Electric power scene tool intelligent complex task disassembling method and system

Abstract

The invention belongs to the technical field of task disassembly, and provides an electric power scene tool intelligent complex task disassembly method and system, and the method comprises the steps: carrying out the semantic analysis of an original task text, and extracting an entity, an operation target and a dependency constraint in the original task text; constructing a semantic triple based on the extracted information, and forming a semantic structure chart by taking the constructed semantic triple as an edge of a task atlas and taking a node of the task atlas as task operation; converting and complementing the structure of the semantic structure graph to form a task triple, and fusing the task triple with the current electric power knowledge graph; performing graph path search and logic rule reasoning on the fused electric power knowledge graph by using a pre-trained large language model, and disassembling to generate a subtask sequence conforming to semantic logic and security constraints; and mapping the disassembled subtask sequence chain into a standardized intention instruction, and issuing the standardized intention instruction to each own intelligent control subsystem. According to the method and the device, the accuracy and controllability of task disassembly can be improved.

Description

Technical Field

[0001] This invention belongs to the field of task decomposition technology, specifically relating to a method and system for decomposing complex tasks embodied in power scenarios. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the development of new power systems and smart grids, tasks such as operation, inspection, and maintenance in power scenarios are becoming increasingly complex and dynamic, placing higher demands on the autonomous understanding, decision-making, and execution capabilities of intelligent agents. Traditional task scheduling and execution systems based on rules or shallow semantics struggle to cope with the diversity and semantic ambiguity of task environments.

[0004] In recent years, Large Language Models (LLMs) have demonstrated powerful capabilities in natural language understanding and generation, while knowledge graph technology excels in structured storage and logical reasoning. The combination of these two technologies can provide end-to-end understanding, planning, and controllable execution capabilities for complex tasks, and holds particularly promising applications in embodied intelligence systems.

[0005] However, current technical solutions cannot fully integrate the semantic generalization ability of large language models with the logical controllability of knowledge graphs because the two are inconsistent in paradigm and representation. LLM makes approximate inferences in a continuous vector space, while knowledge graphs are discrete entity-relationship graphs that require precise matching and verifiability. The inference granularity and robustness of the two are naturally misaligned.

[0006] On the other hand, languages ​​have multiple meanings and numerous alternative names for words; it is quite difficult to reliably map the span of text to a unique knowledge graph entity (including cross-language and cross-ontology). Moreover, there is a conflict between the "old facts" in LLM parameters and the "new facts" of the knowledge graph in terms of knowledge updates and timeliness. Summary of the Invention

[0007] To address the aforementioned problems, this invention proposes an embodied intelligent complex task decomposition method and system for power scenarios. This invention can achieve diversity in applicable task environments and improve the accuracy and controllability of task decomposition.

[0008] According to some embodiments, the present invention adopts the following technical solution: A method for decomposing complex tasks embodied in power scenarios includes the following steps: Obtain the original task text; Perform semantic analysis on the original task text to extract entities, operation targets, and dependency constraints from the original task text; Based on the extracted information, semantic triples are constructed, and the constructed semantic triples are used as edges of the task graph. The nodes of the task graph are task operations, forming a semantic structure graph. The semantic structure graph is transformed and completed according to preset rules to form task triples, and the task triples are then integrated with the current power knowledge graph. A pre-trained large language model is used to perform graph path search and logical rule reasoning on the fused power knowledge graph, and to decompose and generate a sequence of subtasks that conform to semantic logic and security constraints. The disassembled subtask sequence chain is mapped into standardized intent commands and sent to each individual intelligent control subsystem to achieve the final disassembly task distribution.

[0009] As an alternative implementation, the process of semantic analysis of the original task text includes: performing semantic analysis on the original task text using a pre-trained large language model or information extraction model to extract entities, operation targets, and dependency constraints, wherein the entities are operation objects, location or device entities, the operation targets are task actions, and the dependency constraints include any one of action sequence relationships, action trigger relationships, and semantic relationships.

[0010] As an alternative implementation, in the process of constructing semantic triples based on the extracted information, if the original task text contains multiple entities, operation targets, or dependency constraints, it is divided into multiple subtasks, the entities, operation targets, and dependency constraints of each subtask are determined, and each subtask is abstracted into a triple.

[0011] As an alternative implementation method, the process of transforming and completing the structure of the semantic structure graph according to preset rules to form task triples includes: standardizing the names of entities in the structure of the semantic structure graph, normalizing the operation target to ensure that the action of the operation target is consistent with the operation concept in the knowledge graph, and reforming the entities, operation targets and dependency constraints in the processed semantic structure graph into task triples.

[0012] As an alternative implementation, the process of fusing the task triples and the current power knowledge graph includes determining whether the task triples and existing triples in the current power knowledge graph are the same or similar. If they are, the existing triples in the current power knowledge graph are retained. If they are different or similar, the different triples in the task triples are incorporated into the current power knowledge graph. The similarity refers to the case where the entity, operation target, and dependency constraint have the same substantive semantics but different names.

[0013] As an alternative implementation method, the process of using a pre-trained large language model to perform graph path search and logical rule reasoning on the fused power knowledge graph includes: using a pre-trained large language model to identify key sub-tasks of the fused power knowledge graph, labeling their intent, target entities and dependency constraints, ensuring that the decomposition order meets the semantic logic and power industry safety regulations, and outputting a standardized task chain, with each sub-task as a line.

[0014] As an alternative implementation, the security constraints include rules extracted from a domain knowledge graph that conforms to power industry standards, operating procedures, and safety regulations.

[0015] As an alternative implementation method, in the process of decomposing and generating a sequence of subtasks that conform to semantic logic and security constraints, time-dependent constraints, spatial location dependencies, and security procedure constraints within the power industry are formed into a constraint set and integrated into the decomposition process in the form of logical formulas.

[0016] As an alternative implementation, the decomposed subtask sequence chain is mapped to standardized intent commands and distributed to each embodied intelligent control subsystem. The final task distribution process includes: formalizing the task decomposition process as a graph transformation and path planning problem, and defining: the original task statement is... Each of them For sentences or semantic units; The resulting graph after entity extraction is a directed multigraph. ,node Represents an entity or event, edge Indicates semantic relationships; The disassembly process is transformed into... Finding task chains that satisfy constraints Each node For each subtask, a formalized decomposition rule is introduced: the dependencies between subtasks satisfy topological order constraints, and the generation of task paths is expressed as a constrained graph search problem for solution.

[0017] A power scenario-embodying intelligent complex task decomposition system includes: The data acquisition module is configured to acquire the raw task text; The semantic analysis module is configured to perform semantic analysis on the original task text and extract entities, operation targets and dependency constraints from the original task text. The semantic structure graph construction module is configured to construct semantic triples based on the extracted information, and use the constructed semantic triples as edges of the task graph. The nodes of the task graph are task operations, thus forming a semantic structure graph. The semantic and graph fusion module is configured to transform and complete the structure of the semantic structure graph according to preset rules to form task triples, and then fuse the task triples with the current power knowledge graph. The subtask decomposition module is configured to use a pre-trained large language model to perform graph path search and logical rule reasoning on the fused power knowledge graph, and decompose it to generate a sequence of subtasks that conform to semantic logic and security constraints. The subtask distribution module is configured to map the disassembled subtask sequence chain into standardized intent commands and distribute them to each specific intelligent control subsystem, thereby achieving the final disassembly task distribution.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention innovatively provides a method for decomposing complex tasks in power scenarios using embodied intelligence. First, task elements are extracted to form triples, and a structure graph is constructed. The structure graph is expressed in the form of entity-relation triples and integrated into a unified power knowledge graph. This enables the connection between task knowledge and professional knowledge, transforming the semantic-level complex natural language task parsing into structured and operable graph nodes. This solves the problems of poor universality and high maintenance cost of traditional template-based task modeling methods in power scenarios.

[0019] This invention innovatively provides a method for decomposing complex tasks in power scenarios using embodied intelligence. The constructed knowledge graph supports graph path search and logical rule reasoning. Based on this, combined with decomposition template prompts generated by a large language model, the method dynamically infers the task decomposition order and dependency structure, generating a sequence of subtasks that conforms to semantic logic and security constraints. This improves the accuracy and contextual consistency of subtask decomposition and avoids the illusion or unstable decomposition paths that may occur when relying solely on a large language model.

[0020] This invention innovatively provides a method for decomposing complex tasks in an embodied intelligent system for power scenarios. The decomposed task chain is mapped into standardized intent instructions and distributed to each embodied subsystem, realizing decoupling and collaboration between multimodal modules. This avoids the expansion obstacles caused by strong coupling between modules of the embodied intelligent agent and improves the generalization adaptability and maintenance convenience of the intelligent agent when facing different task types.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0022] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0023] Figure 1This is a schematic diagram of a method for decomposing complex tasks embodied in an intelligent power scenario, according to one embodiment. Figure 2 This is a schematic diagram illustrating the content extraction process in one embodiment. Figure 3 This is a schematic diagram of a semantic triple in one embodiment; Figure 4 This is a semantic structure diagram of one embodiment; Figure 5 This is a schematic diagram of task triples converted from a semantic structure graph in one embodiment; Figure 6 This is a schematic diagram of the completion result of one embodiment; Figure 7 This is a schematic diagram illustrating knowledge graph fusion in one embodiment. Figure 8 This is a semantic logic diagram of the subtasks decomposed in one embodiment; Figure 9 This is a schematic diagram illustrating the security constraints of subtasks in one embodiment; Figure 10 This is a schematic diagram of the scheduler execution flow in one embodiment. Figure 11 A schematic diagram of a standard response structure in one embodiment; Figure 12 It is a subsystem response structure of one embodiment. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0026] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0027] Where there is no conflict, the embodiments and features described in this application may be combined with each other.

[0028] Example 1 A method for decomposing complex tasks embodied in power scenarios, such as Figure 1As shown, it includes the following steps: Obtain the original task text; Perform semantic analysis on the original task text to extract entities, operation targets, and dependency constraints from the original task text; Based on the extracted information, semantic triples are constructed, and the constructed semantic triples are used as edges of the task graph. The nodes of the task graph are task operations, forming a semantic structure graph. The semantic structure graph is transformed and completed according to preset rules to form task triples, and the task triples are then integrated with the current power knowledge graph. A pre-trained large language model is used to perform graph path search and logical rule reasoning on the fused power knowledge graph, and to decompose and generate a sequence of subtasks that conform to semantic logic and security constraints. The disassembled subtask sequence chain is mapped into standardized intent commands and sent to each individual intelligent control subsystem to achieve the final disassembly task distribution.

[0029] In this embodiment, the process of semantic analysis of the original task text includes: using a pre-trained large language model or information extraction model to perform semantic analysis on the original task text, extracting entities, operation targets and dependency constraints, wherein the entities are operation objects, location or device entities, the operation targets are task actions, and the dependency constraints include any one of action sequence relationships, action triggering relationships and semantic relationships.

[0030] Starting from the original task text, a task semantic structure graph is constructed by using a large language model or information extraction model (such as UIE) or a combination of multiple models to extract elements such as entities, operation targets, and dependency constraints from the task.

[0031] The key to constructing a task semantic structure graph lies in parsing natural language task instructions into a structured graph structure, where each node represents a task entity or operation target, and each edge represents the semantic relationship between them (such as sequence, dependency, triggering). In this embodiment, this process relies on the semantic understanding capability of a Large Language Model (LLM) and the structured extraction capability of an information extraction model (such as UIE) working together.

[0032] For example, the original task text is: "Please first inspect the high-voltage disconnect switch, then check its contact resistance, and finally upload the results and shut down the switch." Task element extraction Large Language Model (LLM) is used to provide syntactic and semantic understanding and role labeling prompts. UIE information extraction model is used to extract entities, actions, and temporal relationships. Specific results are as follows: Figure 2 As shown.

[0033] Based on the above, semantic triples are constructed, where each subtask and its relation are abstracted into a triple, such as... Figure 3 As shown. These triples form the edges of the task graph, and the nodes represent task operations.

[0034] Finally, a directed semantic graph is constructed, such as Figure 4 As shown in the figure, the nodes are actions (such as inspection, detection, etc.) and entities (such as high-voltage disconnect switches); the edges are semantic relationships (such as subsequent orders, operation objects).

[0035] Based on the extracted information, in the process of constructing semantic triples, if the original task text contains multiple entities, operation targets, or dependency constraints, it is divided into multiple subtasks. The entities, operation targets, and dependency constraints of each subtask are determined, and each subtask is abstracted into a triple.

[0036] Next, the semantic structure graph is transformed and completed according to preset rules to form task triples, such as... Figure 5 As shown, specifically, the names of entities in the semantic structure graph are standardized and the operation goals are normalized to ensure that the actions of the operation goals are consistent with the operation concepts in the knowledge graph. The entities, operation goals and dependency constraints in the processed semantic structure graph are then re-formed into task triples.

[0037] During implementation, the task triples are connected to the existing power knowledge graph (including equipment type, operating procedures, and fault modes) to achieve integrated modeling of task semantics and professional knowledge. Specifically, node alignment is required, and the alignment strategy includes two parts: First, equipment name standardization: mapping "high-voltage disconnector" to a standard entry in the graph, such as equipment: high-voltage disconnector_35kV. Second, action normalization: such as "inspection" → operation: patrol, ensuring that the action node is consistent with the operation concept in the graph.

[0038] Then, the relationship expands and is completed.

[0039] After fusion, more implicit knowledge can be supplemented using graph reasoning, such as: Equipment: High-voltage disconnect switch_35kV → (belongs to) → Bay: Main transformer incoming line bay; Operation: Inspection → (to be followed) → Procedure: Inspection operation specifications.

[0040] The result of the completion is as follows Figure 6 As shown.

[0041] like Figure 7As shown, the process of merging the task triples and the current power knowledge graph includes determining whether the task triples and existing triples in the current power knowledge graph are the same or similar. If they are, the existing triples in the current power knowledge graph are retained. If they are different or similar, the different triples in the task triples are incorporated into the current power knowledge graph. The similarity refers to the case where the entity, operation target, and dependency constraint have the same semantic meaning but different names.

[0042] Additionally, triplets that do not meet the rules can be deleted from the generated triplets.

[0043] The constructed knowledge graph supports graph path search and logical rule reasoning. Based on this, and combined with decomposition template hints generated by LLM, the system dynamically reasons the task decomposition order and dependency structure, generating a sequence of subtasks that conforms to semantic logic and security constraints.

[0044] The process of using a pre-trained large language model to perform graph path search and logical rule reasoning on the fused power knowledge graph includes: using the pre-trained large language model to identify key sub-tasks of the fused power knowledge graph, labeling their intent, target entities and dependency constraints, ensuring that the decomposition order satisfies semantic logic and power industry safety regulations, and outputting a standardized task chain, with each sub-task on a separate line.

[0045] The security constraints include rules extracted from a domain knowledge graph that conforms to power industry standards, operating procedures, and safety regulations.

[0046] In the process of decomposing and generating a sequence of subtasks that conform to semantic logic and security constraints, time-dependent constraints, spatial location dependencies, and security procedure constraints within the power industry are incorporated into a constraint set and integrated into the decomposition process in the form of logical formulas.

[0047] For example, disassembly template prompts: [System Settings] You are a task planning expert in the power industry, specializing in breaking down natural language tasks into a sequence of subtasks. You need to combine power operation procedures, safety constraints, and the semantic logic between task entities to complete the task decomposition.

[0048]

enter

Task

[0049] For example, the original task text is "Please complete the inspection, testing, and anomaly reporting of the high-voltage side disconnect switch of main transformer No. 1." The subtasks broken down as follows: Figure 8 As shown.

[0050] Semantic logic mainly refers to the linguistic meaning, task context dependency, sequence, and logical consistency between subtasks, ensuring that the decomposed subtask chain is semantically understandable, reasonable, and orderly.

[0051] Safety constraints are rules extracted from knowledge graphs in areas such as power industry standards, operating procedures, and safety regulations. These constraints ensure that the breakdown and execution of tasks do not violate physical, electrical, or personnel safety regulations. Figure 9 As shown.

[0052] In practical implementation, the "strong constraint" characteristics of the power industry can be modeled as follows: Time-dependent constraints (e.g., inspection must precede testing); Spatial location dependence (e.g., sensors and devices are not in the same interval); Safety regulations (e.g., high-voltage operations require remote assessment).

[0053] These constraints constitute a constraint set. And integrated in the form of logical formulas:

[0054] The meanings of each symbol are as follows: The "candidate solution / status / configuration / process" being validated. For example, a maintenance work plan includes the time sequence, location / interval, equipment and personnel, and operation method.

[0055] A rules set is a set of constraints containing elements. It is a specific strong constraint, such as time dependence: "inspection" before "inspection"; spatial location: sensors and devices must not be linked if they are not in the same interval; safety procedures: high-voltage operation must be judged remotely, etc.

[0056] : Check each constraint in the set one by one.

[0057] : Verify the predicate. Input "constraint" "and the current plan" Output True / False (or a result with a reason), indicating Does this constraint satisfy? For example: "Inspection before testing" Does it meet the requirements? .

[0058] "Sensor A and device B are not in the same interval and cannot be directly closed-loop." Does it meet the requirements? There may be isolation.

[0059] "Operation at ≥10kV requires remote control". Does it meet the requirements? threshold remote.

[0060] (Uppercase conjunctive operator): Logical AND over all constraints. Meaning: The operator performs a logical AND operation only if each constraint is true. All are true. Only then is it true; otherwise, it is false.

[0061] Overall validation result (Boolean value or expandable to "pass / fail + list of failures").

[0062] In this embodiment, a task graph scheduler is introduced to map the decomposed task chain into standardized intent instructions, which are then sent to various subsystems (such as robotic arm control, image recognition, equipment diagnosis, etc.) through an MCP-style protocol (module communication protocol) to achieve decoupling and collaboration between multimodal modules.

[0063] For example, the original task text is: "Complete the inspection, status identification, and anomaly reporting of the No. 1 main transformer high-voltage switch." Sub-task chain: [1] Inspecting the No. 1 main transformer high-voltage disconnect switch (Intent: Inspection, Target: Switch) [2] Detection state (Intent: Recognition, Target: Switch state, Dependency: 1) [3] Report anomalies (Intent: Report, Goal: Diagnostic results, Dependency: 2) Execution process as follows Figure 10 As shown, the modules do not interfere with each other and rely entirely on the scheduler to manage the context and dependency chain.

[0064] In this method, the task decomposition process is formalized as a graph transformation and path planning problem, and the original task statement is defined as follows: Each of them For sentences or semantic units; The resulting graph after entity extraction is a directed multigraph. ,node Represents an entity or event, edge Indicates semantic relationships; The disassembly process is transformed into... Finding task chains that satisfy constraints Each node For each subtask, a formal decomposition rule is introduced: the dependencies between subtasks satisfy the topological order constraint. ; : Dependency edge set. If , indicating subtask Depends on (Must be completed first) Do it again ).

[0065] : A node (subtask) in a task chain.

[0066] In the generated task sequence middle, Location index in Previously, that is: whenever a dependency edge existed... The sequence must include Ranked forward.

[0067] For example, "inspection" Detection "Operation" corresponds to the dependency edges (monitoring, detection) and (detection, operation). Any legitimate task chain must have these nodes appear in this order to satisfy the strong temporal constraints of the procedure.

[0068] The generation of task paths can be formulated as a constrained graph search problem and solved using algorithms such as A*, DFS+Prune, or LLM-based Guided Search (language model-guided path generation).

[0069] In addition, during the above process, knowledge graphs and large language models work together, and LLM provides semantic generalization capabilities to solve the problem of explaining out-of-vocabulary words and new domain tasks; Knowledge graphs provide logical consistency and constraint verification, such as causal logic between entities and equipment operation specifications.

[0070] In this embodiment, a large language model is used to generate a virtual subgraph candidate set. And through knowledge graphs Figure 1 Consistency verification and path pruning.

[0071] Introducing a graph-Augmented Prompting mechanism:

[0072] in, It is a structured textual representation of the task-related subgraph.

[0073] These are prompts for LLM programs, composed of two parts: Text() is a serialization function; Text(G[T]) serializes the data with the current task. Related knowledge graph subgraphs Linearize it into readable text (or a table / list of triples). Purpose: To feed the entities, relationships, attributes, and constraints in the structured graph into the model, ensuring that it "sees" the correct context and logical edges.

[0074] Instruction: Task instructions and output requirements. Tell the LLM "what you need to do, what format to output, and what rules you must follow".

[0075] Typical content includes: task objective, set of available actions, cost / heuristic information (if guiding search), hard / soft constraints, output schema (e.g., JSON), and avoids conjecture.

[0076] For example, Text(G[T]) is as follows: The entity is: - Device A {Type: Circuit Breaker, Voltage = 220kV, Bay = I-12} - Sensor S1 {type=temperature, bay=I-10} - Interval I-12 Adjacent I-10 The relationship is: - Device A connecting bus M1 - Sensor S1 monitors device A (distance=2 bays) The constraints are: - Timing: Inspection < Testing < Operation - Safety: If the voltage is ≥110kV operation.mode=remote - Space: Non-interval linkage requires approval The resulting task instructions are: Task: Generate the shortest operation path (inspection → detection → operation) for device A, satisfying all hard constraints.

[0077] Action set: {Patrol(x), Detect(x), Operate(x, mode)} Lesson learned: Minimize cross-interval movement and approval processes.

[0078] Output: JSON {steps:[{action, target, mode?, reason}], violates:[]} You may not conjecture about missing entities; if this is not met, return a list of violations instead of fictitious steps.

[0079] The concatenated Prompt is equal to the Text(G[T]) + Instruction above. LLM then generates "virtual subgraph candidate paths" (or multiple candidate paths) based on this prompt.

[0080] In addition, this embodiment sets a predetermined E-MCP (Embodied Modular Communication Protocol) structure during the task breakdown and distribution process, such as the standard intent structure:

[0081] Action: What to do (action / instruction name). Examples: Inspect, Measure, Isolate, Operate (remote)... target: The entity or resource to which action is performed. Examples: Device ID, Interval Number, Work Order ID, Graph Entity URI.

[0082] Context: context and constraints (parameters, preconditions, timing, safety procedures, heuristics, etc.).

[0083] Example: {priority: High, must_follow: [Rule#12, Rule#37], deadline: "2025-09-18T12:00Z"} module_id: The identifier (or routing hint) of the subsystem / module expected to perform this action. Examples include: planner, vision_agent, teleop, and safety_guard.

[0084] Presented in JSON format. Figure 11 As shown.

[0085] Subsystem response structure:

[0086] status: An enumeration of execution statuses. Common ones: Accepted (order accepted), Running, Succeeded, Failed, Blocked (stuck by constraints), Aborted (interrupted by the scheduler), Timeout.

[0087] Result: Result load and evidence (data on success, reasons / error codes / list of defaults on failure).

[0088] For example: measured values, generated path steps, verification logs, failed rule_ids, etc.

[0089] timestamp: The time when the feedback occurred.

[0090] Presented in JSON format, such as Figure 12 As shown.

[0091] All modules are connected to the protocol center (i.e., the scheduler), and a state machine rotation and priority scheduling mechanism is adopted.

[0092] Example 2 A power scenario-embodying intelligent complex task decomposition system includes: The data acquisition module is configured to acquire the raw task text; The semantic analysis module is configured to perform semantic analysis on the original task text and extract entities, operation targets and dependency constraints from the original task text. The semantic structure graph construction module is configured to construct semantic triples based on the extracted information, and use the constructed semantic triples as edges of the task graph. The nodes of the task graph are task operations, thus forming a semantic structure graph. The semantic and graph fusion module is configured to transform and complete the structure of the semantic structure graph according to preset rules to form task triples, and then fuse the task triples with the current power knowledge graph. The subtask decomposition module is configured to use a pre-trained large language model to perform graph path search and logical rule reasoning on the fused power knowledge graph, and decompose it to generate a sequence of subtasks that conform to semantic logic and security constraints. The subtask distribution module is configured to map the disassembled subtask sequence chain into standardized intent instructions and distribute them to each specific intelligent control subsystem, thereby achieving the final disassembly task distribution. 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 be implemented in one or more computer-usable storage media (including but not limited to disk storage, etc.) containing computer-usable program code. CD - ROM It takes the form of a computer program product implemented on (such as optical memory, etc.).

[0093] 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.

[0094] 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.

[0095] 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 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0096] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made by those skilled in the art without creative effort within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A power scenario role intelligent complex task disassembly method, characterized in that, Includes the following steps: Obtain the original task text; Perform semantic analysis on the original task text to extract entities, operation targets, and dependency constraints from the original task text; Based on the extracted information, semantic triples are constructed, and the constructed semantic triples are used as edges of the task graph. The nodes of the task graph are task operations, forming a semantic structure graph. The semantic structure graph is transformed and completed according to preset rules to form task triples, and the task triples are then integrated with the current power knowledge graph. A pre-trained large language model is used to perform graph path search and logical rule reasoning on the fused power knowledge graph, and to decompose and generate a sequence of subtasks that conform to semantic logic and security constraints. The disassembled subtask sequence chain is mapped into standardized intent commands and sent to each individual intelligent control subsystem to achieve the final disassembly task distribution.

2. The method for decomposing complex tasks embodied in power scenarios as described in claim 1, characterized in that, The process of semantic analysis of the original task text includes: using a pre-trained large language model or information extraction model to perform semantic analysis on the original task text, extracting entities, operation targets and dependency constraints. Entities are operation objects, location or device entities, operation targets are task actions, and dependency constraints include any one of action sequence relationships, action trigger relationships and semantic relationships.

3. The method for decomposing complex tasks embodied in power scenarios as described in claim 1, characterized in that, Based on the extracted information, in the process of constructing semantic triples, if the original task text contains multiple entities, operation targets, or dependency constraints, it is divided into multiple subtasks. The entities, operation targets, and dependency constraints of each subtask are determined, and each subtask is abstracted into a triple.

4. The method for decomposing complex tasks embodied in power scenarios as described in claim 1, characterized in that, The process of transforming and completing the structure of the semantic structure graph according to preset rules to form task triples includes: standardizing the names of entities in the structure of the semantic structure graph, normalizing the operation goals to ensure that the actions of the operation goals are consistent with the operation concepts in the knowledge graph, and reforming the entities, operation goals and dependency constraints in the processed semantic structure graph into task triples.

5. The method for decomposing complex tasks embodied in power scenarios as described in claim 1, characterized in that, The process of integrating task triples with the current power knowledge graph includes determining whether there are identical or similar triples in the task triples and existing triples in the current power knowledge graph. If so, the existing triples in the current power knowledge graph are retained. If there are different or similar triples, the different triples in the task triples are incorporated into the current power knowledge graph. The similarity refers to the case where the entity, operation target, and dependency constraint have the same semantic meaning but different names.

6. The method for decomposing complex tasks embodied in power scenarios as described in claim 1, characterized in that, The process of using a pre-trained large language model to perform graph path search and logical rule reasoning on the fused power knowledge graph includes: using the pre-trained large language model to identify key sub-tasks of the fused power knowledge graph, labeling their intent, target entities and dependency constraints, ensuring that the decomposition order satisfies semantic logic and power industry safety regulations, and outputting a standardized task chain, with each sub-task on a separate line.

7. The method for decomposing complex tasks embodied in power scenarios as described in claim 1, characterized in that, The security constraints include rules extracted from a domain knowledge graph that conforms to power industry standards, operating procedures, and safety regulations.

8. The method for decomposing complex tasks embodied in power scenarios as described in claim 1, characterized in that, In the process of decomposing and generating a sequence of subtasks that conform to semantic logic and security constraints, time-dependent constraints, spatial location dependencies, and security procedure constraints within the power industry are incorporated into a constraint set and integrated into the decomposition process in the form of logical formulas.

9. The method for decomposing complex tasks embodied in power scenarios as described in claim 1, characterized in that, The process of mapping the decomposed subtask sequence chain into standardized intent commands and issuing them to the respective embodied intelligent control subsystems, ultimately achieving task decomposition and issuance, includes: formalizing the task decomposition process as a graph transformation and path planning problem, and defining: the original task statement as... Each of them For sentences or semantic units; The resulting graph after entity extraction is a directed multigraph. ,node Represents an entity or event, edge Indicates semantic relationships; The disassembly process is transformed into... Finding task chains that satisfy constraints Each node For each subtask, a formalized decomposition rule is introduced: the dependencies between subtasks satisfy topological order constraints, and the generation of task paths is expressed as a constrained graph search problem for solution.

10. A power scenario-embodying intelligent complex task decomposition system, characterized in that, include: The data acquisition module is configured to acquire the raw task text; The semantic analysis module is configured to perform semantic analysis on the original task text and extract entities, operation targets and dependency constraints from the original task text. The semantic structure graph construction module is configured to construct semantic triples based on the extracted information, and use the constructed semantic triples as edges of the task graph. The nodes of the task graph are task operations, thus forming a semantic structure graph. The semantic and graph fusion module is configured to transform and complete the structure of the semantic structure graph according to preset rules to form task triples, and then fuse the task triples with the current power knowledge graph. The subtask decomposition module is configured to use a pre-trained large language model to perform graph path search and logical rule reasoning on the fused power knowledge graph, and decompose it to generate a sequence of subtasks that conform to semantic logic and security constraints. The subtask distribution module is configured to map the disassembled subtask sequence chain into standardized intent commands and distribute them to each specific intelligent control subsystem, thereby achieving the final disassembly task distribution.

Images