A dual-agent processing method and system based on business decomposition and visualization.

Abstract

This invention discloses a processing method and system based on a dual-agent approach for business decomposition and visualization. The processing method includes: a business decomposition agent receiving a natural language request from a power system; performing semantic parsing of the natural language request using a small language model; performing semantic retrieval based on a power system knowledge graph to achieve knowledge-enhanced intent recognition and task decomposition; and generating a corresponding task instruction chain using a directed acyclic graph structure. The task instruction chain is then sent to an instruction registration center, which schedules a visualization agent to generate an interactive dashboard and natural language summary using a large language model. The technical solution of this application achieves automatic understanding and efficient decomposition of power system business, improves the intelligence and automation level of data processing and visualization, enhances the system's task scheduling and management capabilities, effectively reduces manual intervention, and improves operation and maintenance efficiency and the accuracy of data display.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent power systems, and in particular to a processing method and system based on a dual-agent approach of business decomposition and visualization. Background Technology

[0002] In the current operation of power systems, with the continuous growth in the number of devices and the volume of data, power companies face multiple challenges, including data heterogeneity, complex business processes, and low efficiency of manual processing. Traditional data analysis and business processing methods rely heavily on manual intervention, making it difficult to efficiently respond to diverse business needs. Especially in scenarios such as fault diagnosis, equipment status monitoring, and operational analysis, problems such as information dispersion, data silos, and fragmented business response processes exist, leading to delayed operation and maintenance decisions and increased management costs. In recent years, although some systems have introduced rule-based or simple automated data processing platforms, existing technologies still have many limitations in terms of automatic understanding, intelligent task decomposition, and efficient presentation of analysis results when faced with complex and dynamic business needs and diverse questions raised by users through natural language. These limitations make it difficult to meet the actual needs of intelligent, automated, and refined management of power systems.

[0003] With the rapid development of artificial intelligence and natural language processing technologies, technologies that enable the automatic decomposition and data visualization of complex business processes through intelligent agents are gradually emerging. New-generation intelligent methods, such as multi-agent collaboration mechanisms, domain knowledge graph-driven knowledge enhancement retrieval, task modeling using directed acyclic graph structures, and finite state machine-based process management, provide feasible paths for the semantic understanding, intelligent decomposition, and automatic execution of large-scale power system data. However, the industry currently lacks an end-to-end intelligent processing solution that can deeply integrate business decomposition agents with visualization agents, combining knowledge graphs, task instruction chains, and lifecycle management to achieve multi-dimensional data visualization from user natural language requests. Therefore, there is an urgent need for an intelligent task decomposition and visualization method for power system business to improve the intelligence level of power system data processing. Summary of the Invention

[0004] This invention provides a processing method and system based on a dual-agent approach of business decomposition and visualization, which enables automatic understanding and efficient decomposition of power system business, improves the intelligence and automation level of data processing and visualization, enhances the system's task scheduling and management capabilities, effectively reduces manual intervention, and improves the accuracy of data display.

[0005] According to a first aspect of the present invention, a processing method based on a dual-agent approach of business decomposition and visualization is provided, the method comprising:

[0006] The business decomposition intelligent agent receives natural language requests from the power system, performs semantic parsing of the natural language requests through a small language model, performs semantic retrieval based on the power system knowledge graph, realizes knowledge-enhanced intent recognition and task decomposition, and generates corresponding task instruction chains in a directed acyclic graph structure.

[0007] The task instruction chain is sent to the instruction registration center, which then schedules the visualization presentation agent to generate an interactive dashboard and natural language summary through a large language model, thereby completing the visualization presentation of power system data.

[0008] The instruction registration center uses a finite state machine to manage the entire processing flow throughout its lifecycle, thereby constraining the execution flow of the task instruction chain.

[0009] In one embodiment, it also includes:

[0010] The natural language request is semantically parsed using a small language model.

[0011] Based on the power system knowledge graph, perform enhanced semantic retrieval on the semantic parsing results to obtain power system domain knowledge related to the request;

[0012] The acquired power system domain knowledge is applied to the understanding of the natural language request, and a corresponding task instruction chain is generated.

[0013] In one embodiment, it also includes:

[0014] Pre-train on a general corpus to obtain an initial language model;

[0015] The initial language model is trained based on data from the power system domain to obtain a smaller language model optimized for the power system domain.

[0016] The trained small language model is deployed into the business decomposition agent for intent recognition and task decomposition of natural language requests.

[0017] In one embodiment, it also includes:

[0018] Each node in the task instruction chain corresponds to a power system data analysis subtask or visualization subtask.

[0019] In the directed acyclic graph structure, the directed connections between nodes represent the dependencies between the subtasks;

[0020] The directed acyclic graph structure ensures that there are no circular dependencies during the execution of the task instruction chain.

[0021] In one embodiment, it also includes:

[0022] After receiving the task instruction chain generated by the business decomposition intelligent agent, each sub-task instruction is registered as waiting to be executed;

[0023] When all the prerequisite conditions of a subtask are met, the subtask is switched to the execution state and executed by the visual agent.

[0024] After a subtask is completed, it is switched to the completed state. If an error occurs during execution, it is switched to the error state to trigger the corresponding exception handling.

[0025] In one embodiment, it also includes:

[0026] The visualization-related instructions in the task instruction chain are parsed into interface components of an interactive dashboard to display the power system data analysis results;

[0027] A natural language summary report describing the analysis results is generated based on the large-scale language model.

[0028] The interactive dashboard enables users to interactively query and dynamically visualize power system data.

[0029] According to a second aspect of the present invention, a processing system based on a dual-agent approach of business decomposition and visualization is provided, comprising:

[0030] The decomposition module is used to receive natural language requests from the power system by the business decomposition intelligent agent, perform semantic parsing of the natural language requests through a small language model, perform semantic retrieval based on the power system knowledge graph, realize knowledge-enhanced intent recognition and task decomposition, and generate corresponding task instruction chains in a directed acyclic graph structure.

[0031] The visualization module is used to send the task instruction chain to the instruction registration center, which then schedules the visualization presentation agent to generate interactive dashboards and natural language summaries through a large language model, thereby completing the visualization presentation of power system data.

[0032] The management module is used to manage the entire processing flow through the instruction registration center using a finite state machine, so as to constrain the execution flow of the task instruction chain.

[0033] In one embodiment, the disassembly module, the visualization module, and the management module are controlled to execute any of the above-described processing methods based on a dual-agent approach of business disassembly and visualization.

[0034] According to a third aspect of the present invention, an electronic device is provided, comprising: a communication interface, a processor, and a memory;

[0035] The memory is used to store program instructions, which, when executed by the processor that is connected to the memory via the communication interface, implement any of the above-mentioned processing methods based on business decomposition and visualization of dual intelligent agents.

[0036] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored, which, when executed by a computer (e.g., a processor in the computer), implement any of the above-described processing methods based on a dual-agent approach of business decomposition and visualization.

[0037] In summary, this invention provides a processing method and system based on a dual-agent approach of business decomposition and visualization. The method includes: a business decomposition agent receiving natural language requests from a power system; performing semantic parsing of the natural language requests using a small language model; performing semantic retrieval based on a power system knowledge graph to achieve knowledge-enhanced intent recognition and task decomposition; and generating a corresponding task instruction chain using a directed acyclic graph structure. The task instruction chain is then sent to an instruction registry center, which schedules a visualization agent to generate an interactive dashboard and natural language summary using a large language model, thereby completing the visualization of power system data. The instruction registry center uses a finite state machine to manage the entire processing flow throughout its lifecycle, constraining the execution flow of the task instruction chain. This technical solution, through the collaboration of the business decomposition agent and the visualization agent, achieves intelligent understanding, knowledge enhancement, and automated task decomposition of natural language requests from the power system, effectively improving the response speed and processing intelligence level of business requirements. The use of a directed acyclic graph structure for task modeling, combined with the unified scheduling of the instruction registry center and the lifecycle management of the finite state machine, ensures the efficient execution and process controllability of each sub-task. The overall solution can significantly improve the automation, accuracy, and user experience of power system data analysis and visualization, while significantly reducing manual intervention and management costs.

[0038] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and drawings.

[0039] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0040] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0041] Figure 1 A flowchart of a dual-agent processing method based on business decomposition and visualization is provided for embodiments of the present invention;

[0042] Figure 2 A flowchart of another processing method based on business decomposition and visualization of dual intelligent agents provided as an embodiment of the present invention;

[0043] Figure 3 A flowchart of another processing method based on business decomposition and visualization of dual intelligent agents provided for embodiments of the present invention;

[0044] Figure 4 A flowchart of another processing method based on business decomposition and visualization of dual intelligent agents provided as an embodiment of the present invention;

[0045] Figure 5 A flowchart of another processing method based on business decomposition and visualization of dual intelligent agents provided for embodiments of the present invention;

[0046] Figure 6 A flowchart of another processing method based on business decomposition and visualization of dual intelligent agents provided for embodiments of the present invention;

[0047] Figure 7 A structural diagram of a dual-agent processing system based on business decomposition and visualization is provided for embodiments of the present invention;

[0048] Figure 8 This is a structural diagram of an electronic device provided as an embodiment of the present invention. Detailed Implementation

[0049] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0050] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0051] like Figure 1 As shown, this invention provides a processing method based on a dual-agent approach of business decomposition and visualization, which includes:

[0052] In step S11, the business decomposition intelligent agent receives the natural language request from the power system, performs semantic parsing on the natural language request through a small language model, performs semantic retrieval based on the power system knowledge graph, realizes knowledge-enhanced intent recognition and task decomposition, and generates the corresponding task instruction chain in a directed acyclic graph structure.

[0053] In step S12, the task instruction chain is sent to the instruction registration center, which then schedules the visualization presentation agent to generate an interactive dashboard and natural language summary through a large language model to complete the visualization presentation of power system data.

[0054] In step S13, the instruction registration center uses a finite state machine to manage the entire processing flow throughout its lifecycle, thereby constraining the execution flow of the task instruction chain.

[0055] In one embodiment, intelligent understanding and interactive result display of natural language requests are achieved through multi-agent collaborative work and knowledge-enhanced analysis. Based on a dual-agent architecture of a business decomposition agent and a visualization agent, the processing flow is uniformly coordinated by an instruction registry. The business decomposition agent receives natural language requests from the power system domain, performs semantic parsing of the requests using a small language model, and performs semantic retrieval using a knowledge graph of the power system domain. This enables knowledge-enhanced user intent recognition and task decomposition, generating a corresponding task instruction chain in a directed acyclic graph (DAG) structure. The task instruction chain is sent to the instruction registry, which schedules the visualization agent to execute subsequent steps. The visualization agent processes the received task instruction chain using a large language model, generating interactive dashboards (such as dynamic charts and indicator panels) related to power system data, along with corresponding natural language summaries, ultimately achieving the visualization and intelligent interpretation of the power system data requested by the user. Throughout the entire process, the finite state machine (FSM) built into the instruction registry center manages the lifecycle of the task instruction chain, ensuring that each stage of the task chain proceeds in a controlled manner according to the predetermined state transition sequence, constraining and supervising the execution process of the task chain, and ensuring the stability and reliability of the system processing flow.

[0056] Through the aforementioned architecture, the business decomposition agent, acting as the front-end analysis module, is responsible for transforming user-submitted natural language requests into executable task chains. Utilizing a small language model for semantic understanding, it can parse key intents and required business operations from user requests. Simultaneously, it leverages the semantic retrieval capabilities of the power system knowledge graph to query and compare relevant technical terms, equipment parameters, or historical data related to the request, refining and supplementing task information through knowledge enhancement. Based on the semantic parsing results and the information returned by the knowledge graph, the business decomposition agent automatically breaks down complex requirements into a series of sub-task nodes with causal dependencies, organizing these task nodes into an instruction chain using a Directed Acyclic Graph (DAG) structure. Through DAG modeling, the nodes and their dependencies in the task chain are explicitly represented, avoiding circular dependencies and ensuring the logical consistency and completeness of the task execution flow. The DAG structure also lays the foundation for parallel processing and dynamic adjustment of tasks; for example, some independent sub-tasks can be executed in parallel, improving overall execution efficiency. The task instruction chains generated by the business decomposition agent include, but are not limited to, various types of sub-tasks such as data acquisition, analysis and calculation, and result generation, providing context parameters and execution objectives for each task node.

[0057] The agent acts as the central hub for task chain distribution and process control. After the DAG task instruction chain generated by the business decomposition agent is submitted to the agent, the agent automatically invokes the corresponding visual representation agent instance to enter the execution phase based on the task chain description. The agent uses an internal finite state machine (FSM) to strictly manage the entire execution process: for example, when the task chain is in a pending state, the agent triggers the visual representation agent to begin execution; during subtask execution, the FSM tracks the state transitions of task nodes, ensuring that the next phase only begins after all dependent tasks are completed; if a task fails or the output does not meet expectations, the FSM can transfer the process to an exception handling or retry state according to predefined strategies, ensuring the orderly progress of the processing flow. With the help of the FSM's state transition control, the instruction executor effectively constrains the execution path and rhythm of the task instruction chain, preventing out-of-order execution or infinite waiting, and achieving refined control over complex processes. Meanwhile, the instruction executor also maintains an instruction registry to record descriptions of various task instructions and their corresponding execution modules. This facilitates the dynamic search and invocation of suitable agents or tools to complete specific tasks, thereby achieving decoupled collaboration among the system's internal components. The business decomposition agents and the visualization agents do not interact directly; instead, the task chain is connected from planning to execution through the instruction executor's scheduling, ensuring clear relationships and division of labor among internal components.

[0058] The visualization agent, acting as a backend execution module, focuses on presenting and summarizing power data according to a chain of task instructions. When the instruction registry schedules this agent, it transmits the currently executable tasks authorized by the FSM and their required data context. The large language model built into the visualization agent receives these task instructions and selects appropriate presentation templates and generation strategies based on the instruction type: for data tasks requiring graphical display, the agent can automatically draw corresponding interactive dashboard components, such as trend line charts, pie charts, and geographic distribution maps; simultaneously, leveraging the powerful natural language generation capabilities of the large language model, the visualization agent also generates textual descriptions and business insight summaries of the visualization results. For example, in a power load analysis scenario, the visualization agent can automatically generate dynamic line charts of load curves based on task chain instructions, along with natural language descriptions of peak load periods and the causes of abnormal fluctuations, allowing users to intuitively understand the data through visualization and obtain professional interpretations through textual explanations. Throughout the visualization generation process, the agent can utilize a predefined template library and domain knowledge to adjust the dashboard layout and style, achieving an adaptive dashboard template generation strategy: automatically selecting the optimal chart type and layout based on different data types and user needs, improving the clarity and relevance of the presentation. The large language model not only ensures the fluency and professionalism of the natural language summary but also verifies the generated content using information returned from the knowledge graph, avoiding factual errors and ensuring the accuracy and reliability of the output results. The visualization agent then feeds the generated dashboard and summary back to the instruction registry center, which summarizes the results and returns them to the user, achieving a fully automated response from natural language queries to visualization analysis conclusions.

[0059] Traditional rule-based configuration methods often require manual pre-definition of rules or templates for power business queries and displays, lacking flexibility for new query requirements and making timely expansion difficult. Furthermore, rule-based systems typically can only handle a limited number of known question patterns and cannot understand complex and varied natural language expressions. This embodiment, however, utilizes semantic parsing driven by a small language model and knowledge graph enhancement to enable the system to flexibly understand various natural language requests and automatically plan task chains, eliminating reliance on manual rules and achieving deep semantic understanding of user intent and dynamic task generation. In addition, single-model solutions (e.g., relying solely on a large language model to directly generate answers) suffer from knowledge blind spots and opaque reasoning chains when handling specialized power issues: while large general-purpose models possess powerful language generation capabilities, they lack specialized knowledge in the power field and real-time data access, easily leading to inaccurate or unreliable answers. Moreover, allowing a single model to independently complete all steps from intent recognition to data extraction to result generation lacks modular process control and makes timely error correction and optimization difficult. In contrast, this embodiment employs a dual-agent collaborative mechanism, with two dedicated agents handling task planning and result presentation respectively, forming a closed-loop feedback loop under the coordination of the instruction registry center. The modular decoupled architecture not only fully leverages the advantages of models of different scales—smaller models quickly and accurately understand domain semantics, while larger models generate rich expressions and visualizations—but also achieves monitoring and scheduling of each stage of the process through a finite state machine, ensuring the controllability and robustness of system behavior. Therefore, compared to rigid rule-based systems or black-box single-model systems in existing technologies, this invention can complete complex power data query tasks with higher intelligence density and reliability. It possesses adaptive semantic understanding and knowledge reasoning capabilities not found in rule-based solutions, while avoiding content distortion and process control issues that may occur with single-model solutions, significantly improving the automation level and user experience of power data services. Furthermore, a new mechanism for dual-agent collaborative work is proposed, where the business decomposition agent and the visualization presentation agent each perform their respective functions while cooperating with each other. Information sharing and process integration are achieved through an instruction registry center and a finite state machine (FSM). For example, when generating dashboards and reports, the visualization presentation agent can provide preliminary data analysis feedback to the business decomposition agent as a basis for dynamic adjustment of the task chain, realizing a feedback synchronization mechanism between agents and continuously optimizing task execution strategies. A Directed Acyclic Graph (DAG) structure is introduced to model the task chain, enabling non-linear organization and dependency management of subtasks. Compared to linear sequential execution, the DAG task chain supports concurrent execution and flexible scheduling of subtasks. When power data queries involve multiple independent calculations and analyses, the system can perform different subtasks simultaneously, improving efficiency. Even if new information or environmental changes occur during task execution, the DAG task chain can be dynamically adjusted according to the Functional Graph Synthesis (FSM) strategy (e.g., modifying task priorities or adding new subtask nodes), thereby maintaining optimal responsiveness to user needs.The entire process is controlled by a Finite State Machine (FSM) in the instruction registry center, formalizing the complex and ever-changing task execution process into several controlled states, effectively improving the security and predictability of system operation. FSM control not only allows for the definition of advanced behaviors such as loop waiting and timeout retries, enabling self-healing and fault tolerance in the face of anomalies, but also enables the system to repeatedly execute sub-processes according to expected strategies, meeting the needs of scenarios requiring periodic operation, such as power monitoring. Through knowledge graph-enhanced analysis, this invention introduces professional knowledge from the power field into the natural language processing and task planning stages, greatly improving the accuracy of the system's understanding and handling of professional problems. The knowledge graph provides domain entity relationships and rule constraints, allowing the business decomposition agent to refer to background knowledge, professional terminology explanations, and historical correlation data in the graph when parsing user intent, thereby generating a more reasonable and complete task chain. In the result generation stage, the knowledge graph can also assist large-scale models in verifying key data and conclusions, ensuring the professional correctness of dashboard and summary content. The dynamic adjustment mechanism of the task chain enables the system to adapt to new situations and perform secondary planning in a timely manner. The feedback synchronization between intelligent agents ensures the consistency between the analysis results and the presented content and achieves closed-loop optimization. The adaptive dashboard template generation strategy automatically selects the visualization method based on user preferences and data characteristics, thereby improving the effectiveness and aesthetics of information presentation.

[0060] The technical solution in this embodiment achieves intelligent understanding, knowledge enhancement, and automated task decomposition of natural language requests from the power system through the collaboration of business decomposition intelligent agents and visualization presentation intelligent agents. This effectively improves the response speed and processing intelligence level of business needs. A directed acyclic graph structure is used for task modeling, combined with unified scheduling by an instruction registry center and lifecycle management by a finite state machine, ensuring efficient execution and process controllability of each sub-task. The overall solution significantly improves the automation, accuracy, and user experience of power system data analysis and visualization, while significantly reducing manual intervention and management costs.

[0061] In one embodiment, such as Figure 2 As shown, it also includes the following steps S21-S23:

[0062] In step S21, a small language model is used to perform semantic parsing on the natural language request;

[0063] In step S22, a retrieval-enhanced semantic retrieval is performed on the semantic parsing results based on the power system knowledge graph to obtain power system domain knowledge related to the request;

[0064] In step S23, the acquired power system domain knowledge is applied to the understanding of the natural language request, and a corresponding task instruction chain is generated.

[0065] In one embodiment, the intelligent understanding and automatic business decomposition of natural language requests in the power system domain is divided into three stages: semantic parsing, knowledge enhancement, and task generation. A domain-optimized small language model is used to perform semantic parsing of the user's natural language requests. This small language model, pre-trained and fine-tuned with a large corpus of power system business language, can efficiently identify common power industry terms, equipment names, business scenarios, and complex expressions, thereby accurately converting the user's natural language input into structured semantic expressions. For example, when faced with a request such as "analyze and visualize the load changes of the transformer at substation B during the summer," the model can not only accurately identify key entities and intents such as "transformer," "load change," and "summer," but also distinguish the multiple business requirements, including analysis and visualization, contained in the request.

[0066] Building upon semantic parsing, this system further leverages a power system knowledge graph to enhance semantic retrieval of the parsing results. The power system knowledge graph aggregates structured domain knowledge, including equipment information, operating rules, historical cases, data sensors, and their relationships. By combining semantic parsing results with the knowledge graph, it can automatically retrieve domain knowledge relevant to the current request, such as historical operating data of the target equipment, typical fault modes, and standard analysis methods. This not only completes the implicit information that may exist in the user request but also enhances knowledge for unclear or incomplete needs, enabling the agent to have a deeper understanding of the business environment and context. It can provide personalized and professional task decision-making basis for specific business scenarios, significantly improving the accuracy and practicality of task decomposition.

[0067] This technology applies knowledge augmentation to understand and decompose natural language requests, automatically generating task instruction chains containing multiple business sub-tasks. These task instruction chains connect data acquisition, analysis, and visualization in a structured manner, providing a foundation for subsequent automated scheduling and execution. This technical solution not only achieves intelligent transformation of natural language requests into executable task chains but also overcomes the limitations of traditional rule-based or template-based methods in processing complex industry knowledge through knowledge graph completion and enhancement. This significantly improves the automation, accuracy, and responsiveness of intelligent operation and maintenance and data analysis in power systems.

[0068] In one embodiment, such as Figure 3 As shown, it also includes the following steps S31-S33:

[0069] In step S31, the general corpus is pre-trained to obtain an initial language model;

[0070] In step S32, the initial language model is trained based on power system domain data to obtain a small language model optimized for the power system domain.

[0071] In step S33, the trained small language model is deployed to the business decomposition agent for intent recognition and task decomposition of natural language requests.

[0072] In one embodiment, the complete construction and application path of a small language model used for natural language request processing in the business context of power systems can be divided into three main stages: model pre-training, domain adaptation, and agent deployment. Pre-training on a large-scale general corpus yields an initial language model with broad language comprehension capabilities. This initial language model performs well in basic grammar, general vocabulary, and general expression comprehension, laying a solid foundation for subsequent domain-specific tasks. General pre-training provides an efficient parameter initialization and knowledge transfer basis for subsequent transfer learning and targeted optimization, enabling the model not only to handle everyday language requests but also to adapt to complex and ever-changing contextual structures.

[0073] Building upon the general pre-trained model, a large amount of proprietary data from the power system domain is further incorporated for refined training. This stage includes not only core power industry terms and expressions such as equipment names, operating statuses, business rules, and fault cases, but also scenario-based data such as real-world operation and maintenance logs and user business query history. Through this domain-specific fine-tuning, the model can fully learn the knowledge structure and common expression patterns of the power industry, significantly improving its understanding and adaptability to professional content such as power system business requirements, equipment relationships, and analysis processes. The resulting small language model, while maintaining low computational resource consumption and high inference efficiency, can accurately identify core intents and entities in power system scenarios, providing reliable support for intelligent task decomposition.

[0074] The trained, power-domain-specific small-scale language model is deployed in a business decomposition agent, becoming a fundamental component for the system's natural language intent recognition and automatic task decomposition. When users submit complex business requests such as analysis, queries, or visualizations via natural language, the agent can quickly and accurately perform semantic parsing, mapping the request into structured business intents and task decomposition results. This not only significantly improves the response speed and accuracy to user needs in intelligent operation and maintenance of power systems but also lays a solid technical foundation for subsequent knowledge graph retrieval, task instruction chain generation, and full-process automation.

[0075] In one embodiment, such as Figure 4 As shown, it also includes the following steps S41-S43:

[0076] In step S41, each node in the task instruction chain corresponds to a power system data analysis subtask or a visualization subtask;

[0077] In step S42, in the directed acyclic graph structure, the directed connections between nodes represent the dependencies between the subtasks;

[0078] In step S43, the directed acyclic graph structure ensures that there are no circular dependencies during the execution of the task instruction chain.

[0079] In one embodiment, the task instruction chain is organized using a directed acyclic graph structure to achieve fine-grained modeling and efficient management of complex business processes in power systems. Each node in the task instruction chain corresponds to an independent data analysis subtask or visualization subtask, enabling the decomposition of the user's overall business request into several independently executable and clearly describable functional units. For example, in the scenario of power system operation status analysis, nodes may include data acquisition, anomaly detection, feature extraction, result aggregation, chart generation, etc., with each node carrying different data processing or business logic tasks. Granular design helps the system flexibly configure and reuse various subtask modules according to actual business needs, thereby improving the flexibility of task planning and the overall scalability of the system.

[0080] In this directed acyclic graph (DAG) structure, the directed connections between nodes clearly indicate the dependencies between subtasks. That is, the execution of a node requires the completion of its predecessor node, ensuring not only the correctness of data flow and business processes but also providing fundamental support for automated task scheduling. The instruction registry can automatically identify subtasks that can be executed in parallel or sequentially based on the dependency structure in the DAG, thereby dynamically adjusting scheduling strategies to maximize resource utilization and minimize business latency. Especially against the backdrop of increasing demands for parallel processing of data analysis and visualization tasks, the task chain organization method of the DAG structure significantly improves the system's processing capacity and response efficiency when handling high concurrency and multi-dimensional requests. The DAG structure naturally eliminates circular dependencies, ensuring that the entire task instruction chain execution process can always proceed smoothly without deadlocks or infinite loops caused by task dependencies. This is particularly important for the multi-stage, multi-branch processing flows commonly found in power system operations. For example, in complex data cleaning and multi-layered analysis processes, all nodes without prerequisite dependencies can be automatically identified and executed first. Downstream dependent tasks are then triggered progressively upon node completion, ensuring the efficiency, controllability, and security of the business process. By organizing the task instruction chain using a directed acyclic graph, structured management of complex tasks is achieved.

[0081] In one embodiment, such as Figure 5 As shown, it also includes the following steps S51-S53:

[0082] In step S51, after receiving the task instruction chain generated by the business decomposition intelligent agent, each sub-task instruction is registered as a waiting-to-execute state;

[0083] In step S52, when all the prerequisite conditions of a certain subtask are met, the subtask is switched to the execution state and executed by the visual presentation agent;

[0084] In step S53, after the subtask is completed, it is switched to the completed state. If an error occurs during the execution, it is switched to the error state to trigger the corresponding exception handling.

[0085] In one embodiment, the fine-grained scheduling and full-process lifecycle control mechanism in task instruction chain execution management dynamically manages the state of each sub-task node through an instruction registry center to ensure the orderly progress and controllable anomalies of the overall task. When the task instruction chain generated by the business decomposition agent is submitted to the instruction registry center, each sub-task node in the chain is automatically registered as "waiting to execute." This not only achieves global visual management of the task execution process but also provides a data foundation for subsequent state switching, task scheduling, and dependency analysis. All sub-tasks are statically distributed and can only be activated when their preceding dependencies are met, avoiding problems such as disordered task triggering and resource waste in traditional scheduling. During the actual scheduling phase, the instruction registry center continuously monitors the dependency conditions of each sub-task. When all preceding dependent tasks of a sub-task are completed, the state of that task node is automatically switched from "waiting to execute" to "executing," and it is assigned to the visualization agent or other suitable execution modules for actual operation. For example, in a power system big data analysis scenario, a "visualization" sub-task can only be started after all its preceding data analysis, anomaly detection, and other task nodes are completed. This "dependency-driven, step-by-step activation" scheduling strategy not only improves the accuracy and security of process execution, but also significantly enhances the efficiency of multi-task parallel processing.

[0086] During the actual execution phase of a task node, its status is updated based on the results of the subtask's execution. When a subtask completes successfully, its status is switched to "Completed." If an error or exception occurs during execution, the node's status is automatically changed to "Error," and the corresponding exception handling process is immediately triggered, effectively ensuring the robustness and recoverability of each link in the task chain. For example, if an analysis task fails due to missing data, the system can retry, switch to a backup process, or notify the user to intervene according to the preset exception handling strategy, preventing the exception from spreading to subsequent tasks or causing overall process blockage. This management mechanism, through dynamic switching and fine-grained monitoring of subtask status, achieves closed-loop control of the entire process from task generation to execution, completion, and even exception handling.

[0087] In one embodiment, such as Figure 6 As shown, it also includes the following steps S61-S63:

[0088] In step S61, the visualization-related instructions in the task instruction chain are parsed into interface components of the interactive dashboard to display the power system data analysis results;

[0089] In step S62, a natural language summary report describing the analysis results is generated based on the large language model;

[0090] In step S63, the interactive dashboard enables users to interactively query and dynamically visualize power system data.

[0091] In one embodiment, key technical processes for intelligent visualization and intelligent report generation of power system data involve the automatic parsing of visualization instructions in the task chain, intelligent interpretation of visualization results, and multi-dimensional enhancement of user interaction capabilities. The system can automatically identify data visualization-related instructions in the task instruction chain and parse them into interface components that constitute an interactive dashboard. Specifically, different types of data analysis results will be automatically mapped to appropriate chart components, such as trend line charts, distribution histograms, heat maps, and equipment status dashboards, achieving multi-dimensional visualization of complex analysis conclusions. The automatic component parsing and dashboard generation method eliminates the limitations of traditional manual chart template configuration, enabling the system to flexibly and dynamically organize and display results for different analysis tasks and data types. Building upon the generation of visualization components, the system further leverages the natural language processing capabilities of a large-scale language model to intelligently interpret the analysis results and generate summary reports. The large-scale language model can comprehensively analyze multi-dimensional data and key indicators in the dashboard, automatically generating concise, accurate, and business-insightful natural language summaries. For example, in load anomaly analysis scenarios, the model can not only provide quantitative conclusions such as "the peak transformer load this month is higher than the historical average for the same period," but also automatically generate business explanations and suggestions based on anomalies and related factors, thereby helping users quickly understand the inherent trends and potential risks of the data. The integrated "data-visualization-text" output significantly lowers the professional threshold for users and provides high-value auxiliary evidence for intelligent decision-making. The generated interactive dashboard not only supports static data browsing but also possesses powerful user interaction and dynamic display capabilities. Users can customize filter conditions, adjust data ranges, link different chart perspectives, and even perform drill-down analysis of anomalies and cross-referencing of related data within the dashboard interface. This highly interactive and personalized data exploration experience enables the system to meet the business needs of diverse scenarios such as power operation and maintenance, management, and decision-making, greatly improving data utilization efficiency and information acquisition depth. Through automated visualization analysis, intelligent summary generation, and rich interactive experiences, a new paradigm for intelligent presentation of power system data analysis has been constructed, fully demonstrating technological progress and creativity.

[0092] In one embodiment, Figure 7 This is a block diagram of a dual-agent processing system based on business decomposition and visualization, according to an exemplary embodiment. For example... Figure 7 As shown, the processing system based on business decomposition and visualization dual-agent includes a decomposition module 71, a visualization module 72, and a management module 73.

[0093] The disassembly module 71 is used to receive natural language requests from the power system by the business disassembly intelligent agent, perform semantic parsing of the natural language requests through a small language model, perform semantic retrieval based on the power system knowledge graph, realize knowledge-enhanced intent recognition and task disassembly, and generate corresponding task instruction chains in a directed acyclic graph structure.

[0094] The visualization module 72 is used to send the task instruction chain to the instruction registration center, which then schedules the visualization presentation agent to generate an interactive dashboard and natural language summary through a large language model to complete the visualization presentation of power system data.

[0095] The management module 73 is used to perform lifecycle management of the entire processing flow through the instruction registration center using a finite state machine, so as to constrain the execution flow of the task instruction chain.

[0096] The decomposition module 71, visualization module 72, and management module 73 included in the block diagram of the processing system based on business decomposition and visualization are controlled to execute the processing method based on business decomposition and visualization described in any of the above embodiments.

[0097] like Figure 8 As shown, the present invention provides an electronic device 800, which includes: a communication interface, a processor 801, and a memory 802;

[0098] The memory 802 stores program instructions. When executed by the processor 801, which is connected to the memory 802 via the communication interface, the program instructions are used by a business decomposition agent to receive natural language requests from the power system. The agent performs semantic parsing of the natural language requests using a small language model, performs semantic retrieval based on the power system knowledge graph, and achieves knowledge-enhanced intent recognition and task decomposition. A corresponding task instruction chain is generated using a directed acyclic graph structure. The task instruction chain is then sent to an instruction registration center, which schedules a visualization agent to generate an interactive dashboard and natural language summary using a large language model to complete the visualization of power system data. The instruction registration center uses a finite state machine to manage the entire processing flow throughout its lifecycle, thus constraining the execution flow of the task instruction chain.

[0099] This invention provides a computer-readable storage medium storing computer program instructions. When these instructions are executed by a processor, a business decomposition agent receives natural language requests from a power system, performs semantic parsing of the requests using a small language model, performs semantic retrieval based on a power system knowledge graph, and achieves knowledge-enhanced intent recognition and task decomposition, generating a corresponding task instruction chain using a directed acyclic graph structure. The task instruction chain is then sent to an instruction registry center, which schedules a visualization agent to generate an interactive dashboard and natural language summary using a large language model, thus completing the visualization of power system data. The instruction registry center employs a finite state machine to manage the entire processing flow throughout its lifecycle, constraining the execution flow of the task instruction chain.

[0100] It should be understood that the specific features, operations, and details described above regarding the method of the present invention can also be similarly applied to the assembly system of the present invention, or vice versa. Furthermore, each step of the method of the present invention described above can be performed by a corresponding component or unit of the system of the present invention.

[0101] It should be understood that the various modules / units of the system of the present invention can be implemented wholly or partially through software, hardware, firmware, or a combination thereof. Each module / unit can be embedded in the processor of a computer device in hardware or firmware form or independent of the processor, or it can be stored in the memory of a computer device in software form for the processor to call to execute the operation of each module / unit. Each module / unit can be implemented as an independent component or module, or two or more modules / units can be implemented as a single component or module.

[0102] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores computer instructions executable by the processor, which, when executed by the processor, instruct the processor to perform steps of the methods of embodiments of the present invention. The computer device can be broadly categorized as a server, terminal, or any other electronic device with the necessary computing and / or processing capabilities. In one embodiment, the computer device may include a processor, memory, network interface, communication interface, etc., connected via a system bus. The processor of the computer device can be used to provide the necessary computing, processing, and / or control capabilities. The memory of the computer device may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system, computer programs, etc. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface and communication interface of the computer device can be used to connect and communicate with external devices via a network. When the computer program is executed by the processor, it performs the steps of the methods of the present invention.

[0103] This invention can be implemented as a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, causes the steps of the methods of embodiments of the invention to be performed. In one embodiment, the computer program is distributed across multiple network-coupled computer devices or processors, such that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, may be executed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations may be executed by one or more computer devices or processors, and one or more other method steps / operations may be executed by one or more other computer devices or processors. One or more computer devices or processors may execute a single method step / operation, or execute two or more method steps / operations.

[0104] It will be understood by those skilled in the art that the method steps of the present invention can be performed by a computer program instructing related hardware, such as a computer device or processor. The computer program may be stored in a non-transitory computer-readable storage medium, and its execution causes the steps of the present invention to be performed. Depending on the context, any references herein to memory, storage, databases, or other media may include non-volatile and / or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (GGPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0105] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.

[0106] 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 them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A processing method based on a dual-agent approach of business decomposition and visualization, characterized in that, include: The business decomposition intelligent agent receives natural language requests from the power system, performs semantic parsing of the natural language requests through a small language model, performs semantic retrieval based on the power system knowledge graph, realizes knowledge-enhanced intent recognition and task decomposition, and generates corresponding task instruction chains in a directed acyclic graph structure. The task instruction chain is sent to the instruction registration center, which then schedules the visualization presentation agent to generate an interactive dashboard and natural language summary through a large language model, thereby completing the visualization presentation of power system data. The instruction registration center uses a finite state machine to manage the entire processing flow throughout its lifecycle, thereby constraining the execution flow of the task instruction chain. After receiving the task instruction chain generated by the business decomposition intelligent agent, each sub-task instruction is registered as waiting to be executed; When all the prerequisite conditions of a subtask are met, the subtask is switched to the execution state and executed by the visual agent. After a subtask is completed, it is switched to the completed state. If an error occurs during execution, it is switched to the error state to trigger the corresponding exception handling.

2. The processing method based on dual intelligent agents of business decomposition and visualization as described in claim 1, characterized in that, Also includes: The natural language request is semantically parsed using a small language model. Based on the power system knowledge graph, perform enhanced semantic retrieval on the semantic parsing results to obtain power system domain knowledge related to the request; The acquired power system domain knowledge is applied to the understanding of the natural language request, and a corresponding task instruction chain is generated.

3. The processing method based on business decomposition and visualization of dual intelligent agents as described in claim 1, characterized in that, Also includes: Pre-train on a general corpus to obtain an initial language model; The initial language model is trained based on data from the power system domain to obtain a smaller language model optimized for the power system domain. The trained small language model is deployed into the business decomposition agent for intent recognition and task decomposition of natural language requests.

4. The processing method based on business decomposition and visualization of dual intelligent agents as described in claim 1, characterized in that, Also includes: Each node in the task instruction chain corresponds to a power system data analysis subtask or visualization subtask. In the directed acyclic graph structure, the directed connections between nodes represent the dependencies between the subtasks; The directed acyclic graph structure ensures that there are no circular dependencies during the execution of the task instruction chain.

5. The processing method based on dual intelligent agents of business decomposition and visualization presentation as described in claim 1, characterized in that, Also includes: The visualization-related instructions in the task instruction chain are parsed into interface components of an interactive dashboard to display the power system data analysis results; A natural language summary report describing the analysis results is generated based on the large-scale language model. The interactive dashboard enables users to interactively query and dynamically visualize power system data.

6. A processing system based on a dual-agent approach of business decomposition and visualization, characterized in that, include: The decomposition module is used to receive natural language requests from the power system by the business decomposition intelligent agent, perform semantic parsing of the natural language requests through a small language model, perform semantic retrieval based on the power system knowledge graph, realize knowledge-enhanced intent recognition and task decomposition, and generate corresponding task instruction chains in a directed acyclic graph structure. The visualization module is used to send the task instruction chain to the instruction registration center, which then schedules the visualization presentation agent to generate interactive dashboards and natural language summaries through a large language model, thereby completing the visualization presentation of power system data. The management module is used to manage the entire processing flow through the instruction registration center using a finite state machine to constrain the execution flow of the task instruction chain; it is also used to register each sub-task instruction as waiting to be executed after receiving the task instruction chain generated by the business decomposition agent; when all the preceding dependency conditions of a sub-task are met, the sub-task is switched to the execution state and executed by the visualization agent. After a subtask is completed, it is switched to the completed state. If an error occurs during execution, it is switched to the error state to trigger the corresponding exception handling.

7. The processing system based on business decomposition and visualization of dual intelligent agents as described in claim 6, characterized in that: The disassembly module, the visualization module, and the management module are controlled to execute the processing method based on a dual-agent system of business disassembly and visualization presentation as described in any one of claims 2 to 5.

8. An electronic device, characterized in that, include: Communication interface, processor, memory; The memory is used to store program instructions, which, when executed by the processor that is connected to the memory via the communication interface, enable the electronic device to implement the dual-agent processing method based on business decomposition and visualization as described in any one of claims 1 to 5.

9. A computer-readable storage medium having program instructions stored thereon, characterized in that, When the program instructions are executed by the computer, the computer implements the processing method based on business decomposition and visualization presentation of dual intelligent agents as described in any one of claims 1 to 5.

Images