Intelligent switching operation ticket forming system and method based on large model

Abstract

The invention aims to provide the intelligent switching operation ticket forming system and method based on the large model, which are high in efficiency, high in accuracy and low in cost. The system comprises a data access layer (1), a knowledge support layer (2), an intelligent agent execution layer (3), an application layer (4) and a self-learning optimization layer (5). The method comprises the following steps: S1, a data access layer obtains a scheduling instruction and equipment real-time data and transmits the scheduling instruction and the equipment real-time data to an intelligent agent execution layer; s2, an instruction analysis agent (31) analyzes the scheduling instruction to generate structured feature parameters; s3, the statement generation agent (32) generates an operation ticket first draft in combination with the feature parameters, the topological data and the finished ticket knowledge base; s4, performing logic, normative and multi-ticket comparison verification on the first draft by the verification agent (33), and outputting a verification result; and S5, the application layer displays the operation order passing the verification, and feeds back the data to the self-learning optimization layer to update the system. The method is applied to the technical field of power system operation and maintenance intelligentization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system operation and maintenance intelligentization, and particularly relates to a switching operation order intelligent ticketing system and method based on a large model. BACKGROUND

[0002] The switching operation order is the core safety document of substation operation and maintenance. The traditional manual ticketing mode relies on static account books and personal experience, and has two core pain points: first, it cannot synchronize the device operation mode in real time, resulting in inconsistent operation steps with the actual situation; second, the misoperation prevention system and the ticketing system are independent, and only manual review can be performed afterwards, which is prone to missing risks. Although the existing rule library + expert system mode realizes partial automation, it is not suitable for complex scenarios such as multi-interval linkage and temporary operation mode, and the ticketing accuracy and efficiency are difficult to meet the intelligent operation and maintenance requirements of the power grid. Therefore, it is urgent to build an intelligent ticketing system that integrates large models and real-time data to solve the problems of low efficiency and low accuracy of manual ticketing and lack of standardization. SUMMARY

[0003] The technical problem to be solved by the present application is to overcome the deficiencies of the prior art, provide a switching operation order intelligent ticketing system and method based on a large model, realize accurate analysis of dispatching instructions, second-level generation and real-time verification of operation orders, continuously improve the ticketing capability through a self-learning mechanism, reduce labor costs, and ensure the safety of power grid operation and maintenance.

[0004] The technical solution adopted by the present application is a switching operation order intelligent ticketing system based on a large model, which comprises a data access layer, a knowledge support layer, an intelligent agent execution layer, an application layer, and a self-learning optimization layer. The data access layer is used to interface with the centralized control system, the misoperation prevention system, and the dispatching system, and to obtain real-time state data of devices, graph model data, and dispatching instruction data. The knowledge support layer comprises an intelligent ticketing rule expert library, a ticketing vector knowledge base, and a verification rule knowledge base. The intelligent ticketing rule expert library constructs operation logic rules according to 8 types of dimensions. The ticketing vector knowledge base stores annotated historical operation orders, typical orders, and operation procedure samples. The intelligent agent execution layer comprises an instruction analysis intelligent agent, a statement generation intelligent agent, and a verification intelligent agent. The application layer provides operation order generation, review, and typical order management functions. The self-learning optimization layer iteratively updates the knowledge base and intelligent agent model based on data feedback.

[0005] Further, the data access layer realizes millisecond-level synchronization of device state data through a dispatching data network vertical encryption channel. The device state data includes primary device switching on-off state, secondary protection pressure plate throw-in and throw-out information, and interval operation state.

[0006] Further, the instruction analysis agent comprises a semantic recognition module and a feature extraction module, the semantic recognition module performs semantic analysis on the dispatching instruction based on the Guangming large model, and the feature extraction module extracts structured parameters such as station name, voltage level, operation object, initial state, target state and task order keyword.

[0007] Further, the sentence generation agent comprises a topology analysis module and a text generation module, the topology analysis module identifies the device wiring mode and interval type in combination with the graph data, and the text generation module generates standardized operation sentences based on the retrieval result of the ticket generation vector knowledge base and the large model.

[0008] Further, the verification agent comprises a logic checking module, a specification checking module and a multi-ticket comparison module, the logic checking module checks the anti-misoperation logic based on the topology algorithm, the specification checking module checks the uniformity of operation terms and the completeness of items, and the multi-ticket comparison module realizes the difference identification of power-off tickets and historical tickets, and power-on tickets and power-off tickets.

[0009] Further, the self-learning optimization layer comprises a data filtering module, a model iteration module and a rule updating module, the data filtering module filters compliant historical data to avoid false learning, the model iteration module optimizes the accuracy of large model analysis and generation based on audit feedback, and the rule updating module automatically supplements the operation logic of complex scenes not covered by the rule base.

[0010] An intelligent switching operation ticket generation method based on the above system, the method comprising the following steps: S1, the data access layer acquires dispatching instructions and real-time device data and transmits them to the agent execution layer; S2, the instruction analysis agent analyzes the dispatching instructions to generate structured feature parameters; S3, the sentence generation agent generates an operation ticket draft in combination with the feature parameters, topology data and ticket generation knowledge base; S4, the verification agent checks the draft for logic, specification and multi-ticket comparison, and outputs the verification result; S5, the application layer displays the operation ticket that passes the verification, and feeds back the data to the self-learning optimization layer to update the system.

[0011] Further, in step S2, the instruction analysis agent constructs a structured prompt sentence by retrieving the instruction decomposition sample vector knowledge base, guides the large model to generate feature parameters, and the analysis accuracy of the feature parameters is ≥95%.

[0012] Further, in step S3, the sentence generation agent adopts a rule base and large model dual driving mode, generates operation steps by calling the rule base in a conventional scene, and dynamically deduces through the large model in a complex scene, and the ticket generation time is shortened to seconds.

[0013] Further, in step S5, the self-learning optimization layer updates the ticket generation rule library and the logical audit standard every quarter by comparing and analyzing the ticket generation deviation and the audit problem, so that the ticket generation accuracy is improved to more than 97% annually, and in step S4, the verification agent dynamically verifies the compliance of the operation sequence in real time by calling the device remote signaling data of the centralized control system through the logical verification module, and the verification response time is not more than 100 ms.

[0014] The beneficial effects of the present application are: in the present application, the semantic recognition and text generation capability of the large language model replace manual line-by-line writing, reducing repetitive labor, and through automatic analysis and generation, the ticket generation time is shortened to seconds, and the efficiency is greatly improved; through the accurate analysis of the instruction analysis agent and the support of the vector knowledge base, the dual driving of the rule library + large model and multi-dimensional verification are realized, avoiding errors caused by human experience differences, greatly improving accuracy and significantly reducing the risk of misoperation; through unified terminology and operation logic, the non-standard problems on the ticket surface caused by personnel experience differences are eliminated, reducing rework costs and enhancing standardization; through the self-learning mechanism, the rules and models are dynamically updated to adapt to the upgrade of power grid equipment and changes in operation and maintenance mode, reducing system maintenance costs. BRIEF DESCRIPTION OF DRAWINGS

[0015] Fig. 1 is a simple architecture diagram of the system of the present application; Fig. 2 is a command analysis flowchart of the present application; Fig. 3 is a method flowchart of the present application. DETAILED DESCRIPTION

[0016] As shown in Figs. 1-3 , the present application is an intelligent switching operation ticket generation system based on a large model, which comprises a data access layer 1, a knowledge support layer 2, an agent execution layer 3, an application layer 4 and a self-learning optimization layer 5; The data access layer 1 is used to connect the centralized control system 6, the anti-misoperation system 7 and the command system 8, and obtain real-time state data, graphic model data and dispatching instruction data of the equipment; The knowledge support layer 2 comprises an intelligent ticket generation rule expert library 21, a ticket generation vector knowledge base 22 and a verification rule knowledge base 23, the intelligent ticket generation rule expert library 21 constructs operation logic rules according to 8 dimensions of dispatching order type, equipment type, voltage level, wiring method, equipment operating state, operation task scene, safety anti-misoperation level and industry / enterprise regulation standard, and the ticket generation vector knowledge base 22 stores annotated historical operation tickets, typical tickets and operation regulation samples; The agent execution layer 3 comprises an instruction analysis agent 31, a sentence generation agent 32 and a verification agent 33; The application layer 4 provides functions for generating, reviewing, and managing typical operation tickets; The self-learning optimization layer 5 iteratively updates the knowledge base and agent model based on data feedback.

[0017] Specifically, the data access layer 1 achieves millisecond-level synchronization of device status data through a vertical encrypted channel of the scheduling data network. This device status data includes the on / off status of primary equipment switches, the activation / deactivation information of secondary protection pressure plates, and the interval operation status. The instruction analysis agent 31 includes a semantic recognition module 311 and a feature extraction module 312. The semantic recognition module 311 performs semantic parsing of scheduling instructions based on the Guangming large model, while the feature extraction module 312 extracts structured parameters such as station name, voltage level, operation object, initial state, target state, and task command keywords. The statement generation agent 32 includes a topology analysis module 321 and a text generation module 322. The topology analysis module 321 identifies the device wiring method and interval type by combining graphical model data, while the text generation module 322 generates standardized operation statements based on the results of the ticket vector knowledge base retrieval and the large model deduction. The verification intelligent agent 33 includes a logic verification module 331, a standardization verification module 332, and a multi-ticket comparison module 333. The logic verification module 331 verifies error prevention logic such as "disconnecting the switch before pulling the disconnector" based on topology algorithms. The standardization verification module 332 checks the consistency of operational terminology and the completeness of inspection items. The multi-ticket comparison module 333 identifies the differences between power outage tickets and historical tickets, and between power supply tickets and power outage tickets. The self-learning optimization layer 5 includes a data filtering module 51, a model iteration module 52, and a rule update module 53. The data filtering module 51 filters compliant historical data to avoid erroneous learning. The model iteration module 52 optimizes the parsing and generation accuracy of the large model based on audit feedback. The rule update module 53 automatically supplements complex scenario operation logic not covered by the rule base.

[0018] A method for intelligent ticket generation of switching operation tickets based on the above system, the method includes the following steps: S1. Data access layer 1 obtains scheduling instructions and real-time device data and transmits them to intelligent agent execution layer 3; S2, the instruction analysis agent 31 parses the scheduling instructions and generates structured feature parameters; S3, the statement generation agent 32 combines feature parameters, topological data and ticket knowledge base to generate the first draft of the operation ticket; S4. Verification agent 33 performs logic, standardization, and multi-vote comparison verification on the initial draft and outputs the verification results. S5 and Application Layer 4 display the verified operation tickets and simultaneously feed the data back to the self-learning optimization layer 5 to update the system.

[0019] Further, in step S2, the instruction analysis agent 31 constructs structured prompt statements by retrieving instruction decomposition sample vector knowledge base, guiding the large model to generate feature parameters, with the parsing accuracy of the feature parameters being ≥95%. In step S3, the statement generation agent 32 adopts a dual-drive mode of rule base and large model. In normal scenarios, the rule base is called to generate operation steps, while in complex scenarios, the large model dynamically deduces the process, shortening the ticketing time to the second level. In step S5, the self-learning optimization layer 5 updates the ticketing rule base and logical review standards quarterly by comparing and analyzing invoicing deviations and review issues, thereby increasing the annual ticketing accuracy to over 97%. In step S4, the verification agent 33 calls the equipment remote signaling data of the centralized control system in real time through the logical verification module to dynamically verify the compliance of the operation sequence, with a verification response time of no more than 100ms.

[0020] The present invention will now be described in more detail with reference to specific embodiments.

[0021] I. System Architecture Deployment In this embodiment, the system is deployed at a provincial-level intelligent computing center. Data access layer 1 connects to the centralized control system, anti-misoperation system, and command system of the 220kV Pengshan substation of the Hebei North Power Grid through an encrypted interface to obtain the following data in real time: Real-time equipment status: 220kV section bus voltage 230kV, bus tie 27M switch closed, 271 and 275 line bays in operation, 27A main transformer bay protection pressure plate in operation / disabling status. Diagram data: Double busbar topology, bay numbering and equipment association; Dispatch instruction: "The 220kV section busbar of Pengshan Substation shall be changed from operation to maintenance."

[0022] Knowledge support layer 2 has been completed: Intelligent Ticketing Rules Expert Database 21: Includes preset rules for 220kV double busbar maintenance, such as "first check the status of the bus tie switch → connect the single bus differential pressure plate → disconnect the bus tie control power supply"; Ticket Vector Knowledge Base 22: Stores 5,000 historical operation tickets, 300 sets of typical tickets, and a text sample of the "State Grid Corporation of China Power Safety Work Regulations"; Verification rule knowledge base 23: covers 12 types of error prevention logic, 20 terminology standards and 8 types of inspection item specifications.

[0023] Agent execution layer 3 configuration: Instruction Analysis Agent 31: Based on the Guangming Big Model V2.0, the semantic recognition accuracy reaches 96.2%; Statement generation agent 32: It integrates topology analysis algorithms and supports parsing four types of wiring methods, including double busbars and single busbar segmentation. Verification agent 33: Logic verification response time 80ms, multi-vote comparison difference recognition rate 98%.

[0024] II. Operation Ticket Generation Process The steps for generating an operation ticket for "Pengshan Substation 220kV Section 1 Busbar to be changed from operation to maintenance" using the above system are as follows: S1. Data access layer 1 transmits scheduling instructions and device data to intelligent agent execution layer 3 through an encrypted channel, with a data synchronization delay of 50ms. S2. The intelligent agent 31 analyzes the scheduling instructions and generates characteristic parameters: station name = Pengshan substation, voltage level = 220kV, operation object = 220kV I section bus, initial state = running, target state = maintenance, task command keyword = change to, parsing time 0.3 seconds; S3. The statement generation agent 32 calls the topology analysis module 321 to identify the double busbar connection, retrieves historical tickets with 95% similarity in the ticket vector knowledge base 22, and generates the first draft of the operation ticket by combining the large model. The ticket generation takes 0.8 seconds. S4. Verify that agent 33 performs the verification: Logic verification: Confirmed that there is no risk of "closing the circuit breaker with grounding" and conforms to the logic of "disconnecting the switch before pulling the knife switch"; Standardization verification: Standardize the terminology for "disconnection" and complete the "verify operation panel number" check item; Multiple votes comparison: Compared with historical votes of the same type, there are no differences with missing steps, and the verification time is 0.2 seconds; S5 and Application Layer 4 display the final operation ticket, and at the same time transmit the ticketing data (parsing results, ticketing steps, verification feedback) to the self-learning optimization layer 5 to update the ticketing vector knowledge base 22.

[0025] III. Self-learning optimization effect After the system has been running for 6 months, the statistics for self-learning optimization layer 5 are as follows: Data accumulation: 1200 valid invoice records were collected, including 80 records from complex scenarios (temporary connection adjustments); Model iteration: Instruction parsing accuracy improved from the initial 95% to 97.3%, and parsing accuracy for complex topology scenes improved by 15%; Rule update: Automatically adds 3 new operating rules for protection devices, reducing manual maintenance costs by 70%; Ticket accuracy rate: increased from 92% at the beginning of the launch to 97.1%, achieving the expected goal.

[0026] IV. Benefit Comparison Compared with the manual invoicing method, the benefits of this system are as follows:

[0027] As can be seen from the table above, the present invention can effectively improve the accuracy and efficiency of invoice production, and also greatly reduce the number of rework and the time required for invoice production.

[0028] Finally, it should be emphasized that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart ticketing system for switching operation tickets based on a large model, characterized in that, The system includes a data access layer (1), a knowledge support layer (2), an intelligent agent execution layer (3), an application layer (4), and a self-learning optimization layer (5). The data access layer (1) is used to connect to the centralized control system (6), the anti-misoperation system (7) and the command issuing system (8) to obtain real-time equipment status data, graphic data and scheduling command data; The knowledge support layer (2) includes an intelligent ticketing rule expert library (21), a ticketing vector knowledge library (22), and a verification rule knowledge library (23). The intelligent ticketing rule expert library (21) constructs operation logic rules according to eight dimensions. The ticketing vector knowledge library (22) stores labeled historical operation tickets, typical tickets, and operation procedure samples. The intelligent agent execution layer (3) includes an instruction analysis intelligent agent (31), a statement generation intelligent agent (32), and a verification intelligent agent (33). The application layer (4) provides functions for generating, reviewing and managing typical tickets; The self-learning optimization layer (5) iteratively updates the knowledge base and agent model based on data feedback.

2. The system according to claim 1, characterized in that, The data access layer (1) achieves millisecond-level synchronization of device status data through the longitudinal encryption channel of the scheduling data network. The device status data includes the primary device switch status, secondary protection pressure plate activation / deactivation information, and interval operation status.

3. The system according to claim 1, characterized in that, The instruction analysis agent (31) includes a semantic recognition module (311) and a feature extraction module (312). The semantic recognition module (311) performs semantic parsing of the scheduling instructions based on the Guangming big model. The feature extraction module (312) extracts structured parameters such as station name, voltage level, operation object, initial state, target state and task command keywords.

4. The system according to claim 1, characterized in that, The statement generation agent (32) includes a topology analysis module (321) and a text generation module (322). The topology analysis module (321) identifies the wiring method and interval type of the equipment by combining the graph model data. The text generation module (322) generates standardized operation statements based on the retrieval results of the ticket vector knowledge base and the large model deduction.

5. The system according to claim 1, characterized in that, The verification agent (33) includes a logic verification module (331), a standardization verification module (332), and a multi-ticket comparison module (333). The logic verification module (331) verifies the error prevention logic based on the topology algorithm. The standardization verification module (332) checks the uniformity of operation terms and the completeness of inspection items. The multi-ticket comparison module (333) realizes the difference recognition between power outage tickets and historical tickets, and between power supply tickets and power outage tickets.

6. The system according to claim 1, characterized in that, The self-learning optimization layer (5) includes a data filtering module (51), a model iteration module (52), and a rule update module (53). The data filtering module (51) filters compliant historical data to avoid incorrect learning. The model iteration module (52) optimizes the parsing and generation accuracy of the large model based on audit feedback. The rule update module (53) automatically supplements the operation logic of complex scenarios not covered by the rule library.

7. A method for intelligent ticket generation of switching operation tickets based on the system described in any one of claims 1 to 6, characterized in that, The method includes the following steps: S1, Data Access Layer (1) obtains scheduling instructions and real-time device data and transmits them to the intelligent agent execution layer (3). S2, The instruction analysis agent (31) parses the scheduling instructions and generates structured feature parameters; S3, Statement generation agent (32) combines feature parameters, topological data and ticket knowledge base to generate the first draft of operation ticket; S4. The verification agent (33) performs logical, standardization and multi-vote comparison verification on the draft and outputs the verification results. S5, Application layer (4) displays the operation ticket that has passed the verification, and at the same time feeds the data back to the self-learning optimization layer (5) to update the system.

8. The method according to claim 7, characterized in that, In step S2, the instruction analysis agent (31) constructs a structured prompt statement by retrieving the instruction decomposition sample vector knowledge base, and guides the large model to generate feature parameters. The parsing accuracy of the feature parameters is ≥95%.

9. The method according to claim 7, characterized in that, In step S3, the statement generation agent (32) adopts a dual-drive mode of rule base and large model. In normal scenarios, the rule base is called to generate operation steps, and in complex scenarios, the large model is used for dynamic deduction, reducing the ticket generation time to the second level.

10. The method according to claim 7, characterized in that, In step S5, the self-learning optimization layer (5) updates the invoice rule base and logical audit standards every quarter by comparing and analyzing invoice deviation and audit issues, so that the invoice accuracy rate is increased to more than 97% annually. In step S4, the verification agent (33) calls the remote signaling data of the centralized control system in real time through the logical verification module to dynamically verify the compliance of the operation sequence. The verification response time does not exceed 100ms.

Images