Intelligent agent instruction alignment and database self-completion method and system

Abstract

The invention discloses an agent instruction alignment and database self-completion method and system, and relates to the technical field of artificial intelligence application, and the method comprises the steps: S1, forming an industrial agent, and carrying out incremental input; s2, the large model obtains an operator instruction for interpretation, and retrieves and calls a matching instruction from an instruction library; s3, inquiring the called instruction to an operator in an original text retelling mode, and determining to issue the instruction or collect supplement requirements according to feedback; s4, automatically retrieving the sample knowledge base to provide suggestions for operators, and supplementing the standardized suggestions to the knowledge base; and S5, regularly collecting and analyzing instruction supplement demands, developing corresponding instruction sets in a monitoring system, and performing supplement maintenance on an instruction library. According to the agent instruction alignment and knowledge closed-loop method and system, progressive enrichment and autonomous completeness of information can be realized, and full-life knowledge updating can be efficiently realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence application technology, and in particular to an intelligent agent instruction alignment and database self-completion method and system. Background Art

[0002] The new generation of artificial intelligence application technology represented by large language models is developing rapidly, and the application research in vertical industries in the industrial field is gradually deepening. There is a trend of introducing intelligent agents to assist and gradually replace the traditional manual operation mode of human judgment. The "large model + knowledge base" model based on retrieval-augmented generation (RAG) technology is the most typical industrial intelligent agent application form in the industrial field.

[0003] Industrial production has extremely high requirements for the accuracy of diagnosis and instructions, especially for high-risk parameter process industries such as thermal power generation. Any improper instructions or misjudgment of the state may cause serious accidents such as unit tripping or equipment damage. Therefore, for intelligent applications in the industrial field, accurate alignment of instructions is of great significance to the safe and reliable operation of equipment systems. However, there are understanding deviations in human-computer interaction involving large language models. In general fields, accurate understanding can be obtained through repeated communication and confirmation, but in the field of industrial applications, the process requires simplicity and accuracy. Conventional interaction methods do not meet efficiency requirements. It is necessary to ensure the complete and accurate understanding and efficient matching execution of the issued instructions, and form a continuous instruction correction and supplement mechanism through a gradually complete iterative process.

[0004] In addition, due to the high reliability requirements of industrial production systems, fault samples are very scarce. Taking thermal power units as an example, the unplanned outage rate is only less than 0.5 times / unit-year. The knowledge base constructed using historical cases and data is not complete enough. The existing method of manually collecting and constructing a knowledge base by maintenance developers cannot be connected with actual operation. During regular collation, the context is lost and it is difficult to understand correctly, which is time-consuming and labor-intensive. The human-in-the-loop online closed-loop iterative method assisted by large models can achieve progressive enrichment and autonomous completion of information, and efficiently realize full-life knowledge updates.

[0005] For example, there is a Chinese patent with publication number CN118504683A, which involves a knowledge base-based intelligent fire protection personnel auxiliary inspection method and system. The method includes: the fire protection large language model receives the question information input by the user, obtains the corresponding search results in the preset knowledge base, and outputs the corresponding question and answer results according to the set rules based on the search results, enhanced search information and question information; through the present invention, fire protection practitioners can obtain instant auxiliary information, which reduces the burden of memorizing and querying a large amount of information. The large language model is used to learn and process a large amount of professional knowledge in the fire protection field, which can provide more accurate and professional information, reduce errors in human judgment, improve the accuracy of decision-making, and help practitioners make more reasonable decisions in complex or uncertain situations. However, the Chinese patent with publication number CN118504683A uses a preset knowledge base, which is limited by the completeness of historical cases, and the collection of the knowledge base is time-consuming and labor-intensive. Summary of the invention

[0006] In order to solve the problem of how to achieve efficient knowledge updating and accurate instruction issuance in the process of using intelligent agents for auxiliary decision-making and automatic execution in the industrial field, the present invention proposes an intelligent agent instruction alignment and database self-completion method and system, which can realize the progressive enrichment and autonomous completion of information and efficiently realize full-life cycle knowledge updating.

[0007] In order to achieve the above object, the present invention adopts the following technical solution: a method for agent instruction alignment and database self-completion, comprising the following steps: S1, forming an industrial intelligent entity and performing incremental input; S2, the large model obtains the operator's instructions for interpretation, and retrieves and calls matching instructions from the instruction library; S3, ask the operator to repeat the original instructions and issue the instructions or collect additional requirements based on the feedback; S4, according to the monitoring system parameter warning or mode switching requirements, automatically retrieve the sample knowledge base and make suggestions to the operator. The operator operates according to the suggestions or enters the suggestions, and the standardized suggestions are added to the knowledge base; S5, regularly collect and analyze instruction supplement requirements, develop corresponding instruction sets in the monitoring system, and supplement and maintain the instruction library.

[0008] This technical solution can not only simplify the collection of operating instructions, improve operating efficiency, reduce operating burdens and risks, and effectively solve the potential hazards of semantic deviations in large models of industrial scenarios, but also achieve precise matching of instruction sets and minimize the difficulty of human participation in the human-in-the-loop mode, ensuring accurate and reliable information exchange, thereby achieving efficient knowledge updating and accurate issuance of instructions.

[0009] Preferably, the step S1 comprises: S11: Combine the large model with the knowledge base based on the retrieval-enhanced generation technology to obtain the industrial intelligent agent; S12: Apply industrial intelligence to enhance and incrementally input knowledge in the industry.

[0010] Preferably, step S2 comprises: S21, establish an accurately expressed executable instruction set and a matching automatic execution program; S22, forming key information expressions in the instruction library according to the instruction set and the automatic execution program; S23, using the large model to establish an interactive system to receive operator interaction information; S24, develop the retrieval and interpretation function of voice or text instructions to achieve generalized understanding of received instructions and retrieval and matching with executable instructions in the instruction library.

[0011] Preferably, the step S3 comprises: S31, when a matching instruction is retrieved, the original expression of the executable instruction in the instruction library is output to the voice device or the interactive interface in the form of voice or text; S32, accepting manual confirmation from the operator. When the operator's feedback confirms that the confirmation is correct, an execution instruction is issued to call the automatic execution program; S33, when there is no matching instruction in the search, an output prompt is given that the instruction cannot be executed correctly, and supplementary requirements are collected.

[0012] Preferably, in step S33, collecting supplementary requirements includes: S331, using the big model to standardize and interpret the instruction text and collect it into the instruction library; S332, regularly start the analysis and development tasks, and add executable programs to the instruction library.

[0013] Preferably, step S4 comprises: S41, establishes a knowledge base for fault diagnosis and decision support based on prior knowledge and historical data samples, standardizes the expression of abnormal handling and operation guidance suggestions, and automatically associates the activation information of the instruction set; S42, the large model receives abnormal warnings or mode switching requirements from the monitoring system and automatically retrieves matching sample information from the knowledge base; S43, the large model interprets and summarizes the retrieved matching information and outputs it to the operation interaction interface, and asks whether it is confirmed to be correct; S44, performing operations on the suggestions according to the confirmation result.

[0014] Preferably, the step S44 comprises: S441, when the operator confirms that it is correct, the sample accuracy weight is increased, and the next operation is performed; S442, when the big model prompts that no matching samples are retrieved or the operator's confirmation is incorrect, a knowledge input request is made, and the operator inputs the correct suggestions. The big model standardizes the operator's suggestions and enters them into the knowledge base.

[0015] Preferably, in step S441, executing the next operation includes manual processing by an operator or automatic alignment and issuance of instructions.

[0016] The present invention also adopts the following technical solution: an intelligent agent instruction alignment and database self-completion system, adopting the above-mentioned intelligent agent instruction alignment and database self-completion method, comprising: The intelligent agent, which consists of a large language model and a knowledge base, completes the human-in-the-loop operation suggestion output and the knowledge base self-complete program; Instruction library, which can map the fully automatic operation instruction set developed for regular maintenance in the monitoring system; The large language model works with the instruction library to automatically align instructions and issue them; Operators can call fully automatic instructions by aligning and issuing instructions, or they can manually perform related operations based on operation suggestions.

[0017] The beneficial effects of the present invention are: 1) Through the automated integration of complex operations and pre-processing of the instruction system, the collection and simplification of operation instructions are realized, which improves the operation efficiency and reduces the operation burden and risk; 2) It effectively solves the potential hazards of semantic deviation of large models of industrial scenarios, naturally realizes the precise matching of instruction sets in the semantic interaction link, and minimizes the difficulty of human participation in the human-in-the-loop mode, ensuring the accuracy and reliability of information interaction; 3) Text interaction is mainly applicable to the manual confirmation of intelligent agent decision-making auxiliary instructions, and voice interaction can also be applied to scenarios where manual instructions are issued to achieve fully automatic execution of combined operation modes; 4) The human-in-the-loop closed-loop iterative method with online assisted interpretation of large models can achieve progressive enrichment and autonomous completion of information, and efficiently realize full-life knowledge updating. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of instruction alignment of the present invention.

[0019] Figure 2 It is a self-complete flow chart of the knowledge base of the present invention.

[0020] Figure 3 It is a system block diagram of the present invention. DETAILED DESCRIPTION

[0021] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific implementation method described herein is only an optimal embodiment of the present invention, which is only used to explain the present invention and does not limit the scope of protection of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0022] Example 1 This embodiment provides an intelligent agent instruction alignment and database self-completion method, which is applicable to the entire process of industrial intelligent agent diagnosis and execution. It uses a large language model (LLM) to collaboratively complete the self-completion of the diagnostic knowledge base and the precise matching of execution instructions, so that the industrial intelligent control system can efficiently realize the knowledge accumulation and update and automatic execution of the entire life cycle in actual operation, and finally tend to autonomous decision-making operation. It mainly includes the following steps.

[0023] Step S1, forming an industrial intelligent entity and performing incremental input; the specific process of S1 includes the following sub-steps.

[0024] Step S11, based on the retrieval enhancement generation technology, a large language model is combined with a knowledge base to form an industrial intelligent agent.

[0025] In this embodiment, the large model is a large language model (LLM for short), which refers to a deep learning model trained with a large amount of text data, which can generate natural language text or understand the meaning of language text.

[0026] Large language models can handle a variety of natural language tasks, such as text classification, question answering, and conversation.

[0027] Step S12, apply industrial intelligence to perform knowledge enhancement and incremental entry in the industry field.

[0028] In this embodiment, an industrial intelligent agent that combines a large model and a knowledge base based on RAG technology realizes knowledge enhancement and incremental entry in the industry field, and the intelligent agent and the operator interact in the following two scenarios through voice or text.

[0029] First, in the instruction alignment stage, the large model obtains the operator's instructions for interpretation, retrieves and calls matching instructions from the instruction library, and asks the operator to repeat the retrieved instructions in the original text. Based on the feedback, it determines to issue instructions or collect additional requirements.

[0030] Second, in the self-completion stage of the knowledge base, the big model automatically retrieves the sample knowledge base according to the monitoring system parameter warning or mode switching requirements, and makes suggestions to the operator. The operator can operate manually according to the suggestions or enable the fully automatic instruction set. If the correct sample is not retrieved, the operator enters the suggestion and adds it to the knowledge base after standardization by the big model.

[0031] It should be noted that step S2 and step S3 implement instruction alignment. Figure 1 .

[0032] Step S2, obtain the operator's instructions, interpret them by the large model, and then retrieve and call the matching instructions from the instruction library; the specific process of S2 includes the following sub-steps.

[0033] Step S21, firstly establish an accurately expressed executable instruction set and a matching automatic execution program.

[0034] In this embodiment, an executable instruction set for forming a key information expression in an instruction library and an automatic execution program matching the instruction set are established. The executable instruction set includes several specific instructions, and the automatic execution program is used to execute the instructions in the executable instruction set.

[0035] Step S22, then based on the instruction set and the automatic execution program, a key information expression is formed in the instruction library.

[0036] In this embodiment, while establishing the instruction set and the automatic execution program, a key information expression for instruction retrieval is also formed in the instruction library to facilitate instruction retrieval.

[0037] Step S23, then use the large model to establish an interactive system to receive operator interaction information.

[0038] In this embodiment, the interactive system here includes a voice interactive system and a text interactive system, and the operator's instructions can be voice or text instructions. Text interaction is mainly applicable to the manual confirmation of intelligent agent decision-making auxiliary instructions, and voice interaction can also be applied to manual instruction scenarios to achieve fully automatic execution of combined operation modes.

[0039] Step S24, finally developing the retrieval and interpretation function of voice or text instructions, realizing the generalization of the understanding of the received instructions and the retrieval and matching with the executable instructions in the instruction library.

[0040] In this embodiment, a retrieval and interpretation function is developed for the interactive system, and the interactive system performs retrieval and interpretation of instructions, wherein the interpretation function is used to generalize the understanding of the operator's voice or text instructions, and the retrieval function is used to retrieve and match the generalized instructions with executable instructions in the instruction library.

[0041] Step S3, the retrieved instruction is repeated to the operator in the original text, and the instruction is issued or supplementary requirements are collected based on the feedback; this step mainly includes the following multiple sub-steps.

[0042] Step S31, when a matching instruction is retrieved, the original expression of the executable instruction in the instruction library is output to the voice device or interactive interface in the form of voice or text.

[0043] In this embodiment, if the instruction received from the operator is a voice instruction, the interactive system matches the voice instruction with the instructions in the instruction library through the retrieval and interpretation function. When the interactive system retrieves the matching instruction, the original expression of the executable instruction in the instruction library is output to the voice device in the form of voice for manual confirmation by the operator.

[0044] If the instruction received from the operator is a text instruction, the interactive system matches the text instruction with the instructions in the instruction library through the retrieval and interpretation function. When the interactive system retrieves the matching instruction, the original expression of the executable instruction in the instruction library is output to the interactive interface in text form for manual confirmation by the operator.

[0045] Step S32, accepting manual confirmation from the operator, and when the operator's feedback confirmation is correct, issuing an execution instruction to call the automatic execution program.

[0046] In this embodiment, the operator performs semantic judgment to confirm whether the instructions output by the interactive system are correct and feeds back to the interactive system. The interactive system receives the operator's feedback results. If the feedback obtained is correct, the large model issues instructions and calls the automatic execution program to execute the corresponding instructions in the instruction library, and the instruction alignment link ends; if the feedback result is incorrect, it indicates that there is no matching instruction.

[0047] Step S33: When no matching instruction is found in the search, a prompt is outputted indicating that the instruction cannot be correctly executed, and supplementary requirements are collected.

[0048] In this embodiment, if the interactive system does not retrieve a matching instruction in the instruction library, it means that the instruction entered by the operator is not currently in the instruction library and needs to be supplemented. The interactive system outputs a prompt that the instruction cannot be executed correctly, and the large model collects the requirements that need to be supplemented.

[0049] In step S33, the process of collecting supplementary requirements specifically includes the following steps.

[0050] The first step is to use a large model to standardize the interpretation and organization of instruction texts and collect them into an instruction library.

[0051] The second step is to regularly initiate analysis and development tasks and add executable programs to the instruction library.

[0052] In the technical solution for step S3 of the present embodiment, the retrieval and interpretation function of voice or text instructions is used to achieve a generalized understanding of received instructions and a retrieval and match with executable instructions in the instruction library. The automated integration of complex operations and preprocessing of the instruction system realize the simplification of the collection of operation instructions, improve the operation efficiency, reduce the operation burden and risk, and effectively solve the potential hazards of semantic deviations in large models of industrial scenarios. In the semantic interaction link, accurate matching of instruction sets is naturally achieved, and the difficulty of human participation in the human-in-the-loop mode is minimized, ensuring accurate and reliable information interaction.

[0053] After the instruction alignment phase, the knowledge base is self-completed through steps S4 and S5. The flowchart is shown in Figure 2 .

[0054] Step S4, according to the monitoring system parameter warning or mode switching requirements, automatically retrieve the sample knowledge base and make suggestions to the operator. The operator operates or enters the suggestions according to the suggestions, and the standardized suggestions are added to the knowledge base. The specific process of S4 includes the following sub-steps.

[0055] Step S41, based on prior knowledge and historical data samples, a fault diagnosis and decision-making support knowledge base is established, abnormal handling and operation guidance suggestions are expressed in a standardized manner, and fully automatic instruction set association activation information is generated.

[0056] In this embodiment, it is first necessary to establish a knowledge base. The knowledge base is established based on prior knowledge and historical data samples, and can be used for fault diagnosis and decision-making assistance. After the knowledge base is established, the abnormal handling and operation guidance suggestions, or the fully automatic instruction set association activation information of the instruction alignment link are standardized.

[0057] Step S42: the large model receives abnormal warnings or mode switching requirements from the monitoring system and automatically retrieves matching sample information from the knowledge base.

[0058] In this embodiment, when the monitoring system detects abnormal parameters or changes in conditions, it sends an abnormal warning or mode switching request to the large model. The large model automatically searches the knowledge base and matches the sample information based on the received abnormal warning or mode switching request.

[0059] In step S43, the large model interprets and summarizes the retrieved matching information and outputs it to the operation interaction interface, and asks whether it is confirmed to be correct.

[0060] In this embodiment, the large model interprets the retrieved matching information in terms of phenomenon or demand, and then summarizes and outputs the interpreted operation suggestions to the operation interaction interface for confirmation by the operator.

[0061] Step S44: perform operations on the suggestions according to the confirmation result.

[0062] During step S44, if the following situations exist, the suggestion is operated; when the operator confirms that it is correct, the sample accuracy weight is increased, and the operator manually handles the next operation; when the large model prompts that no matching sample is retrieved or the operator's confirmation is incorrect, the large language model makes a knowledge input request, and then the operator enters the correct suggestion. The large model standardizes the operator's suggestion and enters it into the knowledge base to supplement the knowledge base.

[0063] In the technical solution for step S4 of this embodiment, a human-in-the-loop closed-loop iterative method with online assisted interpretation of a large model is adopted, which can realize the progressive enrichment and autonomous completion of information, efficiently realize the full-life cycle knowledge update, avoid the time-consuming and labor-intensive collection of knowledge bases, and avoid the problem of incomplete cases caused by the use of preset knowledge bases.

[0064] Step S5, regularly collect and analyze instruction supplement requirements, develop corresponding instruction sets in the monitoring system, and supplement and maintain the instruction library.

[0065] In this embodiment, after the instruction library collects instruction supplement requirements, maintenance personnel regularly collect and analyze them, develop corresponding instruction sets in the monitoring system, and supplement and maintain the instruction library.

[0066] The operation method of the traditional industrial system is that the operator directly operates the screen, which has complex positioning and cumbersome operation. An intelligent agent knowledge closed-loop and instruction alignment method in this embodiment realizes the simplification of operation instructions through the automated integration of complex operations and instruction system preprocessing, thereby improving operation efficiency and reducing operation burden and risk.

[0067] The method of aligning the closed loop of intelligent knowledge and instructions in this embodiment effectively solves the potential hazards of semantic deviation of large models of industrial scenarios, naturally realizes the precise matching of instruction sets in the semantic interaction link, and minimizes the difficulty of human participation in the loop mode, ensuring the accuracy and reliability of information interaction. Text interaction is mainly applicable to the manual confirmation of intelligent decision-making auxiliary instructions, and voice interaction can also be applied to the scene of manual instruction issuance to realize the fully automatic execution of the combined operation mode.

[0068] Traditional knowledge base construction and maintenance methods have poor timeliness. During regular maintenance, the context is lost and it is difficult to correctly understand, which is time-consuming and laborious. A method for aligning the closed-loop knowledge and instructions of an intelligent agent in this embodiment adopts a human-in-the-loop closed-loop iteration method with online assisted interpretation of a large model, which can realize the progressive enrichment and autonomous completion of information and efficiently realize full-life knowledge updating.

[0069] Example 2 This embodiment provides an intelligent agent instruction alignment and database self-completion method, which is applicable to the entire process of industrial intelligent agent diagnosis and execution. It uses a large language model (LLM) to collaboratively complete the self-completion of the diagnostic knowledge base and the precise matching of execution instructions, so that the industrial intelligent control system can efficiently realize the knowledge accumulation and update and automatic execution of the entire life cycle in actual operation, and finally tend to autonomous decision-making operation. It mainly includes the following steps.

[0070] Step S1, forming an industrial intelligent entity and performing incremental input; the specific process of S1 includes the following sub-steps.

[0071] Step S11, based on the retrieval enhancement generation technology, a large language model is combined with a knowledge base to form an industrial intelligent agent.

[0072] In this embodiment, the large model is a large language model (LLM for short), which refers to a deep learning model trained with a large amount of text data, which can generate natural language text or understand the meaning of language text.

[0073] Large language models can handle a variety of natural language tasks, such as text classification, question answering, and conversation.

[0074] Step S12, apply industrial intelligence to perform knowledge enhancement and incremental entry in the industry field.

[0075] In this embodiment, an industrial intelligent agent that combines a large model and a knowledge base based on RAG technology realizes knowledge enhancement and incremental entry in the industry field, and the intelligent agent and the operator interact in the following two scenarios through voice or text.

[0076] First, in the instruction alignment stage, the large model obtains the operator's instructions for interpretation, retrieves and calls matching instructions from the instruction library, and asks the operator to repeat the retrieved instructions in the original text. Based on the feedback, it determines to issue instructions or collect additional requirements.

[0077] Second, in the self-completion stage of the knowledge base, the big model automatically retrieves the sample knowledge base according to the monitoring system parameter warning or mode switching requirements, and makes suggestions to the operator. The operator can operate manually according to the suggestions or enable the fully automatic instruction set. If the correct sample is not retrieved, the operator enters the suggestion and adds it to the knowledge base after standardization by the big model.

[0078] It should be noted that step S2 and step S3 implement instruction alignment. Figure 1 .

[0079] Step S2, obtain the operator's instructions, interpret them by the large model, and then retrieve and call the matching instructions from the instruction library; the specific process of S2 includes the following sub-steps.

[0080] Step S21, firstly establish an accurately expressed executable instruction set and a matching automatic execution program.

[0081] In this embodiment, an executable instruction set for forming a key information expression in an instruction library and an automatic execution program matching the instruction set are established. The executable instruction set includes several specific instructions, and the automatic execution program is used to execute the instructions in the executable instruction set.

[0082] Step S22, then based on the instruction set and the automatic execution program, a key information expression is formed in the instruction library.

[0083] In this embodiment, while establishing the instruction set and the automatic execution program, a key information expression for instruction retrieval is also formed in the instruction library to facilitate instruction retrieval.

[0084] Step S23, then use the large model to establish an interactive system to receive operator interaction information.

[0085] In this embodiment, the interactive system here includes a voice interactive system and a text interactive system, and the operator's instructions can be voice or text instructions. Text interaction is mainly applicable to the manual confirmation of intelligent agent decision-making auxiliary instructions, and voice interaction can also be applied to manual instruction scenarios to achieve fully automatic execution of combined operation modes.

[0086] Step S24, finally developing the retrieval and interpretation function of voice or text instructions, realizing the generalization of the understanding of the received instructions and the retrieval and matching with the executable instructions in the instruction library.

[0087] In this embodiment, a retrieval and interpretation function is developed for the interactive system, and the interactive system performs retrieval and interpretation of instructions, wherein the interpretation function is used to generalize the understanding of the operator's voice or text instructions, and the retrieval function is used to retrieve and match the generalized instructions with executable instructions in the instruction library.

[0088] Step S3, the retrieved instruction is repeated to the operator in the original text, and the instruction is issued or supplementary requirements are collected based on the feedback; this step mainly includes the following multiple sub-steps.

[0089] Step S31, when a matching instruction is retrieved, the original expression of the executable instruction in the instruction library is output to the voice device or interactive interface in the form of voice or text.

[0090] In this embodiment, if the instruction received from the operator is a voice instruction, the interactive system matches the voice instruction with the instructions in the instruction library through the retrieval and interpretation function. When the interactive system retrieves the matching instruction, the original expression of the executable instruction in the instruction library is output to the voice device in the form of voice for manual confirmation by the operator.

[0091] If the instruction received from the operator is a text instruction, the interactive system matches the text instruction with the instructions in the instruction library through the retrieval and interpretation function. When the interactive system retrieves the matching instruction, the original expression of the executable instruction in the instruction library is output to the interactive interface in text form for manual confirmation by the operator.

[0092] Step S32, accepting manual confirmation from the operator, and when the operator's feedback confirmation is correct, issuing an execution instruction to call the automatic execution program.

[0093] In this embodiment, the operator performs semantic judgment to confirm whether the instructions output by the interactive system are correct and feeds back to the interactive system. The interactive system receives the operator's feedback results. If the feedback obtained is correct, the large model issues instructions and calls the automatic execution program to execute the corresponding instructions in the instruction library, and the instruction alignment link ends; if the feedback result is incorrect, it indicates that there is no matching instruction.

[0094] Step S33: When no matching instruction is found in the search, a prompt is outputted indicating that the instruction cannot be correctly executed, and supplementary requirements are collected.

[0095] In this embodiment, if the interactive system does not retrieve a matching instruction in the instruction library, it means that the instruction entered by the operator is not currently in the instruction library and needs to be supplemented. The interactive system outputs a prompt that the instruction cannot be executed correctly, and the large model collects the requirements that need to be supplemented.

[0096] In step S33, the process of collecting supplementary requirements specifically includes the following steps.

[0097] The first step is to use a large model to standardize the interpretation and organization of instruction texts and collect them into an instruction library.

[0098] The second step is to regularly initiate analysis and development tasks and add executable programs to the instruction library.

[0099] In the technical solution for step S3 of the present embodiment, the retrieval and interpretation function of voice or text instructions is used to achieve a generalized understanding of received instructions and a retrieval and match with executable instructions in the instruction library. The automated integration of complex operations and preprocessing of the instruction system realize the simplification of the collection of operation instructions, improve the operation efficiency, reduce the operation burden and risk, and effectively solve the potential hazards of semantic deviations in large models of industrial scenarios. In the semantic interaction link, accurate matching of instruction sets is naturally achieved, and the difficulty of human participation in the human-in-the-loop mode is minimized, ensuring accurate and reliable information interaction.

[0100] After the instruction alignment phase, the knowledge base is self-completed through steps S4 and S5. The flowchart is shown in Figure 2 .

[0101] Step S4, according to the monitoring system parameter warning or mode switching requirements, automatically retrieve the sample knowledge base and make suggestions to the operator. The operator operates or enters the suggestions according to the suggestions, and the standardized suggestions are added to the knowledge base. The specific process of S4 includes the following sub-steps.

[0102] Step S41, based on prior knowledge and historical data samples, a fault diagnosis and decision-making support knowledge base is established, abnormal handling and operation guidance suggestions are expressed in a standardized manner, and fully automatic instruction set association activation information is generated.

[0103] In this embodiment, it is first necessary to establish a knowledge base. The knowledge base is established based on prior knowledge and historical data samples, and can be used for fault diagnosis and decision-making assistance. After the knowledge base is established, the abnormal handling and operation guidance suggestions, or the fully automatic instruction set association activation information of the instruction alignment link are standardized.

[0104] Step S42: the large model receives abnormal warnings or mode switching requirements from the monitoring system and automatically retrieves matching sample information from the knowledge base.

[0105] In this embodiment, when the monitoring system detects abnormal parameters or changes in conditions, it sends an abnormal warning or mode switching request to the large model. The large model automatically searches the knowledge base and matches the sample information based on the received abnormal warning or mode switching request.

[0106] Step S43: the large model interprets and summarizes the retrieved matching information and outputs it to the operation interaction interface, and asks whether it is confirmed to be correct.

[0107] In this embodiment, the large model interprets the retrieved matching information in terms of phenomenon or demand, and then summarizes and outputs the interpreted operation suggestions to the operation interaction interface for confirmation by the operator.

[0108] Step S44: perform operations on the suggestions according to the confirmation result.

[0109] During step S44, if the following situations exist, the suggestion is operated; when the operator confirms that it is correct, the sample accuracy weight is increased, and the instruction is automatically aligned and sent to perform the next step; when the large model prompts that no matching sample is retrieved or the operator's confirmation is incorrect, a knowledge input request is made, and the operator enters the correct suggestion. The large model standardizes the operator's suggestion and enters it into the knowledge base.

[0110] The difference from the first embodiment is that when the operator confirms that it is correct, the instruction is automatically aligned and issued to execute the next operation, without the need for manual handling by the operator, further reducing the proportion of human participation.

[0111] Step S5, regularly collect and analyze instruction supplement requirements, develop corresponding instruction sets in the monitoring system, and supplement and maintain the instruction library.

[0112] In this embodiment, after the instruction library collects instruction supplement requirements, maintenance personnel regularly collect and analyze them, develop corresponding instruction sets in the monitoring system, and supplement and maintain the instruction library.

[0113] The operation method of the traditional industrial system is that the operator directly operates the screen, which has complex positioning and cumbersome operation. An intelligent agent knowledge closed-loop and instruction alignment method in this embodiment realizes the simplification of operation instructions through the automated integration of complex operations and instruction system preprocessing, thereby improving operation efficiency and reducing operation burden and risk.

[0114] The method of aligning the closed loop of intelligent knowledge and instructions in this embodiment effectively solves the potential hazards of semantic deviation of large models of industrial scenarios, naturally realizes the precise matching of instruction sets in the semantic interaction link, and minimizes the difficulty of human participation in the loop mode, ensuring the accuracy and reliability of information interaction. Text interaction is mainly applicable to the manual confirmation of intelligent decision-making auxiliary instructions, and voice interaction can also be applied to the scene of manual instruction issuance to realize the fully automatic execution of the combined operation mode.

[0115] Traditional knowledge base construction and maintenance methods have poor timeliness. During regular maintenance, the context is lost and it is difficult to correctly understand, which is time-consuming and laborious. A method for aligning the closed-loop knowledge and instructions of an intelligent agent in this embodiment adopts a human-in-the-loop closed-loop iteration method with online assisted interpretation of a large model, which can realize the progressive enrichment and autonomous completion of information and efficiently realize full-life knowledge updating.

[0116] Example 3 This embodiment provides an intelligent agent instruction alignment and database self-completion system, using the above-mentioned intelligent agent knowledge closed loop and instruction alignment method, reference Figure 3 , including intelligent agents, instruction libraries, large language models and operators.

[0117] Among them, the intelligent agent is composed of a large language model and a knowledge base, which completes the operation suggestion output of the human in the loop and the self-complete program of the knowledge base. The instruction library can map the fully automatic operation instruction set regularly maintained and developed in the monitoring system. The large language model and the instruction library cooperate to perform instruction alignment and automatic issuance operations. The operator can call the fully automatic instructions through the instruction alignment and issuance method, or manually perform related operations based on the operation suggestions.

[0118] The industrial intelligence entity that combines the large model and knowledge base based on RAG technology realizes knowledge enhancement and incremental entry in the industry field.

[0119] The intelligent agent and the operator achieve instruction alignment and knowledge base completion through voice or text.

[0120] In the instruction alignment phase, the large model obtains the operator's instructions for interpretation, retrieves and calls matching instructions from the instruction library, and asks the operator to repeat the retrieved instructions in the original text. Based on the feedback, it determines whether to issue instructions or collect additional requirements.

[0121] In the self-completion stage of the knowledge base, the big model automatically retrieves the sample knowledge base according to the monitoring system parameter warning or mode switching requirements, and makes suggestions to the operator. The operator can operate manually according to the suggestions or enable the fully automatic instruction set. If the correct sample is not retrieved, the operator enters the suggestion and it is standardized by the big model and then added to the knowledge base.

[0122] After the instruction library collects the instruction supplement requirements, maintenance personnel will regularly collect and analyze them, develop corresponding instruction sets in the monitoring system, and supplement and maintain the instruction library.

Claims

1. A method for agent instruction alignment and database self-completion, characterized in that: The following steps are involved: S1, forming an industrial intelligent entity and performing incremental input; S2, the large model obtains the operator's instructions for interpretation, and retrieves and calls matching instructions from the instruction library; S3, ask the operator to repeat the original instructions and issue the instructions or collect additional requirements based on the feedback; S4, automatically retrieves sample knowledge base to make suggestions to operators and adds standardized suggestions to the knowledge base; S5, regularly collect and analyze instruction supplement requirements, develop corresponding instruction sets in the monitoring system, and supplement and maintain the instruction library.

2. According to claim 1, a method for agent instruction alignment and database self-completion is characterized in that: The step S1 comprises: S11: Combine the large model with the knowledge base based on the retrieval-enhanced generation technology to obtain the industrial intelligent agent; S12: Apply industrial intelligence to enhance and incrementally input knowledge in the industry.

3. The method for agent instruction alignment and database self-completion according to claim 1, characterized in that: The step S2 comprises: S21, establish an accurately expressed executable instruction set and a matching automatic execution program; S22, forming key information expressions in the instruction library according to the instruction set and the automatic execution program; S23, using the large model to establish an interactive system to receive operator interaction information; S24, develop the retrieval and interpretation function of voice or text instructions to achieve generalized understanding of received instructions and retrieval and matching with executable instructions in the instruction library.

4. The method for agent instruction alignment and database self-completion according to claim 1, characterized in that: The step S3 comprises: S31, when a matching instruction is retrieved, the original text expression of the executable instruction in the instruction library is output to the voice device or the interactive interface in the form of voice or text; S32, accepting manual confirmation from the operator. When the operator's feedback confirms that the confirmation is correct, an execution instruction is issued to call the automatic execution program; S33, when there is no matching instruction in the search, an output prompt is given that the instruction cannot be executed correctly, and supplementary requirements are collected.

5. The method for agent instruction alignment and database self-completion according to claim 4, characterized in that: In step S33, collecting additional requirements includes: S331, using the big model to standardize and interpret the instruction text and collect it into the instruction library; S332, regularly start the analysis and development tasks, and add executable programs to the instruction library.

6. A method for agent instruction alignment and database self-completion according to claim 1, 3 or 4, characterized in that: The step S4 comprises: S41, establishes a knowledge base for fault diagnosis and decision support based on prior knowledge and historical data samples, standardizes the expression of abnormal handling and operation guidance suggestions, and automatically associates the activation information of the instruction set; S42, the large model receives abnormal warnings or mode switching requirements from the monitoring system and automatically retrieves matching sample information from the knowledge base; S43, the large model interprets and summarizes the retrieved matching information and outputs it to the operation interaction interface, and asks whether it is confirmed to be correct; S44, performing operations on the suggestions according to the confirmation result.

7. The method for agent instruction alignment and database self-completion according to claim 6, characterized in that: The step S44 comprises: S441, when the operator confirms that it is correct, the sample accuracy weight is increased and the next operation is performed; S442, when the big model prompts that no matching samples are retrieved or the operator's confirmation is incorrect, a knowledge input request is made, and the operator inputs the correct suggestions. The big model standardizes the operator's suggestions and enters them into the knowledge base.

8. The method for agent instruction alignment and database self-completion according to claim 7, characterized in that: In step S441, the execution of the next step operation includes manual processing by an operator or automatic alignment and issuance of instructions.

9. An agent instruction alignment and database self-completion system, characterized in that: A method for agent instruction alignment and database self-completion according to any one of claims 1 to 7, comprising: The intelligent agent, which consists of a large language model and a knowledge base, completes the human-in-the-loop operation suggestion output and the knowledge base self-complete program; Instruction library, which can map the fully automatic operation instruction set developed for regular maintenance in the monitoring system; The large language model works with the instruction library to automatically align instructions and issue them; Operators can call fully automatic instructions by aligning and issuing instructions, or they can manually perform related operations based on operation suggestions.

Images