groups are divided into several task groups by grouping related nodes with continuous execution relationships in the critical path of the task and keeping the remaining related nodes as separate task groups. Based on the floating time and resource priority, and combined with the information of resource nodes in the data association graph, multi-objective optimization scheduling is performed within each task group. A non-dominated sorting genetic algorithm with an elitist strategy is used to perform multi-objective optimization scheduling between task groups. The information of the resource nodes includes available capacity, unit cost, and available time.

8. A computational task weaving system, characterized in that, include: The association analysis module is used to obtain the computing target and the corresponding node in the data association graph to obtain the target node, and to obtain all computing task nodes associated with the target node in the data association graph to obtain the association node; wherein, the nodes in the data association graph include data nodes, computing task nodes and computing resource nodes; The graph construction module is used to construct a directed acyclic graph of tasks based on the data association graph and the associated nodes. The multi-objective optimization module is used to obtain the critical path of the task and the floating time of the associated nodes based on the directed acyclic graph of the task, allocate resource priorities to the associated nodes based on the critical path of the task, and perform multi-objective optimization scheduling of the associated nodes based on the critical path of the task, the floating time and resource priorities, combined with the data association graph, to obtain the Pareto optimal frontier of the associated nodes. The output module is used to obtain the computational task weaving scheme based on the optimization requirements and the Pareto optimal frontier of the associated nodes.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the computational task weaving method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the computational task weaving method as described in any one of claims 1 to 7.

Images