Digital robot service intelligent orchestration system and method

Abstract

This invention discloses an intelligent orchestration system and method for digital robot services, relating to the fields of artificial intelligence and data processing technology. It solves the technical problem of excessively cumbersome office processes caused by existing intelligent orchestration systems for digital robot services. The system includes an agent orchestration module, an intelligent query module, a data verification module, and a large-model-driven task generation module. These modules interact via a data bus. The agent orchestration module receives group files or natural language instructions, parses them to generate a task execution tree, schedules and monitors agents to execute atomic tasks, and outputs business query requirement data. The intelligent query module extracts the original retrieval dataset based on this data, the data verification module verifies it and outputs a verified dataset, and the large-model-driven task generation module generates structured tasks based on this, schedules multiple agents to execute collaboratively, and outputs the final orchestration result.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and data processing technology, and in particular to a digital robot service intelligent orchestration system and method. Background Technology

[0002] Currently, digital office is entering a stage of comprehensive and in-depth development. Digital transformation is gradually permeating the entire process of enterprise operation in all industries. Traditional manual office models can no longer meet the development needs of modern enterprises, and intelligent and automated office has become the mainstream direction for enterprise transformation and upgrading.

[0003] In the context of the continuous implementation of digital office solutions across the industry, specific office needs in daily enterprise operations, such as group file processing, cross-departmental task collaboration, and multi-source data querying, are experiencing explosive growth. As the core carrier of enterprise automated office systems, the service capabilities of digital robots directly determine the overall operational efficiency and quality of office work for enterprises.

[0004] The existing intelligent orchestration system for digital robot services relies excessively on manual template configuration for task generation. Users need to define task triggering conditions, execution steps, and output formats in advance. For vague natural language requirements such as "generate a competitive analysis task based on this week's market research report," the system cannot automatically parse and generate a structured task flow, resulting in overly cumbersome office processes. Summary of the Invention

[0005] This invention provides a digital robot service intelligent orchestration system and method, which solves the technical problem that existing digital robot service intelligent orchestration systems cause overly cumbersome office processes.

[0006] The first aspect of the present invention provides a digital robot service intelligent orchestration system, the system comprising an intelligent agent orchestration module, an intelligent query module, a data verification module, and a large model-driven task generation module that interact via a data bus;

[0007] The intelligent agent orchestration module is used to receive group files uploaded by users or input natural language instructions, parse the group files or natural language instructions to obtain a task execution tree, schedule intelligent agents to execute atomic tasks in the task execution tree and monitor the execution status in real time, and output business query requirement data.

[0008] The intelligent query module is used to receive the business query requirement data output by the intelligent agent orchestration module, and extract the original retrieval dataset from multiple databases based on the business query requirement data.

[0009] The data verification module is used to receive the original retrieval dataset output by the intelligent query module, verify it, and output a qualified dataset.

[0010] The large model-driven task generation module is used to receive the qualified dataset output by the data verification module, generate structured tasks based on the qualified dataset, schedule multiple agents to execute collaboratively, and generate the final orchestration result.

[0011] Optionally, the agent orchestration module includes a task parsing unit, an agent scheduling unit, an execution monitoring unit, and a collaborative interaction unit connected in sequence;

[0012] The task parsing unit is used to receive the group file or the natural language instruction, parse the group file or the natural language instruction, generate a task execution tree, and transmit the task execution tree to the agent scheduling unit;

[0013] The agent scheduling unit is used to schedule the agent to execute atomic tasks according to the task execution tree, generate scheduling execution information and transmit it to the execution monitoring unit;

[0014] The execution monitoring unit is used to track the operation process of the corresponding intelligent agent according to the task content and execution requirements in the scheduling execution information, collect the execution status information of the intelligent agent, and then transmit the execution status information to the collaborative interaction unit.

[0015] The collaborative interaction unit is used to generate business query requirement data based on the task progress and execution result in the execution status information, and then transmit the business query requirement data to the intelligent query module.

[0016] Optionally, the intelligent question module includes a question understanding unit and a data retrieval unit connected in sequence;

[0017] The problem understanding unit is used to interpret the business query requirement data, generate standardized query information, and transmit it to the data retrieval unit.

[0018] The data retrieval unit is used to extract the original retrieval dataset from the multiple databases according to the query range defined by the standardized query information and transmit it to the data verification module.

[0019] Optionally, the data verification module includes a data acquisition unit, a rule formulation unit, a verification execution unit, and an error handling unit connected in sequence.

[0020] The data acquisition unit is used to aggregate the original retrieval dataset to obtain the aggregated original retrieval dataset, and then transmit the aggregated original retrieval dataset to the verification execution unit.

[0021] The rule-making unit is used to construct standardized verification rules and transmit them to the verification execution unit;

[0022] The verification execution unit is used to verify the aggregated original retrieval dataset using the standardized verification rules, generate initial screening verification data, and transmit it to the error handling unit.

[0023] The error handling unit is used to correct abnormal data and remove invalid data from the initial screening verification data, generate a qualified verification dataset, and transmit the qualified verification dataset to the large model-driven task generation module.

[0024] Optionally, the large model-driven task generation module includes a task generation principle unit, a task management and execution unit, and an intelligent agent collaboration unit connected in sequence.

[0025] The task generation principle unit is used to parse the business requirements of the qualified dataset, generate structured task assignments, and transmit the structured task assignments to the task management and execution unit.

[0026] The task management and execution unit is used to arrange and plan the structured task assignments, generate task execution plans, and transmit them to the intelligent agent collaboration unit.

[0027] The intelligent agent coordination unit is used to connect with the adapted intelligent agents according to the task execution plan, issue tasks to the intelligent agents and organize multiple intelligent agents to carry out the work collaboratively, generate multiple task execution information, summarize the multiple task execution information, and output the final arrangement result.

[0028] The second aspect of this invention provides a method for intelligent orchestration of digital robot services, comprising:

[0029] Receive group files uploaded by users or input natural language instructions, parse the group files or natural language instructions, and obtain a task execution tree;

[0030] Execute atomic tasks in the task execution tree and monitor their execution status in real time, then output business query requirement data.

[0031] Based on the aforementioned business query requirements data, the original retrieval dataset is extracted from multiple databases;

[0032] The original retrieved dataset is validated, and a valid dataset is output.

[0033] Based on the verified dataset, structured to-do tasks are generated, and multiple agents are scheduled to execute them collaboratively to generate the final orchestration result.

[0034] Optionally, the step of executing atomic tasks in the task execution tree and monitoring their execution status in real time, and outputting business query requirement data, includes:

[0035] The agent is scheduled to execute atomic tasks according to the task execution tree, and scheduling execution information is generated;

[0036] Based on the task content and execution requirements within the scheduling execution information, the operation process of the corresponding intelligent agent is tracked, and the execution status information of the intelligent agent is collected;

[0037] Based on the task progress and execution results in the execution status information, business query requirement data is generated.

[0038] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the digital robot service intelligent orchestration method as described above.

[0039] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the intelligent orchestration method for digital robot services as described above.

[0040] The fifth aspect of the present invention provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein, when the program instructions are executed by a computer, the computer performs the steps of the digital robot service intelligent orchestration method as described above.

[0041] As can be seen from the above technical solutions, the present invention has the following advantages:

[0042] The first aspect of the technical solution of the present invention provides a digital robot service intelligent orchestration system. This system includes an intelligent agent orchestration module, an intelligent query module, a data verification module, and a large model-driven task generation module, all interacting via a data bus. The intelligent agent orchestration module receives user-uploaded group files or input natural language instructions, parses the group files or natural language instructions to obtain a task execution tree, schedules intelligent agents to execute atomic tasks in the task execution tree, monitors the execution status in real time, and outputs business query requirement data. The intelligent query module receives the business query requirement data output by the intelligent agent orchestration module and extracts raw retrieval datasets from multiple databases based on the business query requirement data. The data verification module receives the raw retrieval datasets output by the intelligent query module, verifies them, and outputs a verified dataset. The large model-driven task generation module receives the verified datasets output by the data verification module, generates structured tasks based on the verified datasets, schedules multiple intelligent agents to execute collaboratively, and generates the final orchestration result. Based on the above scheme, this invention uses an intelligent agent orchestration module to parse user-uploaded group files or input natural language commands and construct a task execution tree. It then schedules intelligent agents to execute atomic tasks and output business query requirement data. The data is then processed sequentially by an intelligent query module to extract the original retrieval dataset, a data verification module to perform data verification, and a large model-driven task generation module to generate structured tasks and schedule multiple intelligent agents to execute them collaboratively. Finally, the orchestration result is output. This eliminates the need for manually pre-setting cumbersome rules, and the system allows non-technical personnel to initiate tasks directly through natural language, thereby improving the convenience of office automation.

[0043] The second aspect of the above-mentioned technical solution of the present invention provides a digital robot service intelligent orchestration method, which receives group files uploaded by users or inputs natural language instructions, parses the group files or natural language instructions to obtain a task execution tree; executes atomic tasks in the task execution tree and monitors the execution status in real time, and outputs business query requirement data; extracts original retrieval datasets from multiple databases based on the business query requirement data; verifies the original retrieval datasets and outputs a verified dataset; generates structured to-do tasks based on the verified datasets, schedules multiple intelligent agents to perform collaborative execution, and generates the final orchestration result. Compared with the existing method where users need to define task triggering conditions, execution steps, and output formats in advance, and where the system cannot autonomously parse and generate structured task flows for vague natural language requirements such as "generate a competitor analysis task based on this week's market research report," the present invention eliminates the need for manual pre-setting of cumbersome rules. The system supports non-technical personnel to directly initiate tasks through natural language, thereby improving the convenience of office automation. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the structure of a digital robot service intelligent orchestration system provided in Embodiment 1 of the present invention;

[0046] Figure 2 This is a schematic diagram of the intelligent agent orchestration module provided in Embodiment 1 of the present invention;

[0047] Figure 3 This is a schematic diagram of the intelligent questioning module provided in Embodiment 1 of the present invention;

[0048] Figure 4 This is a schematic diagram of the data verification module provided in Embodiment 1 of the present invention;

[0049] Figure 5 This is a schematic diagram of the structure of the large model-driven task generation module provided in Embodiment 1 of the present invention;

[0050] Figure 6 This is a flowchart illustrating the steps of an intelligent orchestration method for digital robot services provided in Embodiment 2 of the present invention.

[0051] Figure 7 This is a flowchart illustrating a digital robot service intelligent orchestration method provided in Embodiment 2 of the present invention;

[0052] The meanings of the labels in the attached figures are as follows:

[0053] 01. Agent Orchestration Module; 02. Intelligent Questioning Module; 03. Data Validation Module; 04. Large Model-Driven Task Generation Module; 011. Task Parsing Unit; 012. Agent Scheduling Unit; 013. Execution Monitoring Unit; 014. Collaborative Interaction Unit; 021. Problem Understanding Unit; 022. Data Retrieval Unit; 023. Answer Generation Unit; 024. Learning Optimization Unit; 031. Data Acquisition Unit; 032. Rule Formulation Unit; 033. Validation and Execution Unit; 034. Error Handling Unit; 041. Task Generation Principle Unit; 042. Task Management and Execution Unit; 043. Agent Collaboration Unit. Detailed Implementation

[0054] This invention provides a digital robot service intelligent orchestration system and method, which solves the technical problem that existing digital robot service intelligent orchestration systems cause overly cumbersome office processes.

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that in the optional embodiments of the present invention, the object information and other related data involved require the permission or consent of the object when the embodiments of the present invention are applied to specific products or technologies, and the collection, use, and processing of related data must comply with relevant laws, regulations, and standards. That is to say, if the embodiments of the present invention involve data related to the object, it needs to be obtained with the authorization and consent of the object, the authorization and consent of relevant departments, and in compliance with relevant laws, regulations, and standards. If personal information is involved in the embodiments, the acquisition of all personal information requires the consent of the individual. If sensitive information is involved, the separate consent of the information subject is required, and the embodiments also need to be implemented with the authorization and consent of the object.

[0056] Please see Figure 1 , Figure 1 This is a schematic diagram of the structure of a digital robot service intelligent orchestration system provided in Embodiment 1 of the present invention.

[0057] The present invention provides a digital robot service intelligent orchestration system, which includes an intelligent agent orchestration module 01, an intelligent data query module 02, a data verification module 03, and a large model-driven task generation module 04 that interact through a data bus.

[0058] The intelligent agent orchestration module 01 is used to receive group files uploaded by users or input natural language instructions, parse the group files or natural language instructions to obtain the task execution tree, schedule intelligent agents to execute atomic tasks in the task execution tree and monitor the execution status in real time, and output business query requirement data.

[0059] The intelligent query module 02 is used to receive business query requirement data output by the intelligent agent orchestration module 01, and extract the original retrieval dataset from multiple databases based on the business query requirement data;

[0060] The data verification module 03 is used to receive the original search dataset output by the intelligent query module 02, verify it, and output the verified dataset.

[0061] The large model-driven task generation module 04 is used to receive the qualified dataset output by the data verification module 03, generate structured tasks based on the qualified dataset, schedule multiple agents to execute collaboratively, and generate the final orchestration result.

[0062] It should be noted that this invention includes an agent orchestration module 01, an intelligent query module 02, a data verification module 03, and a large model-driven task generation module 04, with each module interacting via a data bus. After system startup, the agent orchestration module 01 first receives group files uploaded by the user or input natural language commands. After parsing the received content, it constructs a task execution tree, simultaneously scheduling atomic tasks within the agent execution tree and monitoring the execution status throughout. It generates and transmits business query requirement data downstream. Upon receiving this data, the intelligent query module 02 extracts the original retrieval dataset from multiple databases and transmits it to the data verification module 03. The data verification module 03 performs verification operations on the received original retrieval dataset and outputs a verified dataset. This dataset flows into the large model-driven task generation module 04, which generates structured tasks based on the dataset, schedules multiple agents to collaboratively complete task execution, and generates the final orchestration result.

[0063] The final orchestration result is the overall output of this round of intelligent orchestration process for digital robot services. It is formed by summarizing the various task execution information generated during the task execution process, organized by the intelligent agent collaboration unit according to the task execution plan, after multiple agents have completed the assigned tasks. This content includes task allocation records, multi-agent collaborative operation data, task running status, and anomaly handling records based on the 3σ principle. It also incorporates supporting data retrieval, data verification, AI model and algorithm outputs, and actual business results, directly supporting business implementation and providing a reference for system strategy optimization. For example, in the substation defect handling scenario, the result is a defect classification table, personnel allocation list, and rectification plan formed by summarizing the defect handling-related task execution information fed back by each agent; in the regional electricity consumption statistics scenario, it is the output of electricity consumption data tables, month-on-month analysis charts, and other business results after integrating the execution information of various electricity consumption statistics tasks.

[0064] As a further improvement, such as Figure 2 As shown, the intelligent agent orchestration module 01 includes a task parsing unit 011, an intelligent agent scheduling unit 012, an execution monitoring unit 013, and a collaborative interaction unit 014 connected in sequence.

[0065] The task parsing unit 011 is used to receive group files or natural language instructions, parse group files or natural language instructions, generate task execution trees, and transmit task execution trees to the agent scheduling unit 012.

[0066] The agent scheduling unit 012 is used to schedule agents to execute atomic tasks according to the task execution tree, generate scheduling execution information and transmit it to the execution monitoring unit 013;

[0067] The execution monitoring unit 013 is used to track the operation process of the corresponding intelligent agent according to the task content and execution requirements in the scheduling execution information, collect the execution status information of the intelligent agent, and then transmit the execution status information to the collaborative interaction unit 014.

[0068] The collaborative interaction unit 014 is used to generate business query requirement data based on the task progress and execution results in the execution status information, and then transmit the business query requirement data to the intelligent query module.

[0069] The scheduling execution information is a set of instruction-type data generated by the agent scheduling unit based on the task execution tree and transmitted to the execution monitoring unit to support the full-process monitoring of agent tasks. Specifically, it includes the unique identifier of each atomic task, the specific content and execution requirements of the task, the corresponding agent number and capability matching result, the task priority coefficient, dependent task node information, the estimated start and end time of execution, the scheduling allocation timestamp, and the cosine similarity matching value between the task vector and the agent capability vector, providing the execution monitoring unit with clear monitoring objects and judgment criteria.

[0070] Execution status information is a set of status data collected, generated, and transmitted to the collaborative interaction unit by the execution monitoring unit to track the operation of the intelligent agents. It reflects the status of the entire task execution process and includes the execution agent number of each atomic task, the actual start and end time of execution, the current execution progress and completion status of key nodes, execution status identifier (pending execution / in execution / completed / abnormal), key output data during execution, details of abnormal events identified based on the 3σ principle (including trigger time, degree of deviation from the mean, alarm level), abnormal handling records (such as the time and reason for task retry and migration), interaction sequence records of multi-agent collaborative operations, and comparison data between the actual task time and the statistical mean and standard deviation, thus fully restoring the actual situation of the task execution process.

[0071] Business query requirement data is a set of instruction-type data generated by the collaborative interaction unit based on execution status information and transmitted to the intelligent query module. It clarifies the subsequent data retrieval goals and rules. Specifically, it includes the business theme corresponding to this orchestration task, the data dimensions to be retrieved (such as task progress, defect details, electricity consumption indicators, etc.), the data source range (such as specified database tables, business systems), the time interval for data retrieval, the filtering conditions and priority rules for data extraction, the data output format requirements, and the association identifier between this orchestration task and user group files / natural language instructions. This provides a clear execution basis for the multi-database retrieval of the intelligent query module.

[0072] It should be noted that the task parsing unit 011 constructs a task execution tree through the task decomposition subunit. The task parsing unit 011 includes a natural language understanding subunit, a task decomposition subunit, a priority determination subunit, and a task template matching subunit. The natural language understanding subunit uses a fine-tuning model based on BERT (Bidirectional Encoder Representations from Transformers) to convert the user's natural language instructions or group files into task vectors that can be understood by machines. The task decomposition subunit decomposes complex tasks into atomic tasks through the graph theory topological sorting algorithm and constructs a task execution tree. The nodes of the task execution tree satisfy the formula T(n) = {I(n), O(n), D(n), P(n)}, where T(n) is the nth atomic task node in the task execution tree, I(n) is the set of input parameters, O(n) is the set of output parameters, D(n) is the set of dependent task nodes, and P(n) is the priority coefficient 0 < P(n) ≤ 1. Among them, the natural language understanding subunit of the task parsing unit 011 uses a fine-tuning model based on BERT, and its training data includes 500,000 natural language instruction-task vector mapping samples in the enterprise office scenario.

[0073] Furthermore, the agent orchestration module 01 also includes an agent scheduling unit 012, an execution monitoring unit 013, and a collaborative interaction unit 014. The ability matching subunit of the agent scheduling unit 012 calculates the similarity between the task vector and the agent ability vector through the cosine similarity algorithm. The agent scheduling unit 012 includes an agent resource pool management subunit, an ability matching subunit, a load balancing subunit, and a dynamic scheduling subunit. The ability matching subunit calculates the similarity between the task vector and the agent ability vector through the cosine similarity algorithm, such as the formula Sim(Vt, Va) = (Vt·Va) / (||Vt||·||Va||), where Vt is the task vector, Va is the agent ability vector, and the larger the Sim value, the higher the matching degree.

[0074] The anomaly detection subunit of the execution monitoring unit 013 uses the 3σ principle to identify task execution anomalies and triggers an alarm when the task execution time deviates from the mean by more than 3 standard deviations.

[0075] The collaborative interaction unit 014 is used to implement group file collaboration and multi-agent communication, and supports content extraction and task association of files in formats such as PDF and Excel.

[0076] Specifically, in the intelligent agent orchestration module 01, the task parsing unit 011 receives the group file uploaded by the user or the input natural language instructions. Through its internal natural language understanding subunit, it employs a BERT-based fine-tuned model. This model has been trained on 500,000 natural language instruction-task vector mapping samples from enterprise office scenarios, transforming the user-input group file or natural language instructions into machine-understandable task vectors. Subsequently, the task decomposition subunit uses a graph theory topological sorting algorithm to decompose complex business tasks into multiple atomic tasks, and then follows the formula... T(n) = {I(n), O(n), D(n), P(n)} constructs a task execution tree. Simultaneously, combining the processing of the priority determination subunit and the task template matching subunit, a complete task execution tree is generated and transmitted to the agent scheduling unit 012. Upon receiving the task execution tree, the agent scheduling unit 012 uses its internal capability matching subunit to calculate the similarity between the task vector and the agent's capability vector using a cosine similarity algorithm. The calculation method is: Sim(Vt, Va) = (Vt·Va) / (||Vt||·||). Va||), in conjunction with the intelligent agent resource pool management subunit, load balancing subunit, and dynamic scheduling subunit, completes the matching, scheduling, and load distribution of intelligent agents. It schedules the corresponding intelligent agents to execute atomic tasks in the task execution tree and generates scheduling execution information containing task content, execution subject, and execution requirements, which is then transmitted to the execution monitoring unit 013. After receiving the scheduling execution information, the execution monitoring unit 013 tracks the operation process of the corresponding intelligent agent based on the task content and execution requirements. Its internal anomaly detection subunit uses the 3σ principle to identify task execution anomalies. When the task execution time deviates from the mean by more than 3 times the standard deviation, an alarm is triggered. At the same time, it collects the intelligent agent's running status, task progress, and other content to form execution status information, which is then transmitted to the collaborative interaction unit 014. After receiving the execution status information, the collaborative interaction unit 014 realizes group file collaboration and multi-agent communication, supports content extraction and task association of PDF, Excel, and other file formats, sorts out the business query direction based on the task progress and execution results in the execution status information, integrates and forms business query requirement data, and transmits it to the intelligent query module.

[0077] As a further improvement, such as Figure 3 As shown, the intelligent question module 02 includes a question understanding unit 021 and a data retrieval unit 022 connected in sequence;

[0078] Problem understanding unit 021 is used to interpret business query requirement data, generate standardized query information and transmit it to data retrieval unit 022;

[0079] The data retrieval unit 022 is used to extract the original retrieval dataset from multiple databases according to the query scope defined by standardized query information and transmit it to the data verification module 03.

[0080] The multi-database system comprises multiple independent database clusters providing multi-source business data support for the intelligent data query module. Divided by business scenarios and data types, it provides a comprehensive data foundation for intelligent orchestration operations, including four core databases: First, the main business database, storing structured core business data such as project ledgers, task lists, equipment ledgers, and metering data; second, the statistical analysis database, storing historical business statistics such as task execution efficiency, anomaly frequency, and electricity consumption statistics; third, the document archive database, storing semi-structured / unstructured data such as group file parsing content and historical collaborative documents; and fourth, the system operation support database, storing system operation logs and configuration data for intelligent agent scheduling, task execution, and data retrieval. Each database is independent and supports parallel retrieval, providing multi-source data support for subsequent stages.

[0081] It should be noted that the intelligent question module 02 includes a question understanding unit 021, a data retrieval unit 022, an answer generation unit 023, and a learning optimization unit 024. The data retrieval unit 022 uses a distributed query subunit to achieve parallel retrieval of multiple databases, and combines the index to build a B-tree index for the subunit to improve query efficiency.

[0082] Among them, Problem Understanding Unit 021: Constructs a business entity relationship network through knowledge graph-related sub-units to achieve in-depth analysis of problem semantics.

[0083] Data retrieval unit 022: It adopts a distributed query subunit to realize parallel retrieval of multiple databases, and combines the index to build a B-tree index for the subunit to improve query efficiency;

[0084] Answer generation unit 023: Integrates multi-source data and generates charts or structured text through the result formatting sub-unit;

[0085] Learning Optimization Unit 024: Update the question number model based on user feedback and optimize the retrieval strategy.

[0086] Specifically, the intelligent query module 02 includes a question understanding unit 021 and a data retrieval unit 022 connected in sequence. After receiving the business query requirement data transmitted by the collaborative interaction unit 014, the question understanding unit 021 constructs a business entity relationship network through its internal knowledge graph association subunit, performs deep semantic analysis on the business query requirement data, clarifies the query target, data dimensions and business association logic, generates standardized query information containing query scope, data fields and filtering conditions, and transmits it to the data retrieval unit 022. After receiving the standardized query information, the data retrieval unit 022 constructs a B-tree index through its internal index construction subunit, and then relies on the distributed query subunit to achieve parallel retrieval of multiple databases. According to the query scope defined by the standardized query information, it quickly locates the target data based on the B-tree index, extracts the corresponding data from multiple databases and summarizes it into the original retrieval dataset, and then transmits the original retrieval dataset to the data verification module 03.

[0087] For example, the problem understanding unit 021 receives the business query request data and constructs a business entity relationship network of "project-collaborative task-executing agent-task progress-job record" through its internal knowledge graph association subunit. It then performs deep semantic parsing of the query request, clarifying that the query target is "collaborative task data for project ID=PROJ20250601", the data dimensions are "task execution progress, completion rate, and corresponding agent job record", and the business association logic is "project ID associated with task ID, task ID associated with agent ID, and agent ID associated with job record". Subsequently, standardized query information is generated, which includes the query scope (project ID: PROJ20250601) and data fields (task ID, task name, execution progress, completion rate, agent number, and job record ID). The filtering criteria (Project ID=PROJ20250601) are transmitted to the data retrieval unit 022. After receiving the standardized query information, the data retrieval unit 022 first constructs a B-tree index for the fields related to Project ID, Task ID, and Agent ID in the business master database and the system operation support database through its internal index construction sub-unit. Then, relying on the distributed query sub-unit, it simultaneously initiates parallel retrieval requests to the business master database and the system operation support database. Based on the B-tree index, it quickly locates the target data, extracts the ledger, execution progress, and completion rate data of all collaborative tasks under the project from the business master database, and extracts the corresponding agent's job record data from the system operation support database. The two types of data are summarized to form a raw retrieval dataset containing all target fields, and finally transmitted to the data verification module 03.

[0088] As a further improvement, such as Figure 4 As shown, the data verification module 03 includes a data acquisition unit 031, a rule formulation unit 032, a verification execution unit 033, and an error handling unit 034 connected in sequence.

[0089] Data acquisition unit 031 is used to collect data from the original retrieval dataset, obtain the collected original retrieval dataset, and transmit the collected original retrieval dataset to the verification execution unit 033.

[0090] Rule-making unit 032 is used to construct standardized verification rules and transmit them to verification execution unit 033;

[0091] The verification execution unit 033 is used to verify the aggregated original retrieval dataset using standardized verification rules, generate initial screening verification data, and transmit it to the error handling unit 034.

[0092] Error handling unit 034 is used to correct abnormal data and remove invalid data from the initial screening and verification data, generate a qualified verification dataset, and transmit the qualified verification dataset to the large model-driven task generation module 04.

[0093] The aggregated raw retrieval dataset is a standardized data set obtained by the data acquisition unit 031 of the data verification module 03 after uniformly accessing, incrementally acquiring, formatting, and standardizing the multi-source raw retrieval data transmitted by the data retrieval unit 022 of the intelligent query module 02. This dataset serves as the direct input data for the verification execution unit 033 to perform batch verification. This dataset integrates various types of target data from the main business database, statistical analysis database, document archive database, and system operation support database. Specifically, it includes structured business field data (such as project ID, task name, execution progress, completion rate, agent number, and job record ID), semi-structured document parsing data (such as collaboration requirements extracted from group files and progress description text), and system operation data (such as data retrieval timestamps and agent job record timestamps). Simultaneously, the timestamp field records the data acquisition and update time, distinguishing between the fully acquired basic data and the incrementally acquired new / changed data. It unifies data field formats, encoding methods, and field definitions, eliminating format differences and field incompatibility issues between different databases, ensuring consistent data structure and complete data sources, and providing a stable and unified processing object for subsequent batch verification of standardized verification rules.

[0094] It should be noted that the data verification module 03 includes a data acquisition unit 031, a rule formulation unit 032, a verification execution unit 033, and an error handling unit 034. The verification execution unit 033 implements parallel verification of data blocks through a parallel verification subunit. The verification pass rate is calculated as follows: R=(NE) / N×100%, where R is the verification pass rate, N is the total number of data entries, and E is the number of data entries that failed the verification.

[0095] Among them, data acquisition unit 031: supports incremental acquisition and multi-source data access, uses timestamp field to realize data change tracking. Data acquisition unit 031 supports incremental acquisition, records data update time through timestamp field. The incremental acquisition SQL statement is SELECT * FROM table WHERE update_time > 'last acquisition timestamp'.

[0096] Rule Formulation Unit 032: Provides interfaces for defining business rules and custom rules, and supports configuration of validation rules such as regular expressions and value ranges;

[0097] Error handling unit 034: Automatically repairs or manually prompts for data that fails the verification, thereby isolating erroneous data.

[0098] Specifically, the data acquisition unit 031 receives the raw retrieval dataset transmitted by the data retrieval unit 022, supports incremental acquisition and multi-source data access, tracks data changes through a timestamp field, and uses the incremental acquisition SQL statement SELECT * FROM table WHERE update_time > The 'last collection timestamp' completes the collection of new / changed data, unifies the multi-source data to obtain the aggregated original retrieval dataset, and transmits the aggregated original retrieval dataset to the verification execution unit 033; the rule formulation unit 032 provides business rule definition and custom rule interface, supports regular expressions, value range and other verification rule configuration, constructs standardized verification rules and transmits them to the verification execution unit 033; the verification execution unit 033 receives the aggregated original retrieval dataset and standardized verification rules, implements parallel verification of data blocks through internal parallel verification subunits, and calculates the verification pass rate using the formula R=(NE) / N×100%. After verifying the aggregated original retrieval dataset, it generates preliminary screening verification data and transmits it to the error handling unit 034; the error handling unit 034 automatically repairs or manually intervenes to prompt the data that fails verification in the preliminary screening verification data, realizes the isolation of erroneous data, completes the correction of abnormal data and the removal of invalid data, generates a qualified verification dataset, and transmits the qualified verification dataset to the large model-driven task generation module 04.

[0099] Among them, the standardized verification rules are a set of structured verification instructions generated by the rule formulation unit 032 for the aggregated original retrieval dataset, providing the verification execution unit 033 with clear and executable data verification judgment basis; each rule includes the verification object (target data field), verification conditions, judgment logic, anomaly marker type and handling priority, covering multiple dimensions of verification requirements such as data format, value range, integrity, and business association; the rules can be configured through business rule definition or custom interface, and can flexibly implement various verification logics such as regular expression matching, numerical range judgment, non-empty verification, and cross-field association verification, ensuring the consistency of data verification and scenario adaptability.

[0100] For example, the standardized validation rules configured for the original retrieval dataset of the collaborative task for the "XX System Upgrade" project include:

[0101] Format validation rules: Configure regular expression validation rules for the "Project ID" field. The regular expression is ^PROJ\d{8}$. Validate whether the field value conforms to the standard format of "PROJ" prefix followed by 8 digits. If it does not conform, mark it as a format error.

[0102] Value range validation rules: For the "task completion rate" field, configure a range validation rule that requires the field value to be ∈ [0,100]. Values ​​outside this range are marked as abnormal.

[0103] Integrity verification rules: Configure non-empty verification rules for the "Task Name" and "Agent Number" fields to determine whether the field values ​​are empty. If they are empty, mark them as missing anomalies.

[0104] Business association verification rules: Configure association verification rules for the "Task ID" and "Project ID" fields to verify whether the project ID corresponding to the task ID is consistent with the project ID currently being queried. If they are inconsistent, they are marked as association anomalies.

[0105] As a further improvement, such as Figure 5 As shown, the large model-driven task generation module 04 includes a task generation principle unit 041, a task management and execution unit 042, and an intelligent agent collaboration unit 043 connected in sequence.

[0106] The task generation principle unit 041 is used to parse the business requirements of the qualified dataset, generate structured task assignments, and transmit the structured task assignments to the task management and execution unit 042.

[0107] The task management and execution unit 042 is used to arrange and plan the execution of structured task assignments, generate task execution plans and transmit them to the intelligent agent collaboration unit 043.

[0108] The agent coordination unit 043 is used to connect with the adapted agents according to the task execution plan, issue tasks to the agents, organize multiple agents to carry out the work collaboratively, generate multiple task execution information, summarize the multiple task execution information, and output the final orchestration result.

[0109] It should be noted that the large model-driven task generation module 04 includes a task generation principle unit 041, a task management and execution unit 042, and an intelligent agent collaboration unit 043. The task generation principle unit 041 adopts the Transformer architecture large model, extracts key information of user needs through the semantic understanding sub-unit, and generates a task framework by combining it with the knowledge base.

[0110] Among them, the task generation principle unit 041 adopts the Transformer architecture large model, extracts key information of user needs (subject, object, time, etc.) through the semantic understanding sub-unit, generates a task framework in combination with the knowledge base, and decomposes it into executable sub-tasks through the task refinement sub-unit.

[0111] Task Management and Execution Unit 042: Includes a task priority sorting subunit (based on the Analytic Hierarchy Process, AHP), a task allocation subunit (using the Hungarian algorithm), a progress tracking subunit, and a feedback adjustment subunit, realizing full lifecycle management of tasks;

[0112] Intelligent Agent Collaboration Unit 043: It realizes information sharing between large models and intelligent agents through standardized interaction interfaces, allocates planning and execution tasks based on the capability complementarity mechanism, and optimizes the collaboration strategy by adopting the collaboration effect evaluation subunit.

[0113] Specifically, after receiving the validated dataset from the data verification module, the task generation principle unit 041 first calls the Transformer architecture model from its internal semantic understanding subunit to perform multiple rounds of parsing on the structured fields and semi-structured text in the dataset: First, it extracts key user requirement information from the validated dataset, including the business entity (e.g., "XX system upgrade project team"), the execution object (e.g., "collaborative task progress data"), time requirements (e.g., "within the current quarter"), constraints (e.g., "statistics by department"), and related business rules (e.g., "project progress update frequency not less than once per day"). Second, it maps the extracted key information to a preset knowledge base, which stores project management models for enterprise office scenarios. The large model combines preset information such as task breakdown rules and execution standards to generate a preliminary task framework, such as "For the XX system upgrade project team, calculate the execution progress of collaborative tasks of each department in the current quarter, summarize by department and generate a progress report"; the third step is to break down the preliminary task framework into directly executable structured tasks by the task refinement sub-unit, clarifying the input parameters (such as "project ID, department list, time interval"), output requirements (such as "progress statistics table, exception description"), execution standards (such as "data format is Excel, including completion rate and deviation rate fields") and dependencies of each task, and finally forming a structured task list containing multiple sub-tasks, and transmitting the structured tasks to the task management and execution unit 042;

[0114] After receiving a structured task, the Task Management and Execution Unit 042 first uses the Analytic Hierarchy Process (AHP) to determine and rank the tasks based on their priorities: Step 1: Define priority dimensions, including urgency (number of days until deadline), importance (weight of impact on overall project progress), dependency level (impact of prerequisite tasks on this task), and resource consumption (estimated computational power required by the agent), and assign weights to each dimension, such as urgency 0.4, importance 0.3, dependency level 0.2, and resource consumption 0.1. Step 2: Quantify and score each dimension for each task, such as assigning 10 points for urgency ≤ 3 days until deadline, 7 points for > 3 days ≤ 7 days, and 3 points for > 7 days. Step 3: Calculate the overall priority score for each task according to the weights, using the formula: "Overall Score = Urgency Score × 0.4 + Importance Score × 0.3 + Dependency Level Score × 0.2 + Resource Consumption Score × 0.1", and rank tasks from highest to lowest score.

[0115] Subsequently, the task allocation subunit uses the Hungarian algorithm to match and allocate tasks to agents: First, a "task-agent" adaptation cost matrix is ​​constructed, where rows represent structured tasks and columns represent agents. The values ​​in the matrix are the reciprocal of the cosine similarity between the task vector and the agent's capability vector (higher similarity means lower adaptation cost). Second, the Hungarian algorithm is used to find the allocation scheme with the lowest adaptation cost, ensuring that each task is matched with the agent whose capability is most suitable and whose load is lowest, while avoiding assigning the same agent to high-priority tasks exceeding its load limit simultaneously. After allocation, the progress tracking subunit establishes a full lifecycle status record for each task, periodically retrieves the agent's execution progress data, and marks it. The task status is "Pending execution / Executing / Completed / Abnormal", and the timestamp and progress percentage of each status are recorded. The feedback adjustment subunit monitors the task execution status in real time. When the task progress is lagging (e.g., the actual completion rate is 20% lower than expected), the agent load rate exceeds 80%, or the task execution error occurs, the adjustment logic is triggered: such as temporarily increasing the priority of the lagging task, reassigning some low-priority tasks of the overloaded agent to the idle agent, and generating retry or manual intervention prompts for the error task. Finally, a task execution plan containing task priority sorting, agent allocation results, execution sequence (e.g., "Start task B after task A is completed") and expected execution nodes is generated and transmitted to the agent coordination unit 043.

[0116] After receiving the task execution plan, the agent coordination unit 043 first achieves information sharing between the large model and the agents through a standardized interaction interface. This interface adopts a unified JSON communication protocol, defining standard message formats for task assignment, status reporting, and result transmission. For example, the task assignment message includes fields such as "task ID, task content, input parameters, execution criteria, and deadline," while the status reporting message includes fields such as "task ID, agent ID, current status, progress percentage, and error information." Subsequently, it organizes multi-agent collaborative operations based on a capability complementarity mechanism: First, according to the capability requirement tags of each task in the task execution plan (such as "data processing," "document generation," and "report analysis"), it matches agents with corresponding capability tags. For example, the "progress data organization" task is assigned to an agent with "data processing" capability, and the "progress report generation" task is assigned to an agent with "document generation" capability. Second, for complex tasks requiring multiple capabilities to cooperate (such as "data organization + report generation"), The system establishes a task flow link between intelligent agents. After the data processing agent completes data organization, it transmits the results to the document generation agent through a standardized interface. The collaboration unit monitors and records the flow process. During the operation, the collaboration effect evaluation subunit synchronously collects the operation data of each intelligent agent, including task completion time, error rate, and interaction response time. After the operation is completed, the collaboration efficiency is evaluated. For example, the average collaboration time of the "data processing agent + document generation agent" combination is calculated. If the time exceeds a preset threshold, the agent combination in subsequent task allocation is optimized. During the multi-agent operation, each intelligent agent reports the corresponding task execution information to the collaboration unit after completing a sub-task or updating its status. This information includes the task ID, agent ID, execution start and end time, actual completion progress, error log, and output results. The collaboration unit collects this information throughout the process and finally summarizes the execution information of multiple tasks to output the final arrangement result containing the entire process operation record, task completion status, and intelligent agent collaboration data.

[0117] For example, this system adopts a distributed deployment architecture to meet the power business needs of a certain city. The server cluster is deployed in the computer room of the city's power dispatch center, with the following configuration: The application server uses two Intel Xeon E5-2690 v4 CPUs, 64GB of memory, and 1TB SSD storage, running CentOS 7.9. It is used to deploy the intelligent agent orchestration module, intelligent query module, and data verification module to ensure the stable operation of core businesses such as defect handling of 110kV and above substations and regional electricity consumption statistics in the city. The database adopts a MySQL 8.0 master-slave architecture (the master database is deployed in the main computer room of the dispatch center, and the slave database is deployed in the backup computer room) to store the city's structured power business data, including substation equipment ledgers, defect order information (covering substations in Duanzhou, Dinghu, Gaoyao, and other districts and counties), and residential and industrial electricity consumption data (distinguishing between Duanzhou urban area, Sihui City, Guangning County, and other areas). MongoDB is also used. The 5.0 replica set stores unstructured data, such as images of defect sites and electricity consumption analysis reports. The agents are deployed using Docker containers and orchestrated using Kubernetes. The resource pool initially configures 25 agent instances, including 5 dedicated agents for power defect handling and 5 for electricity consumption data statistics. It supports elastic scaling based on the city's peak electricity consumption periods (such as summer months like July-August and around the Spring Festival). The large model uses a finely tuned GPT-3.5 architecture. Training data incorporates local power industry terminology (such as "power supply bureau," "Dinghu substation," and "Xijiang River transmission line") and business processes, and is deployed on an NVIDIA A100 GPU server to ensure the accuracy and timeliness of natural language instruction parsing.

[0118] For the substation defect report processing procedure:

[0119] Step 1: The maintenance personnel of Dinghu Substation in a certain city upload the defect report file "Dinghu 110kV Substation October 2024 Defect Report.xlsx" and on-site pictures of the defective equipment (such as the oil leakage photo of the bushing of the No. 1 main transformer) through the "Power Operation and Maintenance Collaboration Group". The file contains information such as the name of the defective equipment, defect type, defect level, discovery time, and maintenance personnel. The upload operation triggers the Enterprise WeChat API (Application Programming Interface) to send a notification to the system collaborative interaction unit, along with the temporary access URL (Uniform Resource Locator) of the file and the uploader's information (employee number: ZQYW2023089, name: Engineer Chen).

[0120] Step 2: The collaborative interaction unit calls the Enterprise WeChat SDK (Software Development Kit) to obtain file permissions and uses the Apache POI (Poor Obfuscation Implementation) component to parse the Excel defect report: extracting file metadata (filename "Dinghu 110kV Substation October 2024 Defect Report.xlsx", creation time 2024-10-09 08:15:30, modification time 2024-10-09 08:20:15, total number of defect reports 12); parsing the table content, extracting "Defect Number (e.g., DQ202410001)", "Equipment Name (e.g., No. 1 Main Transformer)", "Defect Type (e.g., Bushing Oil Leakage)", "Defect Level (Urgent / Important / General)", and "Discovery Time (2024-10-09)". Key fields such as "07:30" and "District / County (Dinghu District)" are converted into JSON structured data. At the same time, the defect site images are parsed to extract information such as shooting time and GPS location (Dinghu Substation, 23.156°N, 112.597°E). The file body is stored in a certain city's power distributed file system, generating a unique identifier "ZQDQ-DEFECT-20241009-001", and synchronized to the "Defect Order Management Table" in the MySQL database.

[0121] Step 3: The collaborative interaction unit sends the defect order identifier and structured data to the task parsing unit. The natural language understanding subunit, combined with the power business scenario (preset rules such as "substation defect handling process" and "emergency defect response within 4 hours"), generates the task instruction "Classify and process the defect orders of Dinghu 110kV substation in October 2024, prioritize the handling of emergency defects and generate a rectification plan"; the task decomposition subunit identifies the task dependencies (defect level determination → defect classification → responsibility allocation → rectification plan generation) through the topology sorting algorithm, and decomposes it into 4 atomic tasks: T1 "Defect level review (confirm 2 emergency defects, important defects) The system consists of four sub-units: T1 (5 general defects), T2 (defect type classification, categorized by 'transformer defects', 'switchgear defects', and 'line defects'), T3 (assignment of rectification responsibilities, matching personnel with the Dinghu District maintenance team), and T4 (generation of rectification plan, specifying the time limit, steps, and safety measures for emergency defect rectification). The priority determination sub-unit assigns priority coefficients to atomic tasks according to power defect management regulations: T1 (1.0, basic determination stage), T2 (0.9, classification and statistical requirements), T3 (0.8, determination of the executing entity), and T4 (0.7, solution output stage). Finally, a task execution tree is constructed and stored in the database.

[0122] Step 4: The agent scheduling unit initiates the scheduling process. The agent resource pool management subunit queries the current status of power-specific agents, where power defect analysis agent A (CPU utilization 45%), personnel allocation agent B (CPU utilization 40%), and planning generation agent C (CPU utilization 38%) are in a low-load state. The capability matching subunit calculates the similarity between atomic tasks and agent capability vectors: T1 has a Sim value of 0.95 with the defect analysis agent, T2 has a Sim value of 0.92 with the defect analysis agent, T3 has a Sim value of 0.89 with the personnel allocation agent, and T4 has a Sim value of 0.91 with the planning generation agent. The load balancing subunit, based on the distribution of maintenance resources in each district and county (the maintenance team in Dinghu District currently has sufficient on-duty personnel), determines the scheduling scheme: T1 and T2 are assigned to defect analysis agent A, T3 is assigned to personnel allocation agent B, and T4 is assigned to planning generation agent C. The task instructions are sent via the MQTT protocol (Message Queuing Telemetry Transport), along with the defect form structured data and a link to the power defect handling specification document.

[0123] Step 5: The monitoring unit tracks the task execution status in real time. The task status acquisition subunit obtains the progress of each atomic task every second. When it is found that the execution time of the T4 "Emergency Defect Rectification Plan Generation" task exceeds twice the standard deviation (5 minutes) of the historical average of similar tasks (20 minutes) after 30 minutes, the anomaly detection subunit triggers the power dispatch center alarm system (audio-visual alarm + SMS notification to the operation and maintenance person in charge). The dynamic scheduling subunit immediately queries the backup plan generation agent D (CPU utilization 35%) and sends a task migration instruction to it, which includes the information of the completed emergency defect equipment (No. 1 main transformer, No. 3 circuit breaker) and rectification requirements (in accordance with the "Guidelines for Defect Rectification of 110kV Substations of Power Supply Bureau"). Agent D continues to generate rectification plans based on the intermediate results to avoid time delays caused by task restart.

[0124] Step 6: After each agent completes its task, it reports the results. Defect analysis agent A returns the defect level review results (confirming DQ202410001 and DQ202410003 as emergency defects) and a type classification table (4 transformer defects, 5 switchgear defects, and 3 line defects). Personnel allocation agent B matches Li (responsible for transformer defects) and Zhang (responsible for switchgear defects) from the Dinghu District maintenance team as the rectification responsible persons. Planning agent D outputs a rectification plan (emergency defects require on-site handling to be completed before 12:00 on October 9, 2024, and the rectification steps include "power outage application → equipment inspection → insulation test → power restoration"). The execution monitoring unit integrates the results and pushes them to the "Power Maintenance Collaboration Group" through the collaborative interaction unit. At the same time, the large model module generates the task "Complete the emergency defect rectification and acceptance of Dinghu 110kV substation before October 10, 2024" and synchronizes it to the power maintenance management system of a certain city, setting a reminder 2 hours before acceptance to ensure closed-loop management of defect handling.

[0125] For the regional electricity consumption statistical analysis process:

[0126] Step 1: Personnel at the municipal power dispatch center input the natural language question "Query the electricity consumption of residents and industries in each district and county of a certain city (Duanzhou, Dinghu, Gaoyao, Sihui, Guangning, Deqing, Fengkai, Huaiji) in October 2024, calculate the month-on-month growth rate compared with September, and sort them in descending order by industrial electricity consumption" through the intelligent query interface on the system's web terminal. At the same time, they select the result display format as "combined chart + data table".

[0127] Step 2: The problem understanding unit initiates semantic parsing. The semantic parsing sub-unit segments words to obtain keywords such as "October 2024", "districts and counties of a certain city", "residential electricity consumption", "industrial electricity consumption", "September", "month-on-month growth rate", and "industrial electricity consumption in descending order". Dependency parsing is used to determine the query dimensions and calculation logic. The intent recognition sub-unit is classified as "composite statistics of power data". The knowledge graph association sub-unit queries the power business knowledge graph (stores related information such as "8 districts and counties under the jurisdiction of a certain city", "residential electricity code 01", "industrial electricity code 02", and "September 2024 is the statistical comparison period"), and supplements localized information such as "electricity consumption data of Duanzhou District including urban areas and surrounding towns" and "industrial electricity consumption data of Sihui City including high-tech industrial parks", generating a standardized query intent: {"Time range": {"Statistical period": "2024-10-01 to 2024-10-31", "Comparison period": "2024-09-01 to 2024-09-30"}. "Regional Scope": ["Duanzhou", "Dinghu", "Gaoyao", "Sihui", "Guangning", "Deqing", "Fengkai", "Huaiji"], "Electricity Consumption Type": ["Residential Electricity Consumption (01)", "Industrial Electricity Consumption (02)"], "Calculation Logic": ["Statistics by Region and Type", "Calculation of Month-on-Month Growth Rate", "Industrial Electricity Consumption in Descending Order"], "Display Format": "Combined Chart + Table"}.

[0128] Step 3: The data retrieval unit performs multi-database retrieval based on the query intent. The data source selection sub-unit determines the primary data source as the real-time electricity consumption business database (stores electricity consumption data from January 2024 to the present, updated every 15 minutes), and the backup data source as the historical electricity consumption database (stores data from 2023 to the present, used for comparative analysis). The index building sub-unit has built B-tree indexes for the "Statistical Time", "District / County Code", and "Electricity Type Code" fields of the "Electricity Data Table". The query optimization sub-unit optimizes the initial SQL statement into an index scan execution plan. The distributed query sub-unit decomposes the query task into 3 sub-tasks using Spark SQL: Sub-task 1 queries the real-time database for October's regional and type-based electricity consumption data, Sub-task 2 queries the historical database for September's regional and type-based electricity consumption data, and Sub-task 3 associates the district / county code with the district / county name mapping table. These tasks are distributed to 3 Executor nodes for parallel execution. After each node completes its execution, the data is aggregated to the Driver node, forming a preliminary dataset containing 160 records (8 districts / counties × 2 types of electricity consumption × 10 days of sampled data).

[0129] Step 4: The data retrieval unit sends the preliminary dataset to the data verification module, and the verification rule formulation unit loads the power data verification rules (such as "the daily average value of residential electricity consumption must not be negative", "the absolute value of the month-on-month growth rate of industrial electricity consumption generally does not exceed 30%", and "data in the same district and county for the same time period must be consistent"). The verification execution unit uses parallel verification sub-units to verify data in groups according to "district / county - electricity consumption type". The total number of data entries is N=160, and 3 abnormal data entries are found: a record of industrial electricity consumption in Sihui City in October exceeds the average value in September by more than 2 times (after verification, it was found that this was caused by the electricity consumption of newly added enterprises in the high-tech industrial park, which is a reasonable fluctuation); one residential electricity consumption data entry in Guangning County has an incorrect format (which has been automatically corrected); and one data entry in Fengkai County has duplicate entries (which have been deduplicated). The verification pass rate is calculated according to the formula R=(NE) / N×100%, where E=0 (all 3 data entries have been corrected or judged to be reasonable), so R=100%. After the verification is completed, a verification report is generated, which includes an explanation of abnormal data processing.

[0130] Step 5: The data verification module sends the verified dataset to the answer generation unit. The data integration sub-unit calculates the total residential and industrial electricity consumption in October, the total consumption in September, and the month-on-month growth rate for each district and county (e.g., the total industrial electricity consumption in Sihui City in October was 520 million kWh, and in September it was 480 million kWh, a month-on-month increase of 8.3%). The results formatting sub-unit generates a combined chart according to user needs (including a bar chart of industrial electricity consumption in each district and county, and a line chart of the month-on-month growth rate of residential and industrial electricity consumption). The chart title is labeled "Statistical Analysis of Electricity Consumption in Each District and County of a Certain City in October 2024", and the axes indicate the district / county name and the electricity consumption unit (100 million kWh). At the same time, a structured table is generated, including "District / County", "Residential Electricity Consumption (October)", and "Month-on-Month Growth Rate of Residential Electricity Consumption". The system displays the fields "Industrial Electricity Consumption (October)" and "Month-on-Month Growth Rate of Industrial Electricity Consumption," sorted in descending order by industrial electricity consumption (Sihui City, Gaoyao District, and Duanzhou District are ranked in the top three). The generated sub-units include a data source description ("Data source: Real-time electricity consumption database, historical electricity consumption database, statistical period 2024-10-01 to 2024-10-31") and a note ("The growth in industrial electricity consumption in Sihui City is mainly due to the new load in the High-tech Industrial Park"). The final results are pushed to the dispatch center's front-end page via WebSocket, supporting chart download (PNG format) and table export (Excel format). The system synchronously records query logs to the power system log database for subsequent business analysis and model optimization.

[0131] As a comparison of technical effects, existing technologies can be used as a reference. Current digital robot systems face several technical bottlenecks in practical applications: First, the agent scheduling mechanism is rigid. Most systems use preset allocation rules, which cannot dynamically adjust based on real-time task priorities and agent resource load. This leads to a contradiction between high-priority tasks queuing and low-load agent resources being idle, severely restricting task execution efficiency. Second, in multi-database collaborative data retrieval scenarios, due to differences in update strategies among various business databases and backup databases, data updates have time differences. Furthermore, network jitter and write anomalies may cause deviations in some database data content. Traditional data retrieval methods either rely solely on a single data source, posing a risk of data errors, or employ simple polling for multi-database comparison, resulting in high latency and difficulty in balancing data accuracy and retrieval efficiency. Third, the task generation process relies excessively on manually configured templates. Users need to predefine task triggering conditions, execution steps, and output formats. For vague natural language requirements such as "generate a competitive analysis task based on this week's market research report," the system cannot autonomously parse and generate a structured task flow, exhibiting extremely poor flexibility and difficulty adapting to dynamic changes in enterprise business scenarios.

[0132] For example, in group file processing scenarios, existing systems can only perform simple file classification and cannot automatically break down tasks and assign them to corresponding agents based on file content; in multi-database collaboration scenarios, traditional data retrieval methods often use a single data source or simple polling, which can easily lead to data errors or data retrieval delays; in terms of generating to-do tasks, existing technologies are unable to automatically generate structured task processes based on natural language instructions.

[0133] To address the aforementioned problems, this invention proposes an intelligent orchestration system for digital robot services, relating to the fields of artificial intelligence and data processing technology. It includes an agent orchestration module, an intelligent query module, a data verification module, and a large-model-driven task generation module, with each module interacting via a data bus. The agent orchestration module includes a task parsing unit, which constructs a task execution tree through task decomposition sub-units. By constructing the task execution tree through the task parsing unit, task dependencies and priorities are clearly defined. Combined with the capability matching and load balancing strategies of the agent scheduling unit, precise adaptation between tasks and agents and dynamic resource allocation are achieved. The system can monitor the agent load status in real time, avoiding overload of single nodes, and adjust execution strategies promptly through an anomaly detection mechanism, effectively improving overall task processing efficiency and resource utilization.

[0134] Compared with existing technologies, it has the following advantages:

[0135] (1) By constructing a task execution tree through the task parsing unit, task dependencies and priorities are clarified. Combined with the capability matching and load balancing strategy of the agent scheduling unit, the system achieves precise adaptation of tasks and agents and dynamic allocation of resources. The system can monitor the agent load status in real time to avoid overload of a single node, and adjust the execution strategy in a timely manner through the anomaly detection mechanism, effectively improving the overall task processing efficiency and resource utilization.

[0136] (2) By adopting distributed parallel retrieval and index optimization technology to improve the data retrieval speed of multiple databases, and combining the block-level parallel verification and automatic repair mechanism of the data verification module, consistency verification of multi-source data is achieved. In response to the problems of database update time difference and content deviation, the accuracy and completeness of data retrieval results are ensured through standardized verification rules and update time priority repair strategy, so as to provide reliable data support for business decision-making.

[0137] (3) By extracting key information about users' natural language needs based on the semantic understanding capabilities of large models, and combining it with the business knowledge base, the system automatically generates a structured task framework and execution process without the need for manual pre-setting of cumbersome rules. The system supports non-technical personnel to initiate tasks directly through natural language, and at the same time realizes the full lifecycle management of task priority sorting, allocation and collaborative execution, which significantly reduces the cost of manual configuration and improves the convenience of office automation.

[0138] In this embodiment of the invention, a digital robot service intelligent orchestration system is provided. The system includes an agent orchestration module, an intelligent query module, a data verification module, and a large model-driven task generation module, all interacting via a data bus. The agent orchestration module receives user-uploaded group files or input natural language instructions, parses the group files or instructions to obtain a task execution tree, schedules agents to execute atomic tasks within the execution tree, monitors the execution status in real time, and outputs business query requirement data. The intelligent query module receives the business query requirement data output by the agent orchestration module and extracts raw retrieval datasets from multiple databases based on the data query requirement data. The data verification module receives the raw retrieval datasets output by the intelligent query module, verifies them, and outputs a verified dataset. The large model-driven task generation module receives the verified datasets output by the data verification module, generates structured tasks based on the verified datasets, schedules multiple agents to execute collaboratively, and generates the final orchestration result. Based on the above scheme, this invention uses an intelligent agent orchestration module to parse user-uploaded group files or input natural language commands and construct a task execution tree. It then schedules intelligent agents to execute atomic tasks and output business query requirement data. The data is then processed sequentially by an intelligent query module to extract the original retrieval dataset, a data verification module to perform data verification, and a large model-driven task generation module to generate structured tasks and schedule multiple intelligent agents to execute them collaboratively. Finally, the orchestration result is output. This eliminates the need for manually pre-setting cumbersome rules, and the system allows non-technical personnel to initiate tasks directly through natural language, thereby improving the convenience of office automation.

[0139] Please see Figure 6 , Figure 6 This is a flowchart illustrating a digital robot service intelligent orchestration method provided in Embodiment 2 of the present invention.

[0140] This invention provides an intelligent orchestration method for digital robot services, comprising:

[0141] Step 601: Receive group files uploaded by users or input natural language instructions, parse the group files or natural language instructions, and obtain the task execution tree.

[0142] Step 602: Execute atomic tasks in the task execution tree and monitor their execution status in real time, and output business query requirement data.

[0143] Step 603: Extract the original retrieval dataset from multiple databases based on business query requirements data.

[0144] Step 604: Validate the original search dataset and output the valid dataset.

[0145] Step 605: Generate structured to-do tasks based on the verified dataset, schedule multiple agents to execute them collaboratively, and generate the final orchestration result.

[0146] Further, step 602 may include the following sub-steps:

[0147] S21. Schedule the agent to execute atomic tasks according to the task execution tree and generate scheduling execution information;

[0148] S22. Based on the task content and execution requirements within the scheduling execution information, track the operation process of the corresponding intelligent agent and collect the execution status information of the intelligent agent;

[0149] S23. Generate business query requirement data based on the task progress and execution results in the execution status information.

[0150] It should be noted that, as Figure 7 As shown, users can upload group files or input natural language commands according to actual business needs. The system first completes task parsing and builds a task execution tree through the intelligent agent orchestration module, and then schedules intelligent agents to execute the corresponding tasks and monitors the overall execution status in real time. Next, the intelligent query module performs multi-database retrieval. During the retrieval process, the business database and backup database are selected as data sources, and the query operations are executed in parallel and the query results are compared. For data with inconsistencies, the data from the data source with the newest update time is used to complete the repair. Then, the data verification module verifies the data consistency. Finally, the system generates to-do tasks based on the large model module, organizes intelligent agents to collaboratively complete the work, and provides feedback on the final execution results.

[0151] In this embodiment of the invention, a method for intelligent orchestration of digital robot services is provided. This method receives user-uploaded group files or input natural language instructions, parses the group files or natural language instructions to obtain a task execution tree, executes atomic tasks in the task execution tree and monitors the execution status in real time, and outputs business query requirement data. Based on the business query requirement data, it extracts raw retrieval datasets from multiple databases, verifies the raw retrieval datasets, and outputs a verified dataset. It generates structured to-do tasks based on the verified datasets, schedules multiple agents to collaboratively execute them, and generates the final orchestration result. Compared to existing methods where users need to predefine task triggering conditions, execution steps, and output formats, and where the system cannot autonomously parse and generate structured task flows for vague natural language requirements such as "generate a competitor analysis task based on this week's market research report," this invention eliminates the need for manually pre-setting cumbersome rules. The system allows non-technical personnel to directly initiate tasks using natural language, thereby improving the convenience of office automation.

[0152] This invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program; when the computer program is executed by the processor, the processor performs the steps of the digital robot service intelligent orchestration method as described in the above embodiments.

[0153] This invention also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the steps of the intelligent orchestration method for digital robot services as described in the above embodiments.

[0154] This invention also provides a computer program product, including a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer performs the steps of the digital robot service intelligent orchestration method as described in the above embodiments.

[0155] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0156] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0157] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0158] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0159] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A digital robotic service intelligent orchestration system, characterized in that, The system includes an intelligent agent orchestration module, an intelligent questioning module, a data verification module, and a large model-driven task generation module that interact via a data bus. The intelligent agent orchestration module is used to receive group files uploaded by users or input natural language instructions, parse the group files or natural language instructions to obtain a task execution tree, schedule intelligent agents to execute atomic tasks in the task execution tree and monitor the execution status in real time, and output business query requirement data. The intelligent query module is used to receive the business query requirement data output by the intelligent agent orchestration module, and extract the original retrieval dataset from multiple databases based on the business query requirement data. The data verification module is used to receive the original retrieval dataset output by the intelligent query module, verify it, and output a qualified dataset. The large model-driven task generation module is used to receive the qualified dataset output by the data verification module, generate structured tasks based on the qualified dataset, schedule multiple agents to execute collaboratively, and generate the final orchestration result.

2. The digital robotic service intelligent orchestration system of claim 1, wherein, The agent orchestration module includes a task parsing unit, an agent scheduling unit, an execution monitoring unit, and a collaborative interaction unit connected in sequence. The task parsing unit is used to receive the group file or the natural language instruction, parse the group file or the natural language instruction, generate a task execution tree, and transmit the task execution tree to the agent scheduling unit; The agent scheduling unit is used to schedule the agent to execute atomic tasks according to the task execution tree, generate scheduling execution information and transmit it to the execution monitoring unit; The execution monitoring unit is used to track the operation process of the corresponding intelligent agent according to the task content and execution requirements in the scheduling execution information, collect the execution status information of the intelligent agent, and then transmit the execution status information to the collaborative interaction unit. The collaborative interaction unit is used to generate business query requirement data based on the task progress and execution result in the execution status information, and then transmit the business query requirement data to the intelligent query module.

3. The digital robotic service intelligent orchestration system of claim 1, wherein, The intelligent questioning module includes a question understanding unit and a data retrieval unit connected in sequence; The problem understanding unit is used to interpret the business query requirement data, generate standardized query information, and transmit it to the data retrieval unit. The data retrieval unit is used to extract the original retrieval dataset from the multiple databases according to the query range defined by the standardized query information and transmit it to the data verification module.

4. The intelligent orchestration system for digital robot services according to claim 1, characterized in that, The data verification module includes a data acquisition unit, a rule formulation unit, a verification execution unit, and an error handling unit connected in sequence. The data acquisition unit is used to aggregate the original retrieval dataset to obtain the aggregated original retrieval dataset, and then transmit the aggregated original retrieval dataset to the verification execution unit. The rule-making unit is used to construct standardized verification rules and transmit them to the verification execution unit; The verification execution unit is used to verify the aggregated original retrieval dataset using the standardized verification rules, generate initial screening verification data, and transmit it to the error handling unit. The error handling unit is used to correct abnormal data and remove invalid data from the initial screening verification data, generate a qualified verification dataset, and transmit the qualified verification dataset to the large model-driven task generation module.

5. The intelligent orchestration system for digital robot services according to claim 1, characterized in that, The large model-driven task generation module includes a task generation principle unit, a task management and execution unit, and an intelligent agent collaboration unit connected in sequence. The task generation principle unit is used to parse the business requirements of the qualified dataset, generate structured task assignments, and transmit the structured task assignments to the task management and execution unit. The task management and execution unit is used to arrange and plan the structured task assignments, generate task execution plans, and transmit them to the intelligent agent collaboration unit. The intelligent agent coordination unit is used to connect with the adapted intelligent agents according to the task execution plan, issue tasks to the intelligent agents and organize multiple intelligent agents to carry out the work collaboratively, generate multiple task execution information, summarize the multiple task execution information, and output the final arrangement result.

6. A method for intelligent orchestration of digital robot services, characterized in that, include: Receive group files uploaded by users or input natural language instructions, parse the group files or natural language instructions, and obtain a task execution tree; Execute atomic tasks in the task execution tree and monitor their execution status in real time, then output business query requirement data. Based on the aforementioned business query requirements data, the original retrieval dataset is extracted from multiple databases; The original retrieved dataset is validated, and a valid dataset is output. Based on the verified dataset, structured to-do tasks are generated, and multiple agents are scheduled to execute them collaboratively to generate the final orchestration result.

7. The intelligent orchestration method for digital robot services according to claim 6, characterized in that, The process involves executing atomic tasks in the task execution tree, monitoring their execution status in real time, and outputting business query requirement data, including: The agent is scheduled to execute atomic tasks according to the task execution tree, and scheduling execution information is generated; Based on the task content and execution requirements within the scheduling execution information, the operation process of the corresponding intelligent agent is tracked, and the execution status information of the intelligent agent is collected; Based on the task progress and execution results in the execution status information, business query requirement data is generated.

8. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the intelligent orchestration method for digital robot services as described in any one of claims 6-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the intelligent orchestration method for digital robot services as described in any one of claims 6-7.

10. A computer program product, characterized in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, wherein when the program instructions are executed by a computer, the computer performs the steps of the intelligent orchestration method for digital robot services as described in any one of claims 6-7.

Images