A power generation enterprise management method and system based on large model dialogue interaction

Abstract

This invention relates to the field of intelligent operation and management technology for power generation enterprises. It provides a management method and system for power generation enterprises based on a large-scale model-based dialogue interaction. The steps are as follows: Collecting multi-source heterogeneous data and performing standardized processing to construct a unified structured data resource pool; training and optimizing a vertical domain dialogue model with power generation industry knowledge based on this resource pool; updating the domain knowledge graph and model parameters in real time through an interaction hub; providing a natural language interaction entry point, supporting operators to initiate intelligent question-and-answer, operation guidance, risk warning, and decision-making suggestion requests via voice or text, achieving cross-departmental data collaborative analysis; and outputting personalized response results based on personnel permissions and historical interaction records. This solves the problems of low decision-making efficiency and collaborative difficulties in existing power generation enterprise management caused by isolated data, unnatural interaction, and lagging knowledge updates.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and management technology for power generation enterprises, and more specifically, to a power generation enterprise management method and system based on large-scale model dialogue interaction. Background Technology

[0002] Traditional power generation companies typically rely on discrete monitoring systems, manual experience-based decision-making, and hierarchical reporting mechanisms for operation and management. Their information systems are largely focused on data acquisition and static analysis, with data from various business modules (such as equipment management, load forecasting, inspection and maintenance, and safety production) isolated, leading to delayed decision-making responses. With the large-scale grid connection of new energy sources and the deepening of power market reforms, power generation companies face challenges such as increasingly complex interactions between power sources, grids, loads, and storage; higher requirements for real-time control; and difficulties in passing on personnel knowledge and experience. While existing technologies have introduced SCADA, EMS, and various expert systems to improve automation, their knowledge base updates rely on manual input, resulting in rigid interaction methods that struggle to adapt to the needs of intelligent decision-making and cross-departmental collaboration in dynamic environments. Furthermore, they fail to provide operators with natural, efficient, and personalized interactive support.

[0003] In recent years, artificial intelligence (AI) technology has been initially applied in the power industry to single scenarios such as fault diagnosis and load forecasting. However, most models are small-scale algorithms trained for specific tasks, with limited generalization capabilities, and require structured data input and professional human-computer interfaces, resulting in high interaction barriers. Although pre-trained large models have made breakthroughs in natural language understanding and generation, their deep application in the vertical field of power generation enterprises is still in the exploratory stage. Existing technologies lack methods to deeply integrate the dialogue interaction capabilities of large language models with the multi-source heterogeneous data of power generation enterprises (real-time operation data, equipment ledgers, procedural documents, and historical cases). This has prevented the construction of a dialogue-based natural language control hub to achieve the organic unity of dynamic knowledge fusion, intelligent question answering, operational guidance, risk warning, and decision-making suggestions. Consequently, significant shortcomings remain in the technology-enabled management process, including unintuitive interaction, lagging knowledge updates, and insufficient system flexibility. Summary of the Invention

[0004] The purpose of this invention is to provide a power generation enterprise management method and system based on large-scale model dialogue interaction, which aims to solve the problems of low decision-making efficiency and difficulty in collaboration caused by data isolation, unnatural interaction and lagging knowledge updates in the existing power generation enterprise management.

[0005] This invention is achieved through the following technical solution: A power generation enterprise management method based on large-scale model dialogue interaction includes the following steps: Collect real-time operational data, equipment ledger data, procedure document data, historical case data, and cross-business module related data, and form a structured data resource pool with a unified format through data standardization processing and heterogeneous data adaptation and conversion; Based on the structured data resource pool, a basic training sample set is constructed. Through domain knowledge injection and fine-tuning, an interactive hub with professional knowledge of the power generation industry is formed. Based on the pre-trained large language model, a vertical domain dialogue model for power generation enterprises is constructed. By receiving new operational data, equipment maintenance records, updated procedures and manual feedback from power generation companies in real time through the interactive hub, the system completes the iteration of the knowledge graph in the power generation field and the parameter optimization of the vertical dialogue model of power generation companies. Based on the interaction hub, a natural language interaction entry point is provided, which allows operators to initiate intelligent Q&A, operation guidance, risk warning and decision suggestion requests in the form of voice or text, and realize the real-time call and collaborative analysis of cross-departmental business data; Personalized response results are output through the interactive hub, and equipment operation status analysis, load forecast results, inspection and maintenance plans and safety production early warning information are presented in a customized manner based on the operator's job authority, operating habits and historical interaction records.

[0006] Optionally, the specific process of data standardization and heterogeneous data adaptation and conversion is as follows: Extract the raw data patterns from real-time operation data, equipment ledger data, procedure document data, historical case data, and cross-business module related data from real-time monitoring systems, equipment management systems, and document databases; Based on the pre-set unified data standard for power generation enterprises, the original data model is mapped and transformed, converting heterogeneous data sources with different formats and protocols into standardized data entries under the unified data model. Perform spatiotemporal alignment and semantic association on standardized data entries to establish links across data sources; Based on the unified data model and the aforementioned link relationships, a structured data resource pool is constructed and updated.

[0007] Optionally, the specific process of constructing the vertical domain dialogue model for power generation enterprises is as follows: Based on the structured data resource pool, a basic training sample set is constructed, which includes equipment operating status, operating procedures, fault cases, and cross-business related question-and-answer pairs. The basic training sample set is integrated with a pre-set power generation terminology database, safety procedure texts, and historical decision logs to extract domain entities, relationships, and rules, generating an enhanced domain knowledge representation. A two-stage fine-tuning process is performed on the initial vertical domain dialogue model for power generation enterprises built based on a pre-trained large language model. The first stage uses instruction fine-tuning to adapt to natural language queries in the power generation domain. The second stage uses reinforcement learning alignment fine-tuning to optimize the accuracy, security, and operability of the model output. The performance of the fine-tuned initial vertical domain dialogue model for power generation enterprises was evaluated using a validation set, and the model parameters were adjusted based on the evaluation results to obtain the vertical domain dialogue model for power generation enterprises.

[0008] Optionally, the specific process of fine-tuning the initial vertical domain dialogue model for power generation enterprises based on the pre-trained large language model in two stages is as follows: Based on the aforementioned basic training sample set, the initial power generation enterprise vertical domain dialogue model is fine-tuned in the first stage to enable the initial power generation enterprise vertical domain dialogue model to understand and generate natural language queries and responses that conform to the professional standards of the power generation field. Using the enhanced domain knowledge representation, a reward model for reinforcement learning is constructed; wherein, the reward model quantitatively evaluates the accuracy, security, and operability of the output of the initial power generation enterprise vertical domain dialogue model based on the power generation field safety regulations and operating procedures; Based on the model output after fine-tuning the first-stage instructions and the evaluation results of the reward model, a reinforcement learning algorithm is used for the second-stage alignment fine-tuning to optimize the model's decision-making logic and response quality in complex power generation business scenarios. The results of the instruction fine-tuning and reinforcement learning alignment fine-tuning are integrated to output the vertical domain dialogue model of the power generation enterprise.

[0009] Optionally, the specific process of completing the iteration of the knowledge graph in the power generation field and the parameter optimization of the vertical domain dialogue model of power generation enterprises is as follows: The newly added operational data, equipment maintenance records, procedure update files and manual feedback information received in real time by the interaction hub are standardized and adapted. The newly added heterogeneous data is converted into incremental data that conforms to the format compatible with the structured data resource pool. At the same time, key information fields are extracted from the incremental data, including equipment ID, timestamp, event type and feedback score. Based on incremental data and extracted entities, relationships and rules in the power generation field, new domain entities are identified from the incremental data, entity attributes are updated, relationships between entities are supplemented, and conflict detection and fusion updates are performed on the existing knowledge graph to generate an iterative power generation domain knowledge graph. The incremental data is associated with the iterative knowledge graph of the power generation field to generate new question-answer pairs. The sample priority is marked by human feedback information to form an incremental training sample set. The incremental training sample set and the basic training sample set together constitute the data foundation for model optimization. Based on the incremental training sample set, the parameters of the vertical domain dialogue model of power generation enterprises are updated through an incremental fine-tuning strategy. The incremental fine-tuning strategy continues the logic of the two-stage fine-tuning. In the first stage, the instruction fine-tuning is performed for the newly added question-answer pairs to adapt to the natural language expression of incremental knowledge. In the second stage, reinforcement learning alignment fine-tuning is adopted based on human feedback information to optimize the model's response accuracy and operability to the new scenarios. The iterated knowledge graph is subjected to consistency verification, the power generation enterprise vertical domain dialogue model with optimized parameters is evaluated, and the verified knowledge graph and the evaluated model parameters are synchronized to the interaction hub.

[0010] Optionally, the implementation of real-time access and collaborative analysis of cross-departmental business data specifically includes: The system receives collaborative analysis requests initiated by operators through the interaction hub and parses the departmental roles, data categories, and analysis objectives involved in the requests. Based on the structured data resource pool and real-time data interface, cross-departmental data queries are initiated to the business systems corresponding to equipment management, safe production, and scheduling operations, and real-time or near-real-time business status data is obtained. By utilizing a vertical dialogue model for power generation companies, we can perform correlation analysis, conflict detection, and trend projection on the acquired cross-departmental business data, and generate collaborative analysis reports and decision-making recommendations. Collaborative analysis reports and decision recommendations are distributed to relevant departmental staff through an interactive hub, and support follow-up questions, corrections, and collaborative confirmations based on natural language, forming a closed-loop cross-departmental collaborative process.

[0011] Optionally, the customized presentation of equipment operation status analysis, load forecast results, inspection and maintenance plans, and safety production early warning information specifically includes: Based on the job permissions of the operators, determine the scope of accessible data and the level of operation instructions, and based on the scope of accessible data and the level of operation instructions, perform information filtering and information desensitization processing on the response results. Adjust the presentation format and level of detail of the response results based on the operating habits of the staff; the presentation formats include text summaries, visual charts, and step-by-step checklists. Based on the historical interaction records of operators, identify common problem types and concerns, and prioritize displaying highly relevant and timely warning and suggestion information in the response results; The customized response result is pushed to the terminal interface of the operator through the interaction hub, and voice broadcast and interactive confirmation are supported.

[0012] Based on the same inventive concept, this invention also provides a power generation enterprise management system based on large-scale model dialogue interaction, used to implement the aforementioned power generation enterprise management method based on large-scale model dialogue interaction, comprising: The data resource pool module is used to collect real-time operation data, equipment ledger data, procedure document data, historical case data, and cross-business module related data, and form a structured data resource pool with a unified format through data standardization processing and heterogeneous data adaptation and conversion. The interaction hub module, connected to the data resource pool module, is used to construct a basic training sample set based on the structured data resource pool, form an interaction hub with professional knowledge of the power generation industry through domain knowledge injection and fine-tuning, and deploy a vertical domain dialogue model for power generation enterprises based on a pre-trained large language model; the interaction hub is used to analyze the received requests in real time and generate responses based on the vertical domain dialogue model for power generation enterprises. The dialogue interaction entry module, connected to the interaction hub module, is used to provide a natural language interaction entry point, receive requests for intelligent Q&A, operation guidance, risk warning and decision suggestions initiated by operators in the form of voice or text, send the requests to the interaction hub module, and return the response results generated by the interaction hub module to the operators; The personalization module connects the interaction hub module and the dialogue interaction entry module. It is used to customize the response results output by the interaction hub module according to the operator's job permissions, operating habits and historical interaction records, so as to form personalized presentation content, and output it through the dialogue interaction entry module. The knowledge iteration and optimization module is connected to the data resource pool module and the interaction hub module respectively. It is used to process the new data and human feedback information received in real time by the interaction hub, drive the iteration of the knowledge graph in the power generation field, and optimize the parameters of the vertical domain dialogue model of the power generation enterprise based on the incremental training sample set.

[0013] Based on the same inventive concept, the present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the above-described power generation enterprise management method based on large-scale model dialogue interaction.

[0014] Based on the same inventive concept, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described power generation enterprise management method based on large-scale model dialogue interaction.

[0015] The technical solution of the present invention has at least the following advantages and beneficial effects: By integrating real-time operational data, equipment ledgers, procedural documents, historical cases, and cross-business module related data, a unified structured data resource pool is formed through standardization and heterogeneous adaptation. This solves the problem of isolated data and difficulty in joint analysis among business modules in the traditional model. The ability to call and collaboratively analyze cross-departmental data in real time enables equipment management, load forecasting, inspection and maintenance, and safe production to form a closed loop, providing complete data support for global management decisions.

[0016] Based on a pre-trained large model combined with domain knowledge injection and fine-tuning, a vertical domain dialogue model for power generation enterprises is constructed. This upgrades artificial intelligence technology from single-scenario application to full-process intelligent empowerment. The model can dynamically integrate multi-source data for real-time analysis and quickly output load forecast results, risk warning information, and scientific decision-making suggestions. It breaks through the response lag caused by traditional human experience-based decision-making and hierarchical reporting, and effectively adapts to the industry needs of complex source-grid-load-storage interaction and increased real-time control requirements after the grid connection of new energy.

[0017] With natural language interaction as the core entry point, it supports diverse interaction methods such as voice and text, replacing the complex operation logic of traditional professional human-machine interfaces and lowering the usage threshold for operators. At the same time, the personalized response mechanism based on job permissions, operating habits and historical interaction records allows personnel in different positions to quickly obtain customized equipment status analysis, operation guidance and inspection plans, improving work efficiency while reducing the learning and adaptation costs for new employees.

[0018] By receiving new operational data, equipment maintenance records, updated procedures and manual feedback in real time through the interactive hub, the knowledge graph is continuously iterated and model parameters are optimized. This solves the problems of traditional expert system knowledge bases relying on manual input and lagging updates. Professional knowledge, historical experience and the latest regulations in the power generation field can be dynamically accumulated and reused through the model, effectively solving the industry pain point of difficulty in the inheritance of personnel knowledge and experience, and ensuring the continuous accumulation of the company's core technologies and management experience. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the power generation enterprise management method based on large-model dialogue interaction, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a power generation enterprise management system based on large-model dialogue interaction according to an embodiment of the present invention. Detailed Implementation

[0020] The following is a detailed description of the embodiments, in conjunction with the accompanying drawings.

[0021] Reference Figure 1 A power generation enterprise management method based on large-scale model dialogue interaction includes the following steps: Step 1: Collect real-time operational data, equipment ledger data, procedure document data, historical case data, and cross-business module related data. Through data standardization processing and heterogeneous data adaptation and conversion, form a structured data resource pool with a unified format.

[0022] In some embodiments, the specific process of data standardization and heterogeneous data adaptation and conversion is as follows: Extract the raw data patterns from real-time operation data, equipment ledger data, procedure document data, historical case data, and cross-business module related data from real-time monitoring systems, equipment management systems, and document databases; Based on the pre-set unified data standard for power generation enterprises, the original data model is mapped and transformed, converting heterogeneous data sources with different formats and protocols into standardized data entries under the unified data model. Perform spatiotemporal alignment and semantic association on standardized data entries to establish links across data sources; Based on the unified data model and the aforementioned link relationships, a structured data resource pool is constructed and updated.

[0023] Step 2: Based on the structured data resource pool, construct a basic training sample set, and through domain knowledge injection and fine-tuning, form an interactive hub with professional knowledge of the power generation industry. Based on the pre-trained large language model, construct a vertical domain dialogue model for power generation enterprises.

[0024] In some embodiments, the specific process of constructing a vertical dialogue model for power generation enterprises is as follows: Based on a structured data resource pool, a basic training sample set is constructed. This set includes equipment operating status, operating procedures, fault cases, and cross-business related question-and-answer pairs. The structured data resource pool can be defined as follows: ;in, , , , and These represent operational data, equipment ledgers, procedural documents, historical cases, and cross-business related data, respectively. (Basic training sample set) This can be represented as a set of question-answer pairs: ;in, Indicates data source from structured data resource pool Natural language queries, such as "Current vibration value of Unit 1?"; This indicates the corresponding structured answer or text guidance; Represents the set of all queries; This represents the set of all answers.

[0025] The basic training sample set is integrated with a pre-defined power generation terminology database, safety procedure texts, and historical decision logs to extract domain entities, relationships, and rules, generating an enhanced domain knowledge representation. External knowledge sources can be defined as: ;in, , and These correspond to a terminology database, safety procedure texts, and historical decision logs, respectively. The enhanced domain knowledge representation is based on entity-relationship-rule triples, as shown in the following equation:

[0026] in, To enhance the representation of domain knowledge, it integrates basic data and external knowledge in a structured form; Represents a triple; From Extracted domain entities (such as "steam turbine" and "vibration threshold"); Relationships between entities (such as "located in", "has parameters"); This refers to the rule logic extracted from procedures and logs (such as "if the vibration is greater than X, then trigger an alarm").

[0027] A two-stage fine-tuning process is performed on the initial vertical domain dialogue model for power generation enterprises built based on a pre-trained large language model. The first stage uses instruction fine-tuning to adapt to natural language queries in the power generation domain. The second stage uses reinforcement learning alignment fine-tuning to optimize the accuracy, security, and operability of the model output. The performance of the fine-tuned initial vertical domain dialogue model for power generation enterprises was evaluated using a validation set, and the model parameters were adjusted based on the evaluation results to obtain the vertical domain dialogue model for power generation enterprises.

[0028] In some embodiments, the specific process of performing two-stage fine-tuning on the initial power generation enterprise vertical domain dialogue model constructed based on the pre-trained large language model is as follows: Based on the aforementioned basic training sample set, the initial power generation enterprise vertical domain dialogue model undergoes a first-stage instruction fine-tuning process. This process enables the initial power generation enterprise vertical domain dialogue model to understand and generate natural language queries and responses that conform to the professional standards of the power generation field. The pre-trained large language model can be set as follows: The initial vertical domain dialogue model is The parameters are The loss function for instruction fine-tuning is shown in the following equation:

[0029] in, This indicates the loss function used during the fine-tuning phase, whose goal is to maximize the log probability of the model generating the correct answer; Indicates that in a given query and enhance knowledge Under the condition that the parameter is Model generates answers The conditional probability; The optimization objective is shown in the following formula:

[0030] in, This represents the model parameters after fine-tuning by the first-stage instructions; Using the enhanced domain knowledge representation, a reward model for reinforcement learning is constructed; wherein, the reward model quantitatively evaluates the accuracy, security, and operability of the output of the initial power generation enterprise vertical domain dialogue model based on the power generation field safety regulations and operating procedures; Based on the model output after fine-tuning the first-stage instructions and the evaluation results of the reward model, a reinforcement learning algorithm is used for the second-stage alignment fine-tuning to optimize the model's decision-making logic and response quality in complex power generation business scenarios. A reward model can be constructed. The parameters are The evaluation criteria include: accuracy bonuses. Security rewards (based on ) and actionable rewards ; The total reward function is shown in the following formula:

[0031] in, , and These correspond to the weighting coefficients of the three rewards mentioned above.

[0032] The PPO algorithm is used for reinforcement learning optimization, and the objective function is shown in the following equation:

[0033] in, This represents the objective function used in the reinforcement learning phase (based on the PPO algorithm). Indicates the current model (parameters are) ) strategy; Indicates the strategy before the update (parameter is) ); This represents the advantage function, used to evaluate performance in a given query. Output the answer below Compared to the average level, the performance of reward-based models is superior. calculate; This is the pruning parameter in the PPO algorithm, used to limit the magnitude of policy updates; The final optimized parameters are shown in the following formula:

[0034] in, This represents the final model parameters after alignment and fine-tuning following the second stage of reinforcement learning; Use validation set Conduct an assessment. Let represent a sample in the validation set, where For the standard answer or expected answer, The total number of samples in the validation set; the performance metrics are shown in the following formula:

[0035] in, Representation Model Overall performance score; Indicates the model on the validation set On the accuracy indicators; This indicates that the model output conforms to safety regulations. Safety indicators; This represents an indicator of the fluency of the model's output answer. , and These correspond to the weighting coefficients of the three performance indicators mentioned above.

[0036] The results of the instruction fine-tuning and reinforcement learning alignment fine-tuning are integrated to output the vertical domain dialogue model of the power generation enterprise.

[0037] Step 3: Receive new operational data, equipment maintenance records, updated procedures, and human feedback from power generation companies in real time through the interactive hub, and complete the iteration of the knowledge graph in the power generation field and the parameter optimization of the vertical domain dialogue model for power generation companies.

[0038] In some embodiments, the specific process of completing the iteration of the knowledge graph in the power generation field and the parameter optimization of the vertical domain dialogue model of power generation enterprises is as follows: The newly added operational data, equipment maintenance records, procedure update files and manual feedback information received in real time by the interaction hub are standardized and adapted. The newly added heterogeneous data is converted into incremental data that conforms to the format compatible with the structured data resource pool. At the same time, key information fields are extracted from the incremental data, including equipment ID, timestamp, event type and feedback score. Based on incremental data and extracted entities, relationships, and rules in the power generation field, new domain entities are identified from the incremental data, entity attributes are updated, relationships between entities are supplemented, and conflict detection and fusion updates are performed on the existing knowledge graph to generate an iterative power generation domain knowledge graph. Building upon steps one and two, an initial power generation domain knowledge graph is constructed using a structured data resource pool and enhanced domain knowledge representation. Power generation domain entities encompass equipment (e.g., "No. 1 steam turbine," "feed water pump A"), parameters (e.g., "main steam pressure," "vibration amplitude"), operating conditions (e.g., "full load operation," "slip-up process"), regulations (e.g., "Regulations R1-Article 3"), fault codes (e.g., "F001 overheat alarm"), personnel roles (e.g., "shift leader," "inspector"), and departments (e.g., "operations department," "maintenance department"). Relationships are richly defined semantic relationships, such as "located in (equipment, plant)", "monitoring (sensor, parameter)", "having state (equipment, plant)". The graph includes elements such as "preparation, operating status", "referenced procedures (operating steps, procedure clauses)", "caused by (abnormal parameters, fault type)", and "belonging to (personnel, department)". Attributes are additional attributes of entities, such as the equipment's "nameplate parameters", "commissioning date", the parameter's "real-time value", "historical trend", and "safety threshold", and the procedure's "version number" and "effective date". Rules include embedding the logic from safety procedures and operating specifications into the graph in a computable form, such as "if the vibration value of equipment X is greater than the threshold Y and the duration is greater than Z minutes, then warning W is triggered". The initial power generation field knowledge graph serves as the static knowledge base for the interaction hub.

[0039] The following is an example of newly added data: New operational data: An alarm was received from the real-time monitoring system: "The high-pressure cylinder of turbine No. 1 is vibrating in the X direction. The current value is 7.8μm, which exceeds the warning threshold of 7.5μm (time: 2023-10-27 10:30:00)". Added equipment maintenance record: Entered from the equipment management system: "Online dynamic balance correction of No. 1 steam turbine was completed on 2023-10-26, and the reference vibration value was adjusted to 5.0μm after processing." New updated regulations: The Safety Supervision Department has released a new version of the "Abnormal Vibration Handling Regulations," which upgrades the response level for "vibration values ​​exceeding the warning threshold for 10 minutes" from "reminder" to "on-site inspection."

[0040] Added manual feedback information: After handling the above vibration alarm, operator Zhang San reported through the interactive hub: "This vibration alarm was found to be due to transient sensor interference, which has now been resolved. It is recommended to mark such short-term fluctuations as 'negligible'." Once the interaction hub receives new data in real time, the knowledge graph is iteratively updated according to the following process: The four types of heterogeneous data—operational data, equipment maintenance records, procedure update documents, and manual feedback information—were uniformly converted into incremental data entries. Key fields were extracted: For alarm data, the entities "No. 1 Steam Turbine" and "High-Pressure Cylinder Vibration X-Direction" were extracted, along with the attributes "Current Value = 7.8μm" and "Timestamp," and the relationship "Trigger (Vibration Value, Warning)." The existing entity "No. 1 Steam Turbine" was identified. Based on the maintenance record, the attribute "Vibration Reference Value" of this entity was updated from the historical value to "5.0μm". A new "Maintenance Event" entity was associated with it, and the relationship "Has experienced (No. 1 Steam Turbine, Dynamic Balance Correction)" was established. The alarm data is associated with the equipment entity to establish "Current Alarm (No. 1 Steam Turbine, High-Pressure Cylinder Vibration Alarm)"; at the same time, the system detects a potential conflict: the current vibration value (7.8μm) has exceeded the new benchmark (5.0μm) and triggered an early warning, but the manual feedback calls it "instantaneous sensor interference"; Update the time condition "lasting 10 minutes" in the new version of the procedure to the corresponding early warning rule logic in the graph; Add an attribute "confidence" to the potential causal relationship "sensor momentary interference caused vibration exceeding the limit", and initially set a high confidence level based on Zhang San's feedback; at the same time, associate "negligible" as a new "event label" entity with this alarm event.

[0041] Through the above iterative update process, an iterative knowledge graph for the power generation field is generated.

[0042] The incremental data is associated with the iterative knowledge graph of the power generation field to generate new question-answer pairs. The sample priority is marked by human feedback information to form an incremental training sample set. The incremental training sample set and the basic training sample set together constitute the data foundation for model optimization. Based on the incremental training sample set, the parameters of the vertical domain dialogue model of power generation enterprises are updated through an incremental fine-tuning strategy. The incremental fine-tuning strategy continues the logic of the two-stage fine-tuning. In the first stage, the instruction fine-tuning is performed for the newly added question-answer pairs to adapt to the natural language expression of incremental knowledge. In the second stage, reinforcement learning alignment fine-tuning is adopted based on human feedback information to optimize the model's response accuracy and operability to the new scenarios. The iterated knowledge graph is subjected to consistency verification, the power generation enterprise vertical domain dialogue model with optimized parameters is evaluated, and the verified knowledge graph and the evaluated model parameters are synchronized to the interaction hub.

[0043] Step 4: Based on the interaction hub, provide a natural language interaction entry point, which allows operators to initiate intelligent Q&A, operation guidance, risk warning and decision suggestion requests in the form of voice or text, so as to realize real-time access and collaborative analysis of cross-departmental business data.

[0044] In some embodiments, the real-time access and collaborative analysis of cross-departmental business data specifically includes: The system receives collaborative analysis requests initiated by operators through the interaction hub and parses the departmental roles, data categories, and analysis objectives involved in the requests. Based on the structured data resource pool and real-time data interface, cross-departmental data queries are initiated to the business systems corresponding to equipment management, safe production, and scheduling operations, and real-time or near-real-time business status data is obtained. By utilizing a vertical dialogue model for power generation companies, we can perform correlation analysis, conflict detection, and trend projection on the acquired cross-departmental business data, and generate collaborative analysis reports and decision-making recommendations. Collaborative analysis reports and decision recommendations are distributed to relevant departmental staff through an interactive hub, and support follow-up questions, corrections, and collaborative confirmations based on natural language, forming a closed-loop cross-departmental collaborative process.

[0045] Step 5: Output personalized response results through the interaction hub. Based on the operator's job authority, operating habits and historical interaction records, present customized equipment operation status analysis, load forecast results, inspection and maintenance plans and safety production early warning information.

[0046] In some embodiments, the customized presentation of equipment operation status analysis, load forecast results, inspection and maintenance plans, and safety production early warning information specifically includes: Based on the job permissions of the operators, determine the scope of accessible data and the level of operation instructions, and based on the scope of accessible data and the level of operation instructions, perform information filtering and information desensitization processing on the response results. Adjust the presentation format and level of detail of the response results based on the operating habits of the staff; the presentation formats include text summaries, visual charts, and step-by-step checklists. Based on the historical interaction records of operators, identify common problem types and concerns, and prioritize displaying highly relevant and timely warning and suggestion information in the response results; The customized response result is pushed to the terminal interface of the operator through the interaction hub, and voice broadcast and interactive confirmation are supported.

[0047] Based on the same inventive concept, and corresponding to any of the above embodiments, refer to... Figure 2 This invention provides a power generation enterprise management system based on large-model dialogue interaction, used to implement the aforementioned power generation enterprise management method based on large-model dialogue interaction, including: The data resource pool module is used to collect real-time operation data, equipment ledger data, procedure document data, historical case data, and cross-business module related data, and form a structured data resource pool with a unified format through data standardization processing and heterogeneous data adaptation and conversion. The interaction hub module, connected to the data resource pool module, is used to construct a basic training sample set based on the structured data resource pool, form an interaction hub with professional knowledge of the power generation industry through domain knowledge injection and fine-tuning, and deploy a vertical domain dialogue model for power generation enterprises based on a pre-trained large language model; the interaction hub is used to analyze the received requests in real time and generate responses based on the vertical domain dialogue model for power generation enterprises. The dialogue interaction entry module, connected to the interaction hub module, is used to provide a natural language interaction entry point, receive requests for intelligent Q&A, operation guidance, risk warning and decision suggestions initiated by operators in the form of voice or text, send the requests to the interaction hub module, and return the response results generated by the interaction hub module to the operators; The personalization module connects the interaction hub module and the dialogue interaction entry module. It is used to customize the response results output by the interaction hub module according to the operator's job permissions, operating habits and historical interaction records, so as to form personalized presentation content, and output it through the dialogue interaction entry module. The knowledge iteration and optimization module is connected to the data resource pool module and the interaction hub module respectively. It is used to process the new data and human feedback information received in real time by the interaction hub, drive the iteration of the knowledge graph in the power generation field, and optimize the parameters of the vertical domain dialogue model of the power generation enterprise based on the incremental training sample set.

[0048] Based on the same inventive concept, and corresponding to any of the above embodiments, the present invention provides an electronic device, including a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the power generation enterprise management method based on large-model dialogue interaction of the embodiments.

[0049] Alternatively, the aforementioned electronic device may be a server.

[0050] In addition, this embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the power generation enterprise management method based on large model dialogue interaction of the embodiment.

[0051] It is understood that the processor in the embodiments of the present invention may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. The general-purpose processor may be a microprocessor or any conventional processor.

[0052] The method steps in the embodiments of the present invention can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0053] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a storage medium or transmitted through a storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive (SSD)).

Claims

1. A power generation enterprise management method based on large-scale model dialogue interaction, characterized in that, Includes the following steps: Collect real-time operational data, equipment ledger data, procedure document data, historical case data, and cross-business module related data, and form a structured data resource pool with a unified format through data standardization processing and heterogeneous data adaptation and conversion; Based on a structured data resource pool, a basic training sample set is constructed. Through domain knowledge injection and fine-tuning, an interactive hub with professional knowledge of the power generation industry is formed. Based on a pre-trained large language model, a vertical domain dialogue model for power generation enterprises is constructed. By receiving new operational data, equipment maintenance records, updated procedures and manual feedback from power generation companies in real time through the interactive hub, the system completes the iteration of the knowledge graph in the power generation field and the parameter optimization of the vertical dialogue model of power generation companies. Based on the interaction hub, a natural language interaction entry point is provided, which allows operators to initiate intelligent Q&A, operation guidance, risk warning and decision suggestion requests in the form of voice or text, and realize the real-time call and collaborative analysis of cross-departmental business data; Personalized response results are output through the interactive hub, and equipment operation status analysis, load forecast results, inspection and maintenance plans and safety production early warning information are presented in a customized manner based on the operator's job authority, operating habits and historical interaction records.

2. The power generation enterprise management method based on large-scale model dialogue interaction as described in claim 1, characterized in that, The specific process of data standardization and heterogeneous data adaptation and conversion is as follows: Extract the raw data patterns from real-time operation data, equipment ledger data, procedure document data, historical case data, and cross-business module related data from real-time monitoring systems, equipment management systems, and document databases; Based on the pre-set unified data standard for power generation enterprises, the original data model is mapped and transformed, converting heterogeneous data sources with different formats and protocols into standardized data entries under the unified data model. Perform spatiotemporal alignment and semantic association on standardized data entries to establish links across data sources; Based on the unified data model and the aforementioned link relationships, a structured data resource pool is constructed and updated.

3. The power generation enterprise management method based on large-scale model dialogue interaction as described in claim 1, characterized in that, The specific process for constructing the vertical domain dialogue model for power generation enterprises is as follows: Based on the structured data resource pool, a basic training sample set is constructed, which includes equipment operating status, operating procedures, fault cases, and cross-business related question-and-answer pairs. The basic training sample set is integrated with a pre-set power generation terminology database, safety procedure texts, and historical decision logs to extract domain entities, relationships, and rules, generating an enhanced domain knowledge representation. A two-stage fine-tuning process is performed on the initial vertical domain dialogue model for power generation enterprises built based on a pre-trained large language model. The first stage uses instruction fine-tuning to adapt to natural language queries in the power generation domain. The second stage uses reinforcement learning alignment fine-tuning to optimize the accuracy, security, and operability of the model output. The performance of the fine-tuned initial vertical domain dialogue model for power generation enterprises was evaluated using a validation set, and the model parameters were adjusted based on the evaluation results to obtain the vertical domain dialogue model for power generation enterprises.

4. The power generation enterprise management method based on large-scale model dialogue interaction as described in claim 3, characterized in that, The specific process of fine-tuning the initial vertical domain dialogue model for power generation enterprises based on the pre-trained large language model is as follows: Based on the aforementioned basic training sample set, the initial power generation enterprise vertical domain dialogue model is fine-tuned in the first stage to enable the initial power generation enterprise vertical domain dialogue model to understand and generate natural language queries and responses that conform to the professional standards of the power generation field. Using the enhanced domain knowledge representation, a reward model for reinforcement learning is constructed; wherein, the reward model quantitatively evaluates the accuracy, security, and operability of the output of the initial power generation enterprise vertical domain dialogue model based on the power generation field safety regulations and operating procedures; Based on the model output after fine-tuning the first-stage instructions and the evaluation results of the reward model, a reinforcement learning algorithm is used for the second-stage alignment fine-tuning to optimize the model's decision-making logic and response quality in complex power generation business scenarios. The results of the instruction fine-tuning and reinforcement learning alignment fine-tuning are integrated to output the vertical domain dialogue model of the power generation enterprise.

5. The power generation enterprise management method based on large-scale model dialogue interaction as described in claim 3, characterized in that, The specific process for completing the iteration of the knowledge graph in the power generation field and the parameter optimization of the vertical domain dialogue model for power generation enterprises is as follows: The newly added operational data, equipment maintenance records, procedure update files and manual feedback information received in real time by the interaction hub are standardized and adapted. The newly added heterogeneous data is converted into incremental data that conforms to the format compatible with the structured data resource pool. At the same time, key information fields are extracted from the incremental data, including equipment ID, timestamp, event type and feedback score. Based on incremental data and extracted entities, relationships and rules in the power generation field, new domain entities are identified from the incremental data, entity attributes are updated, relationships between entities are supplemented, and conflict detection and fusion updates are performed on the existing knowledge graph to generate an iterative power generation domain knowledge graph. The incremental data is associated with the iterative knowledge graph of the power generation field to generate new question-answer pairs. The sample priority is marked by human feedback information to form an incremental training sample set. The incremental training sample set and the basic training sample set together constitute the data foundation for model optimization. Based on the incremental training sample set, the parameters of the vertical domain dialogue model of power generation enterprises are updated through an incremental fine-tuning strategy. The incremental fine-tuning strategy continues the logic of the two-stage fine-tuning. In the first stage, the instruction fine-tuning is performed for the newly added question-answer pairs to adapt to the natural language expression of incremental knowledge. In the second stage, reinforcement learning alignment fine-tuning is adopted based on human feedback information to optimize the model's response accuracy and operability to the new scenarios. The iterated knowledge graph is subjected to consistency verification, the power generation enterprise vertical domain dialogue model with optimized parameters is evaluated, and the verified knowledge graph and the evaluated model parameters are synchronized to the interaction hub.

6. The power generation enterprise management method based on large-scale model dialogue interaction as described in claim 1, characterized in that, The implementation of real-time access and collaborative analysis of cross-departmental business data specifically includes: The system receives collaborative analysis requests initiated by operators through the interaction hub and parses the departmental roles, data categories, and analysis objectives involved in the requests. Based on the structured data resource pool and real-time data interface, cross-departmental data queries are initiated to the business systems corresponding to equipment management, safe production, and scheduling operations, and real-time or near-real-time business status data is obtained. By utilizing a vertical dialogue model for power generation companies, we can perform correlation analysis, conflict detection, and trend projection on the acquired cross-departmental business data, and generate collaborative analysis reports and decision-making recommendations. Collaborative analysis reports and decision recommendations are distributed to relevant departmental staff through an interactive hub, and support follow-up questions, corrections, and collaborative confirmations based on natural language, forming a closed-loop cross-departmental collaborative process.

7. The power generation enterprise management method based on large-scale model dialogue interaction as described in claim 1, characterized in that, The customized presentation of equipment operation status analysis, load forecast results, inspection and maintenance plans, and safety production early warning information specifically includes: Based on the job permissions of the operators, determine the scope of accessible data and the level of operation instructions, and based on the scope of accessible data and the level of operation instructions, perform information filtering and information desensitization processing on the response results. Adjust the presentation format and level of detail of the response results based on the operating habits of the staff; the presentation formats include text summaries, visual charts, and step-by-step checklists. Based on the historical interaction records of operators, identify common problem types and concerns, and prioritize displaying highly relevant and timely warning and suggestion information in the response results; The customized response result is pushed to the terminal interface of the operator through the interaction hub, and voice broadcast and interactive confirmation are supported.

8. A power generation enterprise management system based on large-model dialogue interaction, used to implement the power generation enterprise management method based on large-model dialogue interaction as described in any one of claims 1-7, characterized in that, include: The data resource pool module is used to collect real-time operation data, equipment ledger data, procedure document data, historical case data, and cross-business module related data, and form a structured data resource pool with a unified format through data standardization processing and heterogeneous data adaptation and conversion. The interaction hub module, connected to the data resource pool module, is used to construct a basic training sample set based on the structured data resource pool, form an interaction hub with professional knowledge of the power generation industry through domain knowledge injection and fine-tuning, and deploy a vertical domain dialogue model for power generation enterprises based on a pre-trained large language model; the interaction hub is used to analyze the received requests in real time and generate responses based on the vertical domain dialogue model for power generation enterprises. The dialogue interaction entry module, connected to the interaction hub module, is used to provide a natural language interaction entry point, receive requests for intelligent Q&A, operation guidance, risk warning and decision suggestions initiated by operators in the form of voice or text, send the requests to the interaction hub module, and return the response results generated by the interaction hub module to the operators; The personalization module connects the interaction hub module and the dialogue interaction entry module. It is used to customize the response results output by the interaction hub module according to the operator's job permissions, operating habits and historical interaction records, so as to form personalized presentation content, and output it through the dialogue interaction entry module. The knowledge iteration and optimization module is connected to the data resource pool module and the interaction hub module respectively. It is used to process the new data and human feedback information received in real time by the interaction hub, drive the iteration of the knowledge graph in the power generation field, and optimize the parameters of the vertical domain dialogue model of the power generation enterprise based on the incremental training sample set.

9. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to cause the electronic device to perform the power generation enterprise management method based on large-model dialogue interaction as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the power generation enterprise management method based on large model dialogue interaction as described in any one of claims 1-7.

Images