Operation ticket auditing rule self-learning method, operation ticket auditing method, operation ticket auditing system and related equipment

Abstract

The embodiment of the invention provides an operation ticket auditing method and system and related equipment. Wherein the user performs dialogue interaction with the self-learning agent, the self-learning agent performs intention recognition on current dialogue information input by the user in the dialogue interaction by using a generative AI model, and if the intention category of the current dialogue information is recognized as a newly added operation ticket auditing rule, the current dialogue information is reviewed according to the newly added operation ticket auditing rule. And if yes, analyzing the newly-added operation ticket auditing rule in the current dialogue information by utilizing a generative AI model in combination with a knowledge graph and a professional scheduling knowledge base to obtain rule analysis content of the newly-added operation ticket auditing rule. Therefore, self-learning of the newly added operation ticket auditing rule is completed, and an efficient and intelligent operation ticket auditing rule self-learning solution with a self-learning capability is provided. Therefore, in the subsequent operation ticket auditing stage, the operation ticket can be efficiently, accurately and automatically audited by using the self-learned operation ticket auditing rule.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a self-learning method for operation ticket review rules, an operation ticket review method, a system and related equipment. Background Art

[0002] Operation tickets, as an important written document to ensure safe and standardized operations, are widely used in multiple industries, such as power systems, petrochemicals, and rail transit. Operation tickets record every step of an operation in detail, guiding operators to standardize work processes and ensure safe operations, thereby reducing operational errors, improving efficiency, and protecting the safety of personnel and equipment.

[0003] Currently, the review of operation tickets still relies primarily on manual, item-by-item verification. However, manual review is inefficient. Furthermore, due to the complexity and diversity of operation tickets, reviewers can easily miss key details during the review process due to fatigue, lack of experience, or distraction, which can affect the safety and compliance of operational operations. Therefore, it is necessary to research how to develop efficient and accurate automated review rules for operation tickets. Summary of the Invention

[0004] The embodiments of the present application provide a self-learning method for operation ticket review rules, an operation ticket review method, a system and related equipment, and provide an efficient, intelligent and self-learning solution for operation ticket review rules. In this way, during the operation ticket review stage, the self-learned operation ticket review rules are used to perform efficient and accurate automatic review of the operation ticket.

[0005] An embodiment of the present application provides a self-learning method for operation ticket review rules, which is applied to a self-learning intelligent agent. The method includes: obtaining the current dialogue information input by the user in this dialogue interaction; using a generative AI model to identify the intent of the current dialogue information to obtain the intent category of the current dialogue information; if the intent category of the current dialogue information is a new operation ticket review rule, then using a generative AI model in combination with a knowledge graph and a scheduling professional knowledge base to parse the new operation ticket review rule in the current dialogue information to obtain the rule analysis content of the new operation ticket review rule.

[0006] An embodiment of the present application provides an operation ticket review method, including: obtaining the current dialogue information input by the user in this dialogue interaction; using a generative AI model to identify the intent of the current dialogue information; if the intent category of the current dialogue information is identified as a new operation ticket review rule, then using the generative AI model in combination with the knowledge graph and the scheduling professional knowledge base to parse the new operation ticket review rule in the current dialogue information to obtain the rule analysis content of the new operation ticket review rule; using the generative AI model to review the operation instructions in the operation ticket to be reviewed according to the rule analysis content, and outputting the review result of the operation ticket to be reviewed.

[0007] An embodiment of the present application provides an operation ticket processing system, including: a terminal device and a server, the server running a self-learning agent and an audit agent; the terminal device, for responding to the user's input operation in the dialogue interaction interface, obtaining the current dialogue information input by the user in this dialogue interaction and sending it to the server, and displaying the audit result of the operation ticket to be reviewed returned by the server on the dialogue interaction interface; the server, for using the self-learning agent to use a generative AI model to identify the intent of the current dialogue information, if the intention category of the current dialogue information is identified as a new operation ticket review rule, then using the generative AI model in combination with the knowledge graph and the scheduling professional knowledge base to parse the new operation ticket review rule in the current dialogue information to obtain the rule analysis content of the new operation ticket review rule; using the audit agent to use the generative AI model to review the operation instructions in the operation ticket to be reviewed according to the rule analysis content, and sending the audit result of the operation ticket to be reviewed to the terminal device.

[0008] An embodiment of the present application also provides an electronic device comprising: a memory and a processor; the memory is used to store a computer program; the processor is coupled to the memory, and is used to execute the computer program to execute the steps in the self-learning method of operation ticket review rules or the operation ticket review method.

[0009] An embodiment of the present application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor is enabled to implement the steps in the self-learning method of operation ticket review rules or the operation ticket review method.

[0010] An embodiment of the present application also provides a computer program product, including a computer program / instruction. When the computer program / instruction is executed by a processor, the processor is enabled to implement the steps in the self-learning method of operation ticket review rules or the operation ticket review method.

[0011] The technical solution provided by the embodiment of the present application is that the user conducts a dialogue interaction with the self-learning intelligent agent, and the self-learning intelligent agent uses a generative AI model to identify the intent of the current dialogue information input by the user in the dialogue interaction. If the intent category of the current dialogue information is identified as a new operation ticket review rule, the generative AI model is used in combination with the knowledge graph and the scheduling professional knowledge base to parse the new operation ticket review rule in the current dialogue information to obtain the rule parsing content of the new operation ticket review rule. In this way, the self-learning of the new operation ticket review rule is completed, and an efficient, intelligent and self-learning operation ticket review rule self-learning solution is provided. In this way, in the subsequent operation ticket review stage, the self-learned operation ticket review rules can be used to perform efficient and accurate automatic review of the operation ticket, providing an efficient, intelligent and self-learning operation ticket review solution that can effectively ensure the security, standardization and correctness of the operation ticket. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1a An exemplary application scenario diagram provided for an embodiment of the present application; Figure 1b A flowchart of a self-learning method for operation ticket review rules provided in an embodiment of the present application; Figure 2 A flowchart of an operation ticket review method provided in an embodiment of the present application; Figure 3 A flowchart of another operation ticket review method provided in an embodiment of the present application; Figure 4 Another exemplary application scenario diagram provided for an embodiment of the present application; Figure 5 A system architecture diagram of an operation ticket review system provided in an embodiment of the present application; Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0013] To make the purpose, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the specific embodiments of this application and the corresponding drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0014] In the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the access relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. In the text description of the present application, the character " / " generally indicates that the previous and next associated objects are in an "or" relationship. In addition, in the embodiments of the present application, "first", "second", "third", etc. are only used to distinguish the contents of different objects and have no other special meanings.

[0015] It should be noted that when the embodiments of this application involve user information, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation portals for users to choose to authorize or refuse. In addition, the various models involved in this application (including but not limited to language models or large models) are in compliance with relevant laws and standards.

[0016] First, some terms related to the embodiments of this application are introduced: AI Agent: A software or hardware entity that has the ability to autonomously perceive the environment, make decisions, and execute actions in order to complete specific tasks or achieve goals in the simulated or real world.

[0017] A language model (LM) is a model that learns common language structures and knowledge through unsupervised or self-supervised pre-training on large-scale text data. Examples of language model structures include, but are not limited to, bidirectional encoder representation from transformers (BERT), autoregressive language models (ALMs), and generative pre-trained transformers (GPTs). The Transformer module is a neural network structure based on the self-attention mechanism, significantly improving model performance through parallel processing and self-attention.

[0018] Large language models, also known as large language models (LLMs), are natural language processing models with an extremely large number of parameters. Large language models are typically based on deep learning architectures, particularly the Transformer architecture. They learn the complex structure and rich context of language by pre-training on massive amounts of text data. The Transformer architecture, by introducing a self-attention mechanism, addresses the bottleneck problem of traditional neural network models when processing long sequences. Its highly parallelizable nature significantly improves training efficiency. The Transformer architecture consists of an encoder and a decoder.

[0019] A reasoning model is a language model with reasoning capabilities. It can perform complex thinking processes such as logical inference, causal analysis, and relationship identification based on existing information. It can also draw reasonable conclusions or solutions through reasoning when faced with complex problems.

[0020] A generative AI model is an algorithmic model that uses artificial intelligence technology to automatically generate text, pictures, sounds, videos, codes, and other content.

[0021] Chain of Thought (CoT) is a prompt engineering method that guides language models to gradually decompose complex problems and demonstrate the reasoning process, enabling them to better complete multi-step tasks. Chain of Thought emphasizes the interpretability of language models, making it easier for users to understand and verify their reasoning results.

[0022] Knowledge Graph (KG) is a graph-based knowledge representation used to describe the relationships between entities. It uses nodes (entities) and edges (relationships) to represent entities and the relationships between them.

[0023] A graph database is a database specifically designed to store, manage, and query graph data models (composed of nodes and edges). It efficiently handles complex network structures and is suitable for applications requiring rapid retrieval of related information. Knowledge graphs often use graph databases as their underlying storage mechanism.

[0024] Retrieval Augmented Generation (RAG): is a deep learning architecture that combines information retrieval and text generation, aiming to improve the understanding and generation capabilities of language models in specific tasks.

[0025] The following specific embodiments describe in detail the technical solutions of this application and how the technical solutions of this application solve the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The following detailed description of the technical solutions provided by each embodiment of this application is given in conjunction with the accompanying drawings.

[0026] Figure 1a This is an exemplary application scenario diagram provided by the embodiment of this application. Figure 1a Intelligent agents can include self-learning agents or ticket-checking agents. A self-learning agent is an agent that uses artificial intelligence to learn ticket-checking rules. A ticket-checking agent is an agent that uses artificial intelligence to automatically check ticket checks. Supported by a knowledge graph built on a graph database and a specialized dispatching expertise base, the self-learning agent integrates a generative AI model (which can be an inference model) to build an efficient, intelligent, and self-learning intelligent ticket processing system. Through one or more rounds of interactive feedback with users (such as dispatchers), the self-learning agent forms a closed-loop mechanism of "knowledge learning-inference verification-feedback iteration." This allows it to accumulate new ticket-checking rules and other content into long-term memory, continuously optimize the rule base, and continuously enhance the intelligence level of the ticket-checking system.

[0027] In practical applications, graph databases can provide various operating instructions, various anti-error operation rules, and topological structures (for example, power topological structures). Topological structures reflect the relationships between operating objects and typically represent the connection relationships between operating objects (such as switches, circuit breakers, busbars, transformers, etc.) in the form of a graph. Operating instructions are specific commands issued to achieve specific operating objectives and are the basic unit of an operation ticket. Anti-error operation rules are a set of logical rules used to prevent incorrect operating behaviors during the operation process. Taking power dispatch as an example, these rules are typically formulated based on power grid safety regulations, five-prevention rules, historical accident cases, etc. Based on graph databases, multiple knowledge graphs can be constructed. Knowledge graphs include, but are not limited to: topological structure graphs reflecting the relationships between operating objects, operating instruction graphs, and anti-error operation rule graphs.

[0028] Taking the power dispatching industry as an example, the power topology map reflects the connection relationships and hierarchical structure between various devices in the power grid, including, but not limited to, the electrical connections between substations, lines, transformers, circuit breakers, disconnectors, and other devices. Entities in the power topology map include, but are not limited to, substations, busbars, lines, switches, circuit breakers, and transformers; relationships between entities include, but are not limited to, "connected to," "belongs to," "outgoing," "infeeds in," "powers supplied to," and "parallel operation." Entity attribute information includes, but is not limited to, voltage level, device status (operating / outgoing), and capacity.

[0029] Among them, the "Connect to" relationship type indicates that there is an electrical connection between devices, for example, "5011 switch → Connect to → 110kV busbar". The "Belong to" relationship type indicates the subordinate relationship between devices, for example, "5011 switch → Belong to → Substation A". The "Supply to" relationship type indicates the direction of energy transmission between devices, for example, "110kV line L1 → Supply to → Substation B". The "Parallel operation" relationship type indicates that multiple devices work in parallel, for example, "Main transformer 1 → Parallel operation → Main transformer 2". The "Outgoing line" relationship type indicates the output line from a certain substation, for example, "Substation A → Outgoing line → L1". The "Input / Output" relationship type indicates the voltage conversion relationship of the transformer; for example, "Main transformer high voltage side → Input → 110kV; Main transformer low voltage side → Output → 10kV". The "Feed-in" relationship type indicates that the distribution network supplies power to the load, for example, "10kV feeder F1 → Feed-in → User distribution box".

[0030] In this embodiment, the operation instruction graph is a knowledge graph used to express and manage various operation instructions and their logical relationships; it connects entities such as operation verbs, operation objects, and state descriptions through relationships (such as operation sequence, dependencies, and safety constraints) to form a complex knowledge network that supports intelligent analysis, decision assistance, and other functions.

[0031] Taking the power dispatching industry as an example, the entities in an operation instruction graph include: operation verbs, operation objects, and state descriptions. Examples of operation verbs include, but are not limited to, "close," "open," "engage," "exit," "switch," "test," and "connect ground wire." Operation objects refer to the specific equipment being operated on, including, but not limited to, switches, circuit breakers, lines, busbars, transformers, and protective pressure plates. State descriptions describe the current or expected state of the equipment, including, but not limited to, "operating," "power outage," "disconnected," "opened," and "equipment restored to normal." Relationships between entities include, but are not limited to, operation objects, operation sequences, dependencies, safety constraints, prerequisites, and consequences. The operation object indicates which equipment the operation affects; the operation sequence indicates the order in which operations are performed, for example, "open switch 5011 first, then open circuit break 5011-1." Dependencies indicate whether an operation depends on the completion of other operations, for example, "switching the main transformer can only be performed after all loads have been transferred." Safety constraints define operations that are permitted or prohibited under specific conditions, such as "it is strictly forbidden to open circuit breakers while energized." Preconditions represent the conditions and consequences that must be met before executing an operation; consequences represent the impact or results of executing the operation. Entity attributes include, but are not limited to, execution steps and risk assessment information. Preconditions represent the conditions that must be met before executing an operation, such as "Confirm that the adjacent area has experienced a power outage." Execution steps provide detailed operational instructions, including necessary checkpoints and precautions. Risk assessment information is used to analyze the risks that may arise from the operation and the corresponding control measures.

[0032] For example, the "main transformer power transmission" operation process is represented as follows in the operation instruction map: {[Close]—(operation object)—>[5011 switch]; [Close]—(precondition)—>[5011-1 knife switch has been opened]; [Close]—(consequence)—>[high-voltage side is energized]; [Close]—(operation object)—>[5012 switch]; [Close]—(precondition)—>[5011 switch has been closed]; [Close]—(consequence)—>[low-voltage side is energized]; [Start]—(operation object)—>[No. 1 main transformer protection pressure plate]; [Start]—(precondition)—>[main transformer has been energized and there is no abnormality]; [Start]—(consequence)—>[protection function is enabled]; [Check]—(object)—>[10kV bus voltage]; [Check]—(consequence)—>[confirm that the power supply is normal]}.

[0033] In this embodiment, the anti-misoperation rule graph is a knowledge graph that semantically models the rules for preventing misoperation. It can express which operations cannot be performed, under what conditions they are prohibited, what consequences will be caused by illegal operations, and what control measures should be taken to avoid misoperation.

[0034] Taking the power dispatching industry as an example, the entities of the anti-misoperation rule map include but are not limited to: anti-misoperation rule name, triggering conditions, risk consequences, control measures, operation objects, etc.

[0035] The name of the anti-incorrect operation rule is used to identify a specific anti-incorrect operation rule. For example, the anti-incorrect operation rule includes but is not limited to: preventing the live switch from being pulled, preventing the live compartment from being entered by mistake, preventing the grounding switch from being closed by mistake, preventing the protection pressure plate from being placed by mistake, etc.

[0036] A trigger condition is a condition that causes a specific misoperation prevention rule to take effect. Identifying the trigger condition is crucial for determining whether a specific operation is permitted. Examples of trigger conditions include, but are not limited to: the line is energized, the switch is not disconnected, maintenance is not completed, electrical testing has not been performed, and the equipment is in operation. For example, "[Prevent Live Switch Pulling] - Trigger Condition → [Line is Energized]".

[0037] Risk consequences describe the potential consequences or impacts of violating the corresponding error prevention rules. This helps assess the potential risk level and take appropriate preventative measures. Examples of risk consequences include, but are not limited to, short-circuit tripping, equipment damage, personal injury, malfunctioning protection devices, or regional power outages.

[0038] Control measures are specific measures or procedures designed to prevent incorrect operation. These measures typically require manual confirmation or adherence to specific operating procedures. Examples of control measures include, but are not limited to, requiring a switch to be disconnected beforehand, requiring electrical testing and recording, requiring safety fencing, displaying warning signs, or requiring confirmation by a responsible person.

[0039] An operation object refers to the specific equipment or components involved in performing an operation. Examples of operation objects include, but are not limited to, circuit breakers, grounding switches, switches, busbars, transformers, and protective pressure plates. Relationships between entities in the error prevention rule map include, but are not limited to, trigger conditions, causes, control measures, and belongs to. The "cause" relationship reflects the potential consequences of an illegal operation; the "belongs to" relationship indicates which error prevention system a given error prevention rule belongs to.

[0040] For example, the knowledge information in the anti-misoperation rule map is as follows: {[Prevent pulling the knife switch while it is energized]—(triggering condition)—>[The line is energized]; [Prevent pulling the knife switch while it is energized]—(result)—>[Short-circuit tripping]; [Prevent pulling the knife switch while it is energized]—(control measures)—>[The switch must be disconnected first]; [Prevent accidental closing of the grounding knife switch]—(triggering condition)—>[The line is energized]; [Prevent accidental closing of the grounding knife switch]—(result)—>[Short-circuit fault]; [Prevent accidental closing of the grounding knife switch]—(control measures)—>[The electricity must be tested and recorded first]}.

[0041] In practice, a dispatching expertise database refers to a systematic collection of information established to support efficient dispatching within a specific field (such as power systems, petrochemicals, and rail transit). This expertise database includes, but is not limited to, operational procedures, expert knowledge, and a rule base.

[0042] Operating procedures and standards generally refer to a set of operational guidelines, standards, and regulations established within a specific industry or work environment to ensure the safety, quality, and efficiency of work processes. These procedures and standards may be developed internally by a company or issued by an industry association, national standards body, or international organization to guide employees on how to perform their tasks correctly and safely.

[0043] In practice, each industry has its own operating procedures and specifications. For example, in the power dispatching industry, these procedures include, but are not limited to, power system dispatching procedures, power dispatching operating procedures, power grid accident handling procedures, power system stable operation procedures, relay protection and automatic device operation procedures, power grid dispatching automation system operation procedures, and new energy grid access dispatching procedures.

[0044] The Power System Dispatch Regulations define the power system's dispatch management principles, the responsibilities of dispatch agencies at all levels, and their coordination mechanisms. They include provisions for power system operation arrangements, load forecasting and management, generation planning, and frequency and voltage regulation.

[0045] The power dispatching operating procedures detail the code of conduct for power dispatchers in their daily operations, including specific steps such as operating switchgear and switching lines. They emphasize the safety checks and simulations required before any operation, as well as the principles to be followed during operations, such as the "three comparisons" and "three prohibitions."

[0046] Power grid incident handling procedures guide dispatchers on how to quickly and effectively respond to unexpected incidents or emergencies in the power grid, including processes such as fault identification, isolation, and power restoration. The goal is to minimize the scope of power outages and restore normal power supply as quickly as possible.

[0047] The power system stable operation regulations are formulated to meet the specific requirements for maintaining power system stability, including but not limited to monitoring and adjustment measures in terms of voltage stability and frequency stability.

[0048] The operating procedures for relay protection and automatic devices cover the configuration principles, setting calculations, commissioning and maintenance of relay protection devices and automatic control systems (such as automatic reclosing devices), aiming to protect the power system from faults and quickly restore the system to normal operation.

[0049] The grid dispatching automation system operation procedures describe the relevant technical standards and work processes for realizing grid dispatching automation using modern information technology, including the application and maintenance of energy management systems (EMS) and distribution management systems (DMS).

[0050] With the development of renewable energy, technical requirements and dispatching strategies for the access of new energy sources such as wind power and photovoltaic power to the grid have been specially formulated to ensure the coordinated development of new energy and traditional energy.

[0051] In practice, the expert knowledge within the dispatching expertise base varies across industries. Expert knowledge refers to the accumulated and summarized experience within a specific field by professionals with a solid theoretical foundation and extensive practical experience. In the power dispatching industry, expert knowledge is a valuable resource accumulated through long-term practice. It encompasses not only technical knowledge but also the ability to judge complex situations, experience in handling emergencies, and methods for optimizing grid operations.

[0052] In this embodiment, the rule base mainly provides a plurality of standardized operation ticket review rules and related information of each operation ticket review rule. The operation ticket review rule is a criterion used to ensure that the operation ticket meets all necessary security and operation requirements.

[0053] It should be noted that Figure 1a The application scenario shown is only an example, and the embodiments of the present application do not limit the application scenario.

[0054] Figure 1b This is a flowchart of a self-learning method for operation ticket review rules provided in an embodiment of the present application. This method can be applied to self-learning agents. Figure 1b , the method may include the following steps: 101. Obtain the current dialogue information input by the user in this dialogue interaction; 102. Use the generative AI model to identify the intent of the current conversation information and obtain the intent category of the current conversation information; 103. If the intent category of the current conversation information is to add a new operation ticket review rule, the generative AI model is used in combination with the knowledge graph and the scheduling professional knowledge base to parse the new operation ticket review rule in the current conversation information to obtain the rule analysis content of the new operation ticket review rule.

[0055] Optionally, the rule parsing content includes: operation instruction review type, operation instruction screening thinking chain and operation instruction review thinking chain; the operation instruction review type is used to indicate independent review of a single operation instruction or joint review of multiple operation instructions; the operation instruction screening thinking chain is used to indicate a screening method for screening out at least one target operation instruction to be reviewed from at least one operation instruction included in the operation ticket to be reviewed; the operation instruction review thinking chain is used to indicate the review process of the newly added operation ticket review rules.

[0056] Optionally, the above also includes: in response to the user confirming that the review result of the operation ticket to be reviewed is incorrect, obtaining the next dialogue information input by the user in the next dialogue interaction; wherein, the review result of the operation ticket to be reviewed is obtained by reviewing the operation instructions in the operation ticket to be reviewed according to the rule parsing content; using the generative AI model to identify the intent of the next dialogue information; if it is identified that the intention category of the next dialogue information is to update the rule parsing content of the new operation ticket review rules, then using the generative AI model to update the rule parsing content of the new operation ticket review rules according to the next dialogue information.

[0057] Optionally, the rule parsing content of the newly added operation ticket review rules is updated according to the next dialogue information, including: if the intention category of the next dialogue information is to supplement example information, then example information is added to the rule parsing content of the newly added operation ticket review rules, and the example information includes positive example information and / or negative example information, the positive example information includes the operation instruction example and its review result that passes the operation ticket review rule example review; the negative example information includes the operation instruction example and its review result that fails the operation ticket review rule example review; if the intention category of the next dialogue information is to supplement background knowledge, then background knowledge is added to the rule parsing content of the newly added operation ticket review rules, and background knowledge refers to the knowledge information required to assist in using the newly added operation ticket review rules to review the operation instructions of the operation ticket to be reviewed; if the intention category of the next dialogue information is to modify the operation instruction screening thinking chain, the operation instruction screening thinking chain in the rule parsing content of the newly added operation ticket review rules is modified according to the next dialogue information; if the intention category of the next dialogue information is to modify the operation instruction review thinking chain, the operation instruction review thinking chain in the rule parsing content of the newly added operation ticket review rules is modified according to the next dialogue information.

[0058] Optionally, the above method also includes: if the intention category of the current dialogue information is not identified, outputting guidance information, the guidance information is used to guide the user to enter dialogue information related to the operation ticket review; if the intention category of the current dialogue information is identified as an intention category unrelated to the operation ticket review, outputting a fallback script.

[0059] For details on how to implement the self-learning agent to automatically learn the operation ticket review rules, please refer to the subsequent content.

[0060] The technical solution provided by the embodiment of the present application is that the user conducts a dialogue interaction with the self-learning intelligent agent, and the self-learning intelligent agent uses a generative AI model to identify the intent of the current dialogue information input by the user in the dialogue interaction. If the intent category of the current dialogue information is identified as a new operation ticket review rule, the generative AI model is used in combination with the knowledge graph and the scheduling professional knowledge base to parse the new operation ticket review rule in the current dialogue information to obtain the rule parsing content of the new operation ticket review rule. In this way, the self-learning of the new operation ticket review rule is completed, and an efficient, intelligent and self-learning operation ticket review rule self-learning solution is provided. In this way, in the subsequent operation ticket review stage, the self-learned operation ticket review rules can be used to perform efficient and accurate automatic review of the operation ticket, providing an efficient, intelligent and self-learning operation ticket review solution that can effectively ensure the security, standardization and correctness of the operation ticket.

[0061] Figure 2 This is a flow chart of an operation ticket review method provided in an embodiment of the present application. Figure 2 , the method may include the following steps: 201. Obtain current dialogue information input by the user in this dialogue interaction.

[0062] 202. Use the generative AI model to identify the intent of the current conversation information. If the intent category of the current conversation information is identified as adding new operation ticket review rules, use the generative AI model in combination with the knowledge graph and the scheduling professional knowledge base to parse the new operation ticket review rules in the current conversation information to obtain the rule analysis content of the new operation ticket review rules.

[0063] 203. Use the generative AI model to parse the content according to the rules to review the operation instructions in the operation ticket to be reviewed, and output the review results of the operation ticket to be reviewed.

[0064] In actual applications, steps 201 to 203 can be executed by a self-learning agent, and step 203 can be executed by an operation ticket review agent.

[0065] In practice, a complete operation ticket typically includes the following information: operation ticket number, operation task, operation time, operator, or a list of operation steps. The operation step list includes multiple ordered operation steps, and the step information typically includes: step number (indicating the operation sequence), operation instruction, device name (used to specify the operation target), and operation status (e.g., pending / in progress / completed).

[0066] For example, the pending operation tickets for the power dispatch scenario are as follows: { Operation Ticket Number: OPE-20250604-014 Operation task: L1 line is switched from cold standby state to operation and performs string supply operation Operation time: 2025-06-04 09:00~11:00 Operator: Zhang San; Guardian: Li Si List of steps: Operation step 1: {Operation instruction: Close the isolating switches on both sides of the L1 line; Equipment name: Isolating switches on both sides of the L1 line; Operation status: To be executed}; Operation step 2: {Operation instruction: Check if L1 line has no voltage; Equipment name: L1 line; Operation status: To be executed}; Operation step 3: {Operation instruction: Close L1 circuit breaker 101; Device name: L1 circuit breaker 101; Operation status: To be executed}; Operation step 4: {Operation instruction: L1 circuit breaker 101 turns on and supplies power in series; Device name: L1 circuit breaker 101; Operation status: To be executed}; }.

[0067] To ensure the security and compliance of the operational process, it's necessary to utilize operation ticket review rules to audit the operational instructions within the operation ticket. This ensures the security and compliance of the entire operational process. These rules can review the operational instructions within the operation ticket from one or more dimensions, such as compliance, completeness, or correctness. As you can see, the more dimensions included, the greater the security and compliance of the operation ticket.

[0068] Standardization inspections primarily focus on verifying that the language, terminology, and equipment naming of the operating instructions in the operation ticket comply with relevant operating procedures. By standardizing language and naming, human misunderstandings can be reduced, improving operational safety and compliance.

[0069] Optionally, the standardization check may include one or more of the following check items: terminology consistency check item, double naming check item, ambiguous terminology check item, and format uniformity check item.

[0070] The terminology consistency check is mainly used to check whether standard dispatching terms (such as "pull open", "close", and "disconnect") are used. For example, the operation instruction "open the knife switch" should be written as "pull open the knife switch".

[0071] The double naming check item is mainly used to check whether the device name is in the form of a number and device type. For example, "switch" in the operating instruction should be written as "101 switch".

[0072] The fuzzy wording check item is primarily used to check for uncertain or ambiguous descriptions (such as "maybe," "probably," and "should") in operating instructions. For example, the operating instruction "Open the grounding switch" → "Open the #1 main transformer grounding switch" should be written as "Open the #1 main transformer grounding switch."

[0073] The format consistency check item primarily checks whether each instruction follows a consistent format. For example, the instruction "Confirm the status after disconnecting switch 101" should be written in two steps: "Disconnect switch 101" and "Confirm that switch 101 is disconnected."

[0074] Among them, the integrity check mainly checks whether there are any omissions in the operation ticket or operation instruction, and determines whether the operation steps that must be performed in the operating procedures and specifications have been omitted. Examples of omissions include but are not limited to: failure to open the grounding knife switch, failure to activate the protective device, or failure to check the switch status. Among them, if there is an omission of "failure to open the grounding knife switch", if the grounding knife switch is not opened before the online route maintenance is transferred to operation, there is a risk of closing the circuit breaker with the ground wire. If there is an omission of "failure to activate the protective device", if the "short lead protection" is not activated before the power outage, it will lead to operation in an unprotected state. If there is an omission of "failure to check the switch status", if the "101 switch is disconnected" is not checked before operating the disconnector, it may lead to the opening of the knife switch with load.

[0075] The correctness check is mainly used to check whether the operation sequence of each operation instruction or operation step is reasonable. For example, opening the grounding switch first and then closing the switches on both sides of the switch is a reasonable operation sequence.

[0076] In this embodiment, users can add new operation ticket review rules that are not yet included in the rule base (referred to as newly added operation ticket review rules). If the newly added operation ticket review rule meets the user's expectations for the operation ticket review result, the newly added operation ticket review rule can be added to the rule base for subsequent regular review. For ease of distinction, the operation ticket review rules that have already been added to the rule base are referred to as existing operation ticket review rules.

[0077] In actual applications, operation tickets that need to be reviewed are called pending operation tickets. When a user interacts with a self-learning intelligent agent through a dialogue, the user can enter the current dialogue information in natural language on the dialogue interaction interface. Due to different user needs, the intent category of the current dialogue information is different. For example, the intent category of the current dialogue information is "new operation ticket review rules", "supplementary content" or "irrelevant intent". Among them, when the intent category of the current dialogue information is "supplementary content", it mainly reflects that the user has a need for supplementary explanation of the new operation ticket review rules. When the intent category of the current dialogue information is "intention not related to operation ticket review", it mainly reflects that the current dialogue information entered by the user is not related to operation ticket review. The current dialogue information may be, for example, casual chat information, general questions and answers, etc.

[0078] In some optional embodiments, the current conversation message may include an intent category tag, which can be a keyword identifying the intent category to which the current conversation message belongs. For example, the intent category tags may be "new rules" or "supplementary content." By including the intent category tag in the current conversation message, the self-learning agent can efficiently and accurately identify the intent category to which the current conversation message belongs, thereby improving the efficiency and quality of operation ticket review.

[0079] For example, the current dialogue information is "New rule: When a line with a line knife switch is out of power, short lead protection should be activated before the switch is switched to parallel operation." The "New rule" in the current dialogue information indicates that the intent category of the current dialogue information is to add a new operation ticket review rule.

[0080] For another example, the user enters the following natural language information on the dialogue interaction interface: "Supplementary content: This rule only applies to power outage or power supply operations on lines with line switches, and the operation content is "the line is switched from cold standby status to operation and serial supply." Operation instructions that do not meet the above conditions do not need to be reviewed." The "supplementary content" in the current dialogue information indicates that the intent category of the current dialogue information is to supplement the newly added operation ticket review rules.

[0081] For another example, the user enters natural language information on the dialogue interaction interface as: "Irrelevant content: How is the weather today?" The intent category of the "irrelevant content" in the current dialogue information is "intent not related to the operation ticket review."

[0082] In this embodiment, the self-learning agent obtains the current dialogue information input by the user in this dialogue interaction and performs intent recognition on the current dialogue information.

[0083] Specifically, the generative AI model leverages its powerful semantic understanding and reasoning capabilities to deeply understand the current conversation information input by the user. It can also perform semantic reasoning based on the conversation context to accurately identify the intent category of the current conversation information. Conversation context can be understood as the conversation content or other information recorded in previous conversation interactions prior to the current one. Conversation context can be considered short-term memory information that assists the generative AI model in its decision-making. Furthermore, the generative AI model identifies the intent of the current conversation information based on the intent category tags in the current conversation information.

[0084] Optionally, the knowledge graph includes at least one of the following: a topological structure graph reflecting the association relationship between operation objects, an operation instruction graph, and an anti-misoperation rule graph; the scheduling professional knowledge base includes at least one of the following: operating procedure specifications, expert experience knowledge, and a rule base.

[0085] In practice, if the generative AI model fails to identify the intent category of the current conversation, it outputs guidance information to guide the user to enter conversation information related to the operation ticket review. This guides the user to enter content related to the operation ticket review task in a natural and friendly manner, thereby ensuring the effectiveness of the interaction and the continuity of the process, and preventing users from straying from the core operation ticket review scenario for extended periods of time. For example, if the user enters the current conversation information "Add a rule," the generative AI model outputs guidance information such as "Please describe the rule you want to add, for example, the differential protection should be deactivated before the main transformer is powered off." Another example is, if the user enters the current conversation information "See if there are any problems with this," the generative AI model outputs guidance information such as "Which rule or operation ticket are you referring to? Can you provide more details?"

[0086] In practice, if the generative AI model identifies the intent of the current conversation as unrelated to ticket review, it outputs a fallback message to guide the user back to the ticket review scenario. For example, if the user enters "What's the weather like today?", the fallback message might be "I'm a self-learning assistant. My current task is to add new ticket review rules and self-learn those rules. I'm sorry I can't answer your question."

[0087] In actual applications, if the generative AI model identifies that the intent category of the current conversation information is the addition of new operation ticket review rules, the generative AI model will combine the knowledge graph and the scheduling professional knowledge base to parse the new operation ticket review rules in the current conversation information to obtain the rule analysis content of the new operation ticket review rules.

[0088] Specifically, the rule parsing content of the new operation ticket review rules can be understood as the structured rule description obtained after parsing the new operation ticket review rules described in natural language input by the user.

[0089] In some optional embodiments, in order to improve the efficiency and accuracy of operation ticket review, the rule parsing content may include at least one of the following information: operation instruction review type, operation instruction screening thought chain and operation instruction review thought chain.

[0090] In some optional embodiments, for the instruction screening thought chain and the operation instruction review thought chain, if the generative AI model has a deep thinking function, the generative AI model can be regarded as an inference model, and the generative AI model under the deep thinking function can be used to parse the instruction screening thought chain and the operation instruction review thought chain. If the generative AI model does not have a deep thinking function, the generative AI model can also directly parse the instruction screening thought chain and the operation instruction review thought chain. Of course, the generative AI model under the deep thinking function can more accurately parse the instruction screening thought chain and the operation instruction review thought chain.

[0091] In practical applications, prompt engineering can be used to design appropriate prompts, which can then be used to guide generative AI models in parsing newly added operation ticket review rules. Prompts can guide the generative AI model in analyzing various aspects of the newly added operation ticket review rules, including the operation instruction review type, the operation instruction screening thought chain, and the operation instruction review thought chain.

[0092] To ensure the effectiveness of the prompt word information, it is necessary to supplement the prompt word information with some knowledge information from the knowledge graph and the scheduling professional knowledge base. In actual applications, the generative AI model analyzes the newly added operation ticket review rules to determine preliminary rule analysis results, such as the operation instructions involved in the newly added operation ticket review rules and the association between the operation instructions; based on the preliminary rule analysis results, the knowledge graph interface is used to query relevant knowledge information, including equipment topology information; based on the preliminary rule analysis results, the retrieval augmentation generation (RAG) technology is used to recall relevant knowledge information from the scheduling professional knowledge base, and the recalled relevant knowledge information is summarized to obtain the recalled summary knowledge information; the prompt word information is supplemented with background knowledge such as the knowledge information retrieved from the knowledge graph and the recalled summary knowledge information from the scheduling professional knowledge base; in this way, under the guidance of this background knowledge, the generative AI model can more accurately perform structured analysis of the newly added operation ticket review rules.

[0093] For example, a new operation ticket review rule, "A power test must be performed before closing the circuit breaker," was added. The background knowledge captured from the knowledge graph includes the following: 1. Anti-misoperation rule: "Prevent pulling the circuit breaker while energized." The trigger condition is that the line is energized, and the control measure requires a power test beforehand; 2. Operation verbs: "Test power," "Close"; 3. Equipment information: "Circuit breaker status," "Bus voltage level." The regulatory clause, captured from the dispatching expertise database, summarizes Article X of the "Electric Power Dispatching Operating Procedures," which states: "Closing the circuit breaker without a power test is strictly prohibited." The generative AI model combines this background knowledge to analyze the rule.

[0094] In this embodiment, the operation instruction review type is used to indicate whether to conduct an independent review of a single operation instruction or a joint review of multiple operation instructions.

[0095] Specifically, if the Operation Instruction Review Type indicates that a single operation instruction should be reviewed independently, this indicates that the newly added Operation Ticket Review Rule has a single review target, focusing only on the single operation instruction itself. Consequently, each operation instruction is reviewed separately using the newly added Operation Ticket Review Rule. For example, if there are five operation instructions that require review, each will be reviewed separately using the newly added Operation Ticket Review Rule. Taking the newly added Operation Ticket Review Rule of "Drafting an Operation Instruction Ticket should ensure clear tasks and legible text, and correctly use dual equipment nomenclature and dispatch terminology" as an example, it is only necessary to check whether a single operation instruction correctly uses dual equipment nomenclature and dispatch terminology.

[0096] If the Operation Instruction Review Type indicates a joint review of multiple Operation Instructions, this indicates that the newly added Operation Ticket Review Rules will diversify their scope, focusing on multiple Operation Instructions and the order, dependencies, or logical relationships between them. Furthermore, the newly added Operation Ticket Review Rules will be used to review the combined actions of two or more Operation Instructions within the Operation Ticket. For example, if the newly added Operation Ticket Review Rule states "Open Isolator A first, then Close Isolator B; Reverse Operation is Prohibited," multiple Operation Instructions will need to be reviewed simultaneously.

[0097] In this embodiment, the operation instruction screening thinking chain is used to indicate a screening method for screening at least one target operation instruction to be reviewed from at least one operation instruction included in the operation ticket to be reviewed. Specifically, the operation instruction screening thinking chain is a thinking chain used to screen operation instructions. For example, the newly added operation ticket review rule is "the power must be tested before the circuit breaker is closed", and the operation instruction screening thinking chain is "identifying and extracting all operation instructions involving "circuit breaker closing" from the operation ticket for subsequent review and judgment of "whether to test the power". For another example, the newly added operation ticket review rule is "first open the disconnector A, then close the disconnector B, and reverse operation is prohibited", and the operation instruction screening thinking chain is "screening out all "open" operations involving "disconnector A" and "close" operations involving "disconnector B" from the operation ticket. This process ensures that only operation instructions related to the newly added operation ticket review rules are selected for subsequent joint review of sequence and compliance.

[0098] In this embodiment, the operation instruction review thinking chain is used to indicate the review process of the newly added operation ticket review rules. Specifically, the operation instruction review thinking chain is a thinking chain used to indicate the review process of the newly added operation ticket review rules. For example, the newly added operation ticket review rule is "the power must be tested before the circuit breaker is closed", and the operation instruction review thinking chain is "If the current operation instruction is the circuit breaker closing operation, step 1: find out whether there is a corresponding power test operation before the circuit breaker closing operation; step 2: determine whether the power test operation occurs before the closing operation; step 3: verify whether the interval between the power test time and the closing time is within the allowable range (such as within 5 minutes); step 4: If any of the above conditions are not met, the operation instruction is judged to be non-compliant (that is, the audit result is failed to pass the audit). If all of the above conditions are met, the operation instruction is judged to be compliant (that is, the audit result is passed). ". For example, the newly added operation ticket review rule is "open the disconnector A first, then close the disconnector B, and the reverse operation is prohibited", and the operation instruction review thinking chain is "Step 1: Confirm whether there is an operation instruction to open the disconnector A in the operation ticket; Step 2: Confirm whether there is an operation instruction to close the disconnector B in the operation ticket; Step 3: Verify the operation sequence to ensure that the opening operation of the disconnector A occurs before the closing operation of the disconnector B; Step 4: If any of the above conditions is not met, multiple operation instructions are judged to be non-compliant (that is, the review result is failed). If all of the above conditions are met, multiple operation instructions are judged to be compliant (that is, the review result is passed)."

[0099] In this embodiment, after the self-learning intelligent agent uses the generative AI model to complete the analysis of the new operation ticket review rules, it provides the rule analysis content of the new operation ticket review rules to the inference model. The inference model is combined with the knowledge graph and the scheduling professional knowledge base to review the operation instructions in the operation ticket to be reviewed according to the rule analysis content, and output the review results of the operation ticket to be reviewed.

[0100] In some scenarios, the self-learning agent can provide the rule analysis content of the newly added operation ticket review rules to the operation ticket review agent. The operation ticket review agent uses the generative AI model to review the operation instructions in the operation ticket to be reviewed according to the rule analysis content, and output the review results of the operation ticket to be reviewed.

[0101] In practice, the review results for pending operation tickets can include: approved or rejected. Approved indicates that the pending operation ticket has passed the newly added operation ticket review rules and is considered a qualified operation ticket. Failed indicates that the pending operation ticket has failed the newly added operation ticket review rules and is considered an unqualified operation ticket. The review results can also include an analysis of the reasons for approval or rejection, detailed information about the review process, and more.

[0102] In practical applications, the review results for pending operation tickets may include the review results for each target operation instruction. Target operation instructions are the operation instructions that are screened out from the pending operation tickets and require review. If the review result for a target operation instruction is "approved," it indicates that the target operation instruction has passed the newly added operation ticket review rules and is considered qualified. If the review result for a target operation instruction is "failed," it indicates that the target operation instruction has failed the newly added operation ticket review rules and is considered unqualified.

[0103] In practice, the audit results for pending operation tickets can be aggregated based on the audit results of each target operation instruction. If all target operation instructions are approved, the pending operation ticket will be deemed approved. If any target operation instruction is rejected, the pending operation ticket will be deemed rejected.

[0104] In some optional embodiments, in order to further improve the security and standardization of operation tickets, the reasoning model reviews the operation instructions in the operation ticket to be reviewed from multiple dimensions according to the rule analysis content; the multiple dimensions include: integrity check, standardization check or correctness check.

[0105] In some optional embodiments, in order to further improve the security and standardization of operation tickets, the implementation method of using a generative AI model to parse the content according to rules to review the operation instructions in the operation ticket to be reviewed is: using a generative AI model to screen out at least one target operation instruction to be reviewed from at least one operation instruction included in the operation ticket to be reviewed according to the operation instruction screening thinking chain; if the operation instruction review type indicates that a single operation instruction is to be reviewed independently, then the single target operation instruction is to be reviewed independently according to the operation instruction review thinking chain to obtain the review result of the single target operation instruction; and the review result of the operation ticket to be reviewed is generated based on the review result of at least one target operation instruction; if the operation instruction review type indicates that multiple operation instructions are to be reviewed jointly, then the multiple target operation instructions are to be reviewed jointly according to the operation instruction review thinking chain to obtain the review results of multiple target operation instructions; and the review result of the operation ticket to be reviewed is generated based on the review results of multiple target operation instructions.

[0106] The technical solution provided by the embodiment of the present application is to conduct a dialogue interaction with the user and use a generative AI model to identify the intent of the current dialogue information input by the user in the dialogue interaction. If the intent category of the current dialogue information is identified as a new operation ticket review rule, the new operation ticket review rule in the current dialogue information is parsed in combination with the knowledge graph and the scheduling professional knowledge base to obtain the rule parsing content of the new operation ticket review rule; the generative AI model is used to review the operation instructions in the operation ticket to be reviewed according to the rule parsing content, and the review result of the operation ticket to be reviewed is output. Thus, an efficient, intelligent and self-learning operation ticket review solution is provided, which can effectively ensure the security, standardization and correctness of the operation ticket.

[0107] Figure 3 This is a flowchart of another operation ticket review method provided in the embodiment of this application. Figure 3 , the method may include the following steps: 301. Obtain current dialogue information input by the user in this dialogue interaction.

[0108] 302. Use the generative AI model to identify the intent of the current conversation information. If the intent category of the current conversation information is identified as adding new operation ticket review rules, use the generative AI model in combination with the knowledge graph and the scheduling professional knowledge base to parse the new operation ticket review rules in the current conversation information to obtain the rule analysis content of the new operation ticket review rules.

[0109] 303. Use the generative AI model to review the operation instructions in the operation ticket to be reviewed according to the rule analysis content, and output the review result of the operation ticket to be reviewed, and execute step 304 or step 307.

[0110] Steps 301-302 can be executed by a self-learning agent, and step 303 can be executed by an operation ticket review agent.

[0111] In actual applications, after the operation ticket review agent outputs the review results of the operation ticket to be reviewed, the user can confirm whether the review results of the operation ticket to be reviewed meet expectations. If the user confirms that the review results of the operation ticket to be reviewed meet expectations, it can be considered that the review results of the operation ticket to be reviewed are correct, indicating that the review results of the operation ticket to be reviewed are credible. If the user confirms that the review results of the operation ticket to be reviewed do not meet expectations, it can be considered that the review results of the operation ticket to be reviewed are incorrect, indicating that the review results of the operation ticket to be reviewed are unreliable. Based on user feedback, the rule parsing content of the newly added operation ticket review rules can be continuously optimized, thereby improving the accuracy and adaptability of subsequent reviews.

[0112] 304. In response to the user confirming that the review result of the pending operation ticket is wrong, obtaining the next dialogue information input by the user in the next dialogue interaction.

[0113] 305. Use the generative AI model to identify the intent of the next conversation information. If the intent category of the next conversation information is identified as updating the rule parsing content of the newly added operation ticket review rules, then update the rule parsing content of the newly added operation ticket review rules based on the next conversation information.

[0114] Optionally, a generative AI model is used in combination with a knowledge graph and a scheduling expertise base to update the rule parsing content of the newly added operation ticket review rules based on the next conversation information.

[0115] 306. Use the generative AI model to parse the content according to the updated rules to review the operation instructions in the operation ticket to be reviewed, and output the review results of the operation ticket to be reviewed.

[0116] Steps 304-305 can be executed by a self-learning agent, and step 306 can be executed by an operation ticket review agent.

[0117] In this embodiment, if the user confirms that the review result of the pending operation ticket is incorrect, the user can continue to interact with the self-learning agent, and the self-learning agent obtains the next dialogue information entered by the user in the next dialogue interaction. The next dialogue interaction is the dialogue interaction after the current dialogue interaction, and the next dialogue information is the dialogue information entered by the user in the next dialogue interaction.

[0118] In actual applications, users enter the next conversation information as needed. Different next conversation information will update the rule parsing content of the newly added operation ticket review rules differently, thereby achieving continuous optimization and personalized adaptation of the rule parsing content.

[0119] In some optional embodiments, the implementation method for updating the rule parsing content of the newly added operation ticket review rules based on the next conversation information is: if the intention category of the next conversation information is to supplement example information, then example information is added to the rule parsing content of the newly added operation ticket review rules, and the example information includes positive example information and / or negative example information. The positive example information includes the operation instruction example and its review result of passing the operation ticket review rule example review; the negative example information includes the operation instruction example and its review result of failing to pass the operation ticket review rule example review.

[0120] Specifically, examples of operational instructions can be understood as illustrative operational instructions, and examples of operational ticket review rules can be understood as illustrative operational ticket review rules. These examples are specific operational instructions in operational tickets that have been reviewed using these examples. By introducing both positive and negative examples, generative AI models and inference models can better understand the newly added operational ticket review rules, improving the accuracy and efficiency of operational ticket reviews.

[0121] In some optional embodiments, the implementation method for updating the rule parsing content of the newly added operation ticket review rules based on the next conversation information is: if the intention category of the next conversation information is to supplement background knowledge, then background knowledge is added to the rule parsing content of the newly added operation ticket review rules. Background knowledge refers to the knowledge information required to assist in using the newly added operation ticket review rules to review the operation instructions of the operation ticket to be reviewed.

[0122] Specifically, supplementing background knowledge can help generative AI models and reasoning models better understand the newly added operation ticket review rules and improve the accuracy and efficiency of operation ticket review.

[0123] In some optional embodiments, the implementation method for updating the rule parsing content of the newly added operation ticket review rules based on the next conversation information is: if the intention category of the next conversation information is to modify the operation instruction filtering thought chain, the operation instruction filtering thought chain in the rule parsing content of the newly added operation ticket review rules is modified according to the next conversation information.

[0124] Specifically, it supports users to dynamically update the operation instruction screening thinking chain, so as to more accurately identify the operation instructions that should be reviewed by the newly added operation ticket review rules, and realize the personalized adaptation and continuous optimization of the rule parsing content.

[0125] In some optional embodiments, the implementation method of updating the rule parsing content of the newly added operation ticket review rules based on the next conversation information is: if the intention category of the next conversation information is to modify the operation instruction review thinking chain, the operation instruction review thinking chain in the rule parsing content of the newly added operation ticket review rules is modified according to the next conversation information.

[0126] Specifically, it supports users to dynamically update the operation instruction review thinking chain, so as to more accurately review the operation tickets and achieve personalized adaptation and continuous optimization of the rule parsing content.

[0127] In actual applications, after each round of dialogue interaction, the operation ticket review agent outputs the review result of the operation ticket to be reviewed. The user can then confirm whether the review result of the operation ticket to be reviewed meets expectations. If not, the process returns to steps 304 to 306 until it is confirmed that the review result of the operation ticket to be reviewed meets expectations.

[0128] 307. In response to the user confirming that the audit result of the operation ticket to be audited is correct, the newly added operation ticket audit rule and the correct rule parsing content are associated and stored in the rule library.

[0129] Among them, step 307 can be performed by a self-learning agent.

[0130] In actual applications, after each round of dialogue interaction, the operation ticket review agent outputs the review results of the operation ticket to be reviewed. The user can then confirm whether the review results of the operation ticket to be reviewed meet expectations. If they meet expectations, the newly added operation ticket review rules and their expected rule parsing content can be associated and stored in the rule library, and then the newly added operation ticket review rules and their expected rule parsing content can be deposited into long-term memory to continuously optimize the rule library. In the daily management stage, the operation ticket review rules in the rule library can be used to automatically review the operation tickets. The rule parsing content that meets expectations can be considered as the correct rule parsing content confirmed by the user; the rule parsing content that does not meet expectations can be considered as the incorrect rule parsing content confirmed by the user.

[0131] In practical applications, the rule base can include at least one existing operation ticket review rule and the rule parsing content for each existing operation ticket review rule. Optionally, the rule base can also include approved operation tickets for each existing operation ticket review rule. Approved operation tickets are those that have been reviewed using existing operation ticket review rules. This way, when newly added operation ticket review rules are stored in long-term memory, the newly added operation ticket review rules, their expected rule parsing content, and the pending operation tickets can be stored in the rule base.

[0132] It is worth noting that the audited operation tickets are actual cases that have been audited based on the existing operation ticket audit rules. They can be used to assist in rule understanding and continuous optimization, and to improve the intelligence level of self-learning agents.

[0133] The technical solution provided by the embodiment of this application is that the intelligent agent interacts with the user through dialogue, and uses generative AI models and reasoning models combined with knowledge graphs and scheduling professional knowledge bases to provide an efficient, intelligent, and self-learning operation ticket review solution that can effectively ensure the safety, standardization, and correctness of operation tickets. Through one or more rounds of interactive feedback with the user, a closed-loop mechanism of "knowledge learning-reasoning verification-feedback iteration" is formed, realizing the dynamic update of rules enhanced by human-computer collaboration, and continuously improving the intelligent level of operation ticket review.

[0134] For ease of understanding, the following Figure 4 Introducing a specific scenario implementation. Figure 4 ,First, the self-learning agent obtains the current dialogue ,information input by the user in the dialogue interaction ,interface, and performs intent recognition on the current dialogue ,information.

[0135] If the current conversation doesn't clearly express the user's intent, the intent recognition result is a follow-up question, which can be understood as not recognizing the intent category of the current conversation. At this point, the self-learning agent outputs guidance information on the dialogue interface. This guidance information guides the user to enter conversation information related to the operation ticket review, essentially guiding the user back to the main task (i.e., the operation ticket review task).

[0136] If the intent recognition result is irrelevant, the self-learning agent will output a fallback message in the dialogue interface. If the intent recognition result is the main task, the self-learning agent can determine whether the main task's intention is to add new rules or supplementary content.

[0137] If the main task is to add new rules, the self-learning agent parses the newly added operation ticket review rules and obtains the parsed content of the newly added operation ticket review rules. If the main task is to supplement content, the self-learning agent updates the parsed content. The operation ticket review agent uses the final parsed content to review the operation ticket and outputs the review results on the dialogue interface.

[0138] If the user confirms that the review result of the operation ticket does not meet expectations, the user can continue to enter supplementary content for the newly added operation ticket review rules. The self-learning intelligent agent uses the supplementary content to update the rule parsing content, and continues to use the updated rule parsing content to review the operation ticket and output the review result of the operation ticket on the dialogue interaction interface for the user to confirm whether it meets expectations.

[0139] If the user confirms that the review result of the operation ticket meets expectations, he can execute the rule storage operation, that is, save the newly added operation ticket review rules and their rule analysis content to the rule library, so that the operation ticket can be reviewed using the operation ticket review rules in the subsequent daily management stage.

[0140] In this embodiment, efficient, intelligent, and self-learning operation ticket review can be achieved, which has the following advantages: 1. Utilize the contextual learning and thought chain reasoning capabilities of the generative AI model to achieve semantic understanding and rule parsing of the operation ticket review rules in natural language form, complete self-learning of the operation ticket review rules, and support dynamic updating of the rule parsing content of the operation ticket review rules.

[0141] 2. Through one or more rounds of interactive feedback with users, the self-learning agent forms a closed-loop mechanism of "knowledge learning - reasoning verification - feedback iteration," enabling dynamic rule updates enhanced by human-machine collaboration and continuously improving the intelligent level of operation ticket review. Newly added operation ticket review rules can be stored in long-term memory, continuously optimizing the rule base.

[0142] 3. The knowledge graph and scheduling professional knowledge base dynamically verify the rationality of operation instructions to ensure the real-time and accuracy of the audit results.

[0143] Figure 5 This is a system architecture diagram of an operation ticket processing system provided in an embodiment of the present application. Figure 5 ,The operation ticket processing system may include: a terminal device 10 and a server 20 , the server 20 running a self-learning agent and an operation ticket review agent; The terminal device 10 is configured to, in response to user input operations on the dialogue interaction interface, obtain the current dialogue information input by the user in the dialogue interaction and send it to the server, and display the review results of the pending operation ticket returned by the server on the dialogue interaction interface; The server 20 is used to use a self-learning intelligent agent to use a generative AI model to identify the intent of the current dialogue information. If the intent category of the current dialogue information is identified as a new operation ticket review rule, the generative AI model is used in combination with the knowledge graph and the scheduling professional knowledge base to parse the new operation ticket review rule in the current dialogue information to obtain the rule analysis content of the new operation ticket review rule; the operation ticket review intelligent agent uses the generative AI model to review the operation instructions in the operation ticket to be reviewed according to the rule analysis content, and sends the review result of the operation ticket to be reviewed to the terminal device.

[0144] The detailed implementation and beneficial effects of each step in this embodiment have been described in detail in the aforementioned embodiments and will not be elaborated here.

[0145] It should be noted that the execution entity of each step of the method provided in the above embodiment can be the same device, or the method can be executed by different devices. For example, the execution entity of steps 201 to 203 can be device A; for another example, the execution entity of steps 201 and 202 can be device A, and the execution entity of step 203 can be device B; and so on.

[0146] In addition, in some of the processes described in the above embodiments and the accompanying drawings, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations, such as 201, 202, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0147] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 6 As shown, the electronic device includes: a memory 61 and a processor 62; Memory 61 is used to store computer programs and may be configured to store various other data to support operations on the computing platform. Examples of such data include instructions for any application or method operating on the computing platform, data structures, contact data, phone book data, messages, images, videos, etc.

[0148] The processor 62 is coupled to the memory 61 and is used to execute the computer program in the memory 61 to: execute the steps in the self-learning method of operation ticket review rules or the operation ticket review method.

[0149] Optional, such as Figure 6 As shown, the electronic device also includes: a communication component 63, a display 64, a power component 65, an audio component 66 and other components. Figure 6 Only some components are shown schematically, which does not mean that the electronic device only includes Figure 6 In addition, Figure 6The components in the dotted box are optional components, not mandatory components, and the specific components may depend on the product form of the electronic device. The electronic device of this embodiment can be implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone or an IOT (Internet of Things) device, or a server device such as a conventional server, a cloud server or a server array. If the electronic device of this embodiment is implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone, etc., it may include Figure 6 If the electronic device of this embodiment is implemented as a conventional server, cloud server or server array and other server-side devices, it may not include Figure 6 Components within the dotted box.

[0150] The above-mentioned memory can be implemented by any type of volatile or non-volatile memory device or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0151] The communication component is configured to facilitate wired or wireless communication between the device containing the communication component and other devices. The device containing the communication component can access a wireless network based on a communication standard, such as 2G (2nd Generation), 3G (3rd Generation), 4G (4th Generation) / LTE (Long Term Evolution), 5G (5th Generation), or other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel.

[0152] The display includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, it may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, slides, and gestures on the touch panel. The touch sensors can detect not only the boundaries of a touch or slide action, but also the duration and pressure associated with the touch or slide operation.

[0153] The power supply assembly provides power to various components of the device in which the power supply assembly is located. The power supply assembly may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which the power supply assembly is located.

[0154] The above-mentioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC). When the device where the audio component is located is in an operating mode, such as call mode, recording mode, and voice recognition mode, the microphone is configured to receive external audio signals. The received audio signal can be further stored in the memory or sent via the communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0155] Accordingly, embodiments of the present application further provide a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor is enabled to implement the steps of the above-described method embodiments. The computer-readable storage medium may be volatile, non-volatile, or a combination thereof, and may be removable or non-removable. Examples of computer-readable storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, disk storage or other magnetic storage devices, or any other non-transmission media.

[0156] Accordingly, the present application embodiment also provides a computer program product, which includes a computer program or instructions. When the computer program or instructions are executed by a processor, the processor is enabled to implement the steps in the above-mentioned method embodiment. It should be understood that each process or a combination of multiple processes in the above-mentioned method flow can be implemented by a computer program or instruction. In addition, these computer programs or instructions can be applied to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device, so that the processor of the general-purpose computer, the special-purpose computer, the embedded processor or other programmable data processing device can be implemented as a device for implementing the corresponding functions in the above-mentioned method embodiment.

[0157] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0158] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A self-learning method for operation ticket review rules, characterized in that: Applied to a self-learning agent, the method comprises: Get the current conversation information entered by the user in this conversation interaction; Using a generative AI model to perform intent recognition on the current conversation information to obtain an intent category of the current conversation information; If the intent category of the current conversation information is new operation ticket review rules, the generative AI model is used in combination with the knowledge graph and the scheduling professional knowledge base to parse the new operation ticket review rules in the current conversation information to obtain the rule analysis content of the new operation ticket review rules.

2. The method according to claim 1, characterized in that The rule parsing content includes: operation instruction review type, operation instruction screening thinking chain and operation instruction review thinking chain; the operation instruction review type is used to indicate independent review of a single operation instruction or joint review of multiple operation instructions; the operation instruction screening thinking chain is used to indicate a screening method for screening at least one target operation instruction to be reviewed from at least one operation instruction included in the operation ticket to be reviewed; the operation instruction review thinking chain is used to indicate the review process of the newly added operation ticket review rules.

3. The method according to claim 1, characterized in that Also includes: In response to the user confirming that the review result of the pending operation ticket is incorrect, obtaining next dialogue information input by the user in the next dialogue interaction; wherein the review result of the pending operation ticket is obtained by reviewing the operation instructions in the pending operation ticket according to the rule parsing content; Using the generative AI model to identify the intent of the next conversation information; If it is identified that the intention category of the next conversation information is to update the rule parsing content of the newly added operation ticket review rules, the generative AI model is used to update the rule parsing content of the newly added operation ticket review rules according to the next conversation information.

4. The method according to claim 3, characterized in that The rule parsing content of the newly added operation ticket review rule is updated according to the next conversation information, including: If the intent category of the next conversation message is to supplement example information, then the example information is added to the rule parsing content of the newly added operation ticket review rule, and the example information includes positive example information and / or negative example information. The positive example information includes an operation instruction example and its review result of passing the operation ticket review rule example review; the negative example information includes an operation instruction example and its review result of failing the operation ticket review rule example review; If the intent category of the next conversation message is to supplement background knowledge, then the background knowledge is added to the rule parsing content of the newly added operation ticket review rule. The background knowledge refers to the knowledge information required to assist in reviewing the operation instructions of the operation ticket to be reviewed using the newly added operation ticket review rule; If the intention category of the next conversation information is to modify the operation instruction screening thought chain, then modify the operation instruction screening thought chain in the rule parsing content of the newly added operation ticket review rule according to the next conversation information; If the intention category of the next dialogue information is to modify the operation instruction review thought chain, then the operation instruction review thought chain in the rule analysis content of the newly added operation ticket review rule is modified according to the next dialogue information.

5. The method according to any one of claims 1 to 4, characterized in that Also includes: If the intention category of the current dialogue information is not identified, outputting guidance information, the guidance information being used to guide the user to input dialogue information related to the operation ticket review; If it is identified that the intent category of the current dialogue information is an intent category unrelated to the operation ticket review, a fallback speech is output.

6. A method for reviewing an operation ticket, characterized in that: include: Get the current conversation information entered by the user in this conversation interaction; Using a generative AI model to identify the intent of the current conversation information; If the intent category of the current conversation information is identified as a new operation ticket review rule, the generative AI model is used in combination with the knowledge graph and the scheduling professional knowledge base to parse the new operation ticket review rule in the current conversation information to obtain the rule analysis content of the new operation ticket review rule; The generative AI model is used to review the operation instructions in the operation ticket to be reviewed according to the rule analysis content, and the review result of the operation ticket to be reviewed is output.

7. The method according to claim 6, characterized in that The rule parsing content includes: an operation instruction review type, an operation instruction screening thought chain, and an operation instruction review thought chain; the operation instruction review type is used to indicate whether a single operation instruction is to be reviewed independently or multiple operation instructions are to be reviewed jointly; the operation instruction screening thought chain is used to indicate a screening method for screening at least one target operation instruction to be reviewed from at least one operation instruction included in the operation ticket to be reviewed; the operation instruction review thought chain is used to indicate the review process of the newly added operation ticket review rule; Accordingly, the generative AI model is used to parse the content according to the rules to review the operation instructions in the pending operation ticket, including: Utilizing the generative AI model to screen out at least one target operation instruction to be reviewed from at least one operation instruction included in the operation ticket to be reviewed according to the operation instruction screening thought chain; If the operation instruction review type indicates that a single operation instruction is to be independently reviewed, then the single target operation instruction is independently reviewed according to the operation instruction review thought chain to obtain the review result of the single target operation instruction; and the review result of the operation ticket to be reviewed is generated based on the review result of at least one target operation instruction; If the operation instruction review type indicates a joint review of multiple operation instructions, then the multiple target operation instructions are jointly reviewed according to the operation instruction review thinking chain to obtain the review results of the multiple target operation instructions; and the review results of the operation ticket to be reviewed are generated based on the review results of the multiple target operation instructions.

8. The method according to claim 6, characterized in that After outputting the audit result of the operation ticket to be audited, the following is also included: In response to the user confirming that the review result of the to-be-reviewed operation ticket is wrong, obtaining next dialogue information input by the user in the next dialogue interaction; Using the generative AI model to identify the intent of the next conversation information, if it is identified that the intent category of the next conversation information is to update the rule parsing content of the newly added operation ticket review rules, then updating the rule parsing content of the newly added operation ticket review rules according to the next conversation information; The generative AI model is used to review the operation instructions in the pending operation ticket according to the updated rule parsing content, and the review result of the pending operation ticket is output.

9. The method according to claim 8, characterized in that The rule parsing content of the newly added operation ticket review rule is updated according to the next conversation information, including: If the intent category of the next conversation message is to supplement example information, then the example information is added to the rule parsing content of the newly added operation ticket review rule, and the example information includes positive example information and / or negative example information. The positive example information includes an operation instruction example and its review result of passing the operation ticket review rule example review; the negative example information includes an operation instruction example and its review result of failing the operation ticket review rule example review; If the intent category of the next conversation message is to supplement background knowledge, then the background knowledge is added to the rule parsing content of the newly added operation ticket review rule. The background knowledge refers to the knowledge information required to assist in reviewing the operation instructions of the operation ticket to be reviewed using the newly added operation ticket review rule; If the intention category of the next conversation information is to modify the operation instruction screening thought chain, modify the operation instruction screening thought chain in the rule parsing content of the newly added operation ticket review rule according to the next conversation information; If the intention category of the next dialogue information is to modify the operation instruction review thought chain, the operation instruction review thought chain in the rule analysis content of the newly added operation ticket review rule is modified according to the next dialogue information.

10. The method according to claim 6, characterized in that Also includes: In response to the user confirming that the audit result of the pending operation ticket is correct, the newly added operation ticket audit rule and the correct rule parsing content are associated with the pending operation ticket and stored in the rule library.

11. The method according to any one of claims 6 to 10, characterized in that Also includes: If the intention category of the current dialogue information is not identified, outputting guidance information, the guidance information being used to guide the user to input dialogue information related to the operation ticket review; If it is identified that the intent category of the current dialogue information is an intent category unrelated to the operation ticket review, a fallback speech is output.

12. The method according to any one of claims 6 to 10, characterized in that Reviewing the operation instructions in the pending operation ticket according to the rule parsing content includes: The operation instructions in the operation ticket to be reviewed are reviewed from multiple dimensions according to the rule parsing content; the multiple dimensions include: completeness check, standardization check or correctness check.

13. An operation ticket processing system, characterized in that: include: Terminal devices and a server, wherein the server runs a self-learning agent and an operation ticket review agent; The terminal device is configured to, in response to the user's input operation on the dialogue interaction interface, obtain the current dialogue information input by the user in the current dialogue interaction and send it to the server, and display the review result of the pending operation ticket returned by the server on the dialogue interaction interface; The server is configured to use the self-learning agent to perform intent recognition on the current dialogue information using a generative AI model. If the intent category of the current dialogue information is identified as a new operation ticket review rule, the server utilizes the generative AI model in combination with the knowledge graph and the scheduling professional knowledge base to parse the new operation ticket review rule in the current dialogue information to obtain the rule analysis content of the new operation ticket review rule. The operation ticket review agent uses the generative AI model to review the operation instructions in the operation ticket to be reviewed according to the rule analysis content, and sends the review result of the operation ticket to be reviewed to the terminal device.

14. An electronic device, characterized in that: include: memory and processor; The memory is used to store computer programs; The processor is coupled to the memory and configured to execute the computer program to perform the steps of the method according to any one of claims 1 to 12.

15. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor is enabled to implement the steps of the method according to any one of claims 1 to 12.

16. A computer program product, characterized in that The method comprises a computer program / instruction, which, when executed by a processor, enables the processor to implement the steps of the method according to any one of claims 1 to 12.

Images