Thermal power plant equipment fault diagnosis measuring point screening method, system, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence and thermal power production, in particular to a thermal power plant equipment fault diagnosis measuring point screening method and system, equipment and a medium, and the method comprises the steps: obtaining the measuring point data of all production equipment in a thermal power plant when the production equipment of the thermal power plant breaks down; inputting the measuring point data into a trained measuring point screening model, and screening out measuring points related to fault points by using the measuring point screening model; measuring point screening suggestions and supplementary measuring point types are generated, and fault equipment is diagnosed by using the screened measuring points, the measuring point screening suggestions and the supplementary measuring point types; and the measuring point screening model adopts a large language model optimized by a fine tuning or RAG method. According to the method, fault related measuring points can be quickly and accurately screened out.

Description

Technical Field

[0001] The present invention relates to the technical fields of artificial intelligence and thermal power production, and specifically relates to a method, system, device, and medium for screening measurement points for equipment fault diagnosis in thermal power plants. Background Art

[0002] During the process of fault diagnosis in thermal power production, it usually relies on manual reasoning and judgment by combining the actual situation at the fault site after the fault occurs, and then obtaining solutions and suggestions for prevention and control measures. In recent years, the application of deep learning time series anomaly detection models has begun to bring benefits to the power generation industry, but the time series model needs to rely on fault-related measurement point data for prediction. The selection result and quality of measurement points have a great impact on the judgment result of the deep learning time series anomaly detection model. However, the on-site unit structure is complex, the types of equipment and measurement points are numerous, there are differences in the naming methods of measurement points in different power plants, and the fault types and occurrence locations usually require professional knowledge and long-term experience accumulation to make accurate judgments. If measurement points are blindly selected to troubleshoot faults when the artificial professional experience is insufficient or the thermal power field experts cannot arrive at the scene immediately, situations such as misjudgment and missed detection are likely to occur, resulting in losses of time or economic benefits. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method, system, device, and medium for screening measurement points for equipment fault diagnosis in thermal power plants to solve the technical problems of low efficiency and low accuracy in fault diagnosis of thermal power plant production equipment in view of the above-mentioned prior art.

[0004] The object of the present invention is achieved by the following technical solutions: In a first aspect, the present invention provides a method for screening measurement points for equipment fault diagnosis in thermal power plants, including: When a fault occurs in the production equipment of a thermal power plant, obtaining the measurement point data of all production equipment in the thermal power plant; Inputting the measurement point data into a trained measurement point screening model, and using the measurement point screening model to screen out the measurement points related to the fault point; and generating measurement point screening suggestions and supplementary measurement point types, and diagnosing the faulty equipment by using the screened measurement points, measurement point screening suggestions, and supplementary measurement point types; The measurement point screening model is a large language model optimized by the fine-tuning or RAG method.

[0005] As a further improvement of the present invention, the data set for training the measurement point screening model includes: corpus data from various sources such as a large number of historical fault cases in the thermal power field, expert analysis reports, maintenance manuals, equipment operation logs, power generation industry technical manuals, operation guides, safety regulations, engineering and technical papers, industry standards and specifications, etc.

[0006] As a further improvement of the present invention, the large language model uses an open-source large language model, and the open-source large language model includes Qwen2, Qwen2.5, and DeepSeek.

[0007] As a further improvement of the present invention, the way of fine-tuning the measurement point screening model adopts a low-load fine-tuning method; the low-load fine-tuning method includes the LOMO or LoRA method; The LOMO method specifically includes: during the fine-tuning process of the measurement point screening model, the LOMO optimizer is used to fuse gradient calculation and parameter update.

[0008] The LoRA method specifically includes: during the fine-tuning process of the measurement point screening model, a trainable low-rank matrix is added to the weights of the measurement point screening model to achieve fine-tuning.

[0009] As a further improvement of the present invention, the RAG method includes: By performing vector quantization encoding on the expert knowledge data, the embedding model is used to convert the vectorized data into a high-dimensional vector representation and construct an efficiently retrievable vector database; When the query statement input to the measurement point screening model, the system will also vectorize it, match the semantically related paragraphs or rules from the vector library based on algorithms such as cosine similarity, and finally perform weighted ranking on the retrieval results in combination with the logic of the domain knowledge base to generate accurate measurement point screening suggestions.

[0010] As a further improvement of the present invention, in the step of generating the measurement point screening suggestions, the measurement point screening model also performs at least one of the following functions: Generate an explanatory text for the measurement point type and name corresponding to the fault phenomenon; Integrate multi-source knowledge base data and infer the correlation between measurement point anomalies and fault types; Provide a troubleshooting priority suggestion based on historical cases.

[0011] As a further improvement of the present invention, it also includes feedback optimization of the measurement point screening model; the feedback optimization specifically includes: The user inputs feedback data into the measurement point screening model, adjusts the measurement point association weight and inference logic according to the feedback data, and then dynamically adjusts the output measurement point screening suggestions and supplementary measurement point types; The feedback data includes the matching degree between the measurement point screening result and the actual fault.

[0012] In the second aspect, the present invention provides a measurement point screening system for fault diagnosis of thermal power plant equipment, which is used to implement the above-mentioned measurement point screening method for fault diagnosis of thermal power plant equipment, and includes: A data acquisition module, when a fault occurs in the production equipment of the thermal power plant, acquires the measurement point names of all the production equipment related to the fault in the thermal power plant; A measuring point screening module, the measuring point screening module includes a measuring point screening model, and the measuring point screening model is used to screen out the measuring points related to the fault point according to the input measuring point name data, so as to obtain the measuring point data and use the screened measuring point data to diagnose the faulty equipment; The measuring point screening model adopts a large language model optimized by the fine-tuning or RAG method.

[0013] In a third aspect, the present invention provides a computer-readable storage medium storing one or more programs, the one or more programs include instructions, and when the instructions are executed by a computing device, the computing device is caused to execute the above-mentioned method for screening measuring points for fault diagnosis of thermal power plant equipment.

[0014] In a fourth aspect, the present invention provides a computing device, including: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include steps for executing the above-mentioned method for screening measuring points for fault diagnosis of thermal power plant equipment.

[0015] The beneficial effects of the present invention are as follows: The method for screening measuring points for fault diagnosis of thermal power plant equipment provided by the present invention is mainly used before fault diagnosis. Through the measuring point screening model, the measuring points related to the fault point or faulty equipment can be quickly and accurately screened out, and then the screened measuring points can be further diagnosed in a targeted manner. The present invention uses the trained measuring point screening model to quickly and accurately screen out the measuring points related to the fault point from a large amount of measuring point data. This avoids the cumbersome process of manual screening and possible omissions or misjudgments, and improves the accuracy of diagnosis. Through efficient measuring point screening, the fault point can be quickly located, the fault diagnosis time can be shortened, the equipment downtime can be reduced, and the production efficiency can be improved. Through the fine-tuning or RAG method, the model of the present invention can continuously absorb new data and experience and continuously optimize its own diagnostic ability. This model can not only screen out relevant measuring points, but also generate measuring point screening suggestions to help engineers more comprehensively understand the possible influence range and potential causes of the fault. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a schematic flow chart of the method for screening measuring points for fault diagnosis of thermal power plant equipment in the embodiment of the present invention; Figure 2 It is a schematic structural diagram of the fault diagnosis measurement point screening system for thermal power plant equipment in an embodiment of the present invention; Figure 3 It is a schematic structural diagram of an electronic device in an embodiment of the present invention. Detailed implementation manners

[0018] In order to make the objectives and technical solutions of the present invention clearer and easier to understand, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings and specific embodiments. Among them, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments.

[0020] Embodiment 1 As Figure 1 shown, this embodiment provides a method for screening fault diagnosis measurement points of thermal power plant equipment. By combining the large language model with the expert knowledge in the thermal power production field and utilizing the powerful reasoning and knowledge integration capabilities of the large language model, the measurement points related to faults are effectively screened out from a large number of measurement points, and professional diagnosis suggestions are generated. The following are the specific implementation manners.

[0021] First, when a fault occurs in the production equipment of the thermal power plant, obtain the names of the measurement points of all production equipment that may be related to the fault in the thermal power plant.

[0022] In this embodiment, the measurement point data of the production equipment includes various types, mainly including temperature, humidity, hydrogen leakage amount of the generator enclosed bus, three-phase circuit of the generator, vibration, flow rate, pressure, and the measurable operation data related to each equipment are all measurement point data. Since the measurement point data of the production equipment is relatively large and of various types, this embodiment also conducts a preliminary screening manually to screen out the measurement points that are completely irrelevant to the faulty equipment, thereby reducing the screening time cost.

[0023] Then, in this embodiment, the measurement point screening model quickly and accurately screens out the measurement points related to the fault from the massive measurement point data. In this embodiment, the measurement point screening model uses a domestic open-source large language model. And the large language model uses any one of qwen2, qwen2.5, and DeepSeek. For example, this embodiment selects qwen2.5 as the measurement point screening model.

[0024] Specifically, input the measurement point names into the trained measurement point screening model, use the measurement point screening model to screen out the measurement points related to the fault point; and generate measurement point screening suggestions and supplementary measurement point types, and use the screened measurement points, measurement point screening suggestions and supplementary measurement point types to diagnose the faulty equipment; Among them, the fine-tuning dataset and RAG corpus for training the measurement point screening model include corpus data from multiple sources such as a large number of historical fault cases, expert analysis reports, maintenance manuals, equipment operation logs, power generation industry technical manuals, operation guides, safety regulations, engineering and technical papers, industry standards and specifications in the thermal power field. Among them, historical fault cases refer to extracting the time, type, and corresponding solutions of equipment failures from historical fault records. To ensure the data quality in the dataset, all the collected data is standardized, cleaned, and labeled.

[0025] During the training process, the measurement point screening model is fine-tuned using the low-load fine-tuning method; the low-load fine-tuning method includes the LOMO or LoRA method.

[0026] The LOMO method specifically includes: during the fine-tuning process of the measurement point screening model, the LOMO optimizer is used to fuse gradient calculation and parameter update. In this embodiment, the backpropagation mechanism is used to update the parameters. During the backpropagation process, parameter update is immediately executed after each calculation layer completes gradient calculation, avoiding storing intermediate gradients, reducing video memory occupancy, and immediately releasing the intermediate variables of this layer after the update.

[0027] The LoRA method specifically includes: during multiple rounds of training of the measurement point screening model, a trainable low-rank matrix is added to the weights of the measurement point screening model to achieve fine-tuning. On the original weight matrix of the measurement point screening model, a low-rank matrix is introduced. During the training process, the size of the rank is adjusted using the progressive rank growth method. Using an open-source large language model as the basis, a low-rank matrix is added beside the weight matrix of the pre-trained model, and these low-rank matrices are trained to adapt to downstream tasks while keeping the original weights unchanged, thereby reducing the number of trainable parameters, saving computing resources, and avoiding overfitting at the same time. Further optimization is carried out through low-load fine-tuning. By fine-tuning the large language model in combination with the expert knowledge base in the power generation industry under limited computing resources, it can be efficiently applied in the thermal power production scenario, effectively improving the accuracy and professionalism of the model in the fault diagnosis of thermal power equipment.

[0028] The RAG method specifically includes: By vectorizing and encoding expert knowledge data, using an embedding model to convert the vectorized data into a high-dimensional vector representation and constructing a vector database that can be efficiently retrieved; When the query statement input to the measurement point screening model is received, the system will also vectorize it, match the semantically related paragraphs or rules from the vector library based on algorithms such as cosine similarity, and finally perform weighted sorting on the retrieval results in combination with the logic of the domain knowledge base to generate accurate measurement point screening suggestions.

[0029] Specifically, RAG technology is used to process data through data loading, document segmentation, vectorization, and data storage. Based on the preparations in the above steps, the received user query statement is vectorized and then the vector database is searched for knowledge text or historical conversation records that are semantically similar to the question vector to complete the RAG process.

[0030] Furthermore, in the step of generating the measuring point screening suggestion, the measuring point screening model also performs at least one of the following functions: Generate explanatory text of the measurement point type and name corresponding to the fault phenomenon; Integrate multi-source knowledge base data to infer the correlation between measurement point anomalies and fault types; Provide troubleshooting priority suggestions based on historical cases.

[0031] In this embodiment, the large language model interacts with the user to implement dynamic adjustment of screening suggestions. The large language model receives the natural language fault description and existing measurement point information input by the user, parses the key fault characteristics through natural language processing, and generates standardized input. The large language model allows users to explore the potential relationship between the measurement point data through interactive queries and dynamically adjust the screening suggestions. Use the analysis program to generate natural language to interact with the large model to describe the faults occurring on site and all existing measurement point information. Based on the fault description and existing measurement points input by the user, the large model will analyze the fault characteristics, intelligently filter out the measurement points associated with the fault phenomenon, and provide users with selection suggestions for the locations and measurement point types where the fault may occur. It is convenient for users to more efficiently identify the measurement points related to the fault and to improve and supplement the existing measurement points.

[0032] Use the analysis program to generate natural language to interact with the system, describing the abnormal operation or failure of the equipment and the names of the existing measurement points collected by the user. The measurement point screening model first processes the user's input and converts it into a standard format that can be analyzed by the model. The specific operation process is as follows: Input content: Users can describe the operating status of the equipment (such as "generator hydrogen leakage high alarm") or fault characteristics (such as "the hydrogen leakage of the generator stator cooling water tank has reached 20%~25%").

[0033] Input parsing: Through the natural language processing module, key fault characteristics, equipment types and related measurement point information in the input are identified, and key elements in the user description are automatically marked.

[0034] Measuring point screening suggestions: Based on the description entered by the user, after analysis by the large language model, fault-related measuring point screening suggestions are output. For example, when the user enters "the hydrogen leakage of the generator stator cooling water tank has reached 20%", the system may suggest checking the measuring points related to the hydrogen and stator cooling water system, such as the hydrogen leakage of the generator stator cooling water tank, the hydrogen pressure of the generator, the hydrogen purity of the generator, and other data.

[0035] This embodiment also includes feedback optimization of the measurement point screening model; the feedback optimization specifically includes: the user inputs feedback data into the measurement point screening model, adjusts the measurement point association weight and inference logic according to the feedback data, and then dynamically adjusts the output measurement point screening suggestions and supplementary measurement point types; the feedback data includes the matching degree between the measurement point screening result and the actual fault. Specifically, in actual applications, after the user operates according to the system's fault diagnosis and troubleshooting suggestions, the result can be fed back to the system. The system further optimizes the model by collecting user feedback data. The specific methods include: Feedback data collection: Record the matching degree between the diagnosis result after measurement point screening by the user in actual applications and the actual fault type, as well as the effectiveness of the troubleshooting suggestions.

[0036] Model adjustment: According to user feedback, adjust the inference mechanism of the large language model to optimize the accuracy of fault judgment and measurement point screening.

[0037] The large language model can integrate knowledge from multiple data sources (such as equipment manuals, historical fault records, expert reports, etc.) and conduct in-depth reasoning based on this information to help analyze the relevant measurement points of the factors leading to abnormal equipment phenomena. For example, the model can explain the correlation between abnormal changes in certain measurement points and specific fault types, thus providing more accurate measurement point selection for maintenance personnel. After the large language model analyzes the time series data and identifies potential anomalies, it provides targeted troubleshooting measurement point suggestions and guidance to maintenance personnel based on historical data and similar cases. It helps maintenance personnel make more accurate and scientific judgments when facing complex fault phenomena, and avoid misselection or omission of measurement points caused by insufficient experience or incomplete data in traditional methods.

[0038] Embodiment 2 As a preferred embodiment in this embodiment, the application of this measurement point screening model is described in detail with an example. The specific implementation steps include: First, obtain the fault-related records of the generator and the existing measurement point data. For example, in the generator fault record of a power plant, the following key information is included: Measurement point 20:35:57.202 NCS system came "#4 generator transformer protection set B-protection start", "#4 generator transformer protection set B-stator grounding protection start", measurement point 20:35:57.728 "#4 generator transformer protection set B-protection start", "#4 generator transformer protection set B-stator grounding protection start" reset, on-site inspection found that the three-phase voltage at the #4 machine end fluctuated instantly, because the protection start duration did not reach 1s (generator stator grounding protection setting value-zero sequence voltage protection delay 1s), the protection did not operate, contact maintenance electrical inspection and processing. Measurement point 20:40 #4 generator hydrogen pressure 331Kpa. At 20:49, the duty officer found that the hydrogen gas in the #4 generator dropped rapidly from 331Kpa to 310Kpa, and the generator gave a "generator hydrogen leakage high alarm" and reported to the duty officer. The duty officer immediately sent patrol personnel to check the hydrogen system of the #4 generator on site, arranged personnel to make emergency hydrogen replenishment, and checked the hydrogen system, sealing oil system, and cooling water system of the #4 machine at the same time. At the same time, personnel were sent to monitor the hydrogen concentration on site to check the location of the hydrogen leakage. At 20:55, the duty officer found that the liquid level of the cooling water tank of the #4 machine dropped from 528mm to 480mm, and the speed of decline became faster. The patrol personnel were ordered to check the cooling water system of the #4 machine. The patrol feedback showed that there was no leakage on site; personnel were immediately arranged to check the oil-water detector. The patrol personnel reported that water was discharged from the oil-water detector in the middle of the #4 generator. Subsequently, the patrol personnel reported that water was discharged from the oil-water detector at the steam end of the #4 generator and the bottom sewage of the #4 generator hydrogen bus. At the same time, #4 gave an alarm of "liquid leakage in the middle shell of the generator". Pre-disk inspection found that the hydrogen leakage of the generator stator cooling water tank had reached 20% (about 0.08% in normal operation). The generator hydrogen dew point monitor showed a dew point temperature of 38.6℃ (about -17℃ in normal operation). Comprehensively judged that there was a water leak in the stator cooling water pipeline inside the generator. At 21:30:31, the provincial dispatching #4 generator was immediately informed that water was entering the internal part of the generator and an emergency shutdown was required. Then the #4 unit was shut down. At 21:32, the #4 generator was urgently discharged of hydrogen and gas replacement was performed. After the temperature of the steam inlet volute of the high and medium pressure cylinder of the steam turbine cooled to 150℃, the turntable was stopped, the sealing oil system was stopped, and the generator cover was inspected. The estimated construction period is 7-10 days. The basis for the unit shutdown: According to the hydrogen leakage monitoring data of the cold water tank, it should be based on the situation when no water replenishment and drainage are carried out and the water tank liquid level is stable. When the hydrogen content (volume content) exceeds 2%, an alarm should be sounded, and the monitoring of the generator should be strengthened. If it exceeds 10%, the generator should be shut down immediately to eliminate the fault. Article 10.5.3.7 When a water-cooled unit issues a water leakage alarm signal and is confirmed to have an internal water leakage, the unit should be shut down immediately for processing.

[0039] Based on the analysis of the acquired data, the measuring point information related to generator faults is extracted. For example, the existing measuring points include: hydrogen leakage amount at the neutral point of the generator enclosed busbar 1, hydrogen leakage amount at the neutral point of the generator enclosed busbar 2, flue gas temperature at the inlet of the B-side reactor rectifier, flue gas temperature at the inlet of the A-side reactor rectifier, hydrogen leakage amount of phase A of the generator enclosed busbar, hydrogen leakage amount of phase B of the generator enclosed busbar, hydrogen leakage amount of phase C of the generator enclosed busbar, return water temperature of the generator stator cooling water, temperature of the generator stator cooling water tank, liquid level of the generator stator cooling water tank, excitation current of unit 4, temperature 1 of the #1 rectifier cabinet of the excitation system of the #3 generator-transformer unit, phase A current of generator 4, phase B current of generator 4, phase C current of generator 4, relative vibration of bearing 1 in the X direction. The above measuring points have exceeded the measuring point threshold very early.

[0040] Based on the analysis results, key measuring points, supplementary measuring points and other relevant measuring points are selected.

[0041] For example, the key measuring points include: Hydrogen leakage amount at the neutral point of the generator enclosed busbar: including "hydrogen leakage amount at the neutral point of the generator enclosed busbar 1" and "hydrogen leakage amount at the neutral point of the generator enclosed busbar 2", which are used to monitor the hydrogen leakage at the neutral point of the generator enclosed busbar and timely detect potential hydrogen leakage problems.

[0042] Return water temperature of the generator stator cooling water, which is used to monitor the return water temperature of the generator stator cooling water. Abnormal temperature changes may indicate cooling system failures or internal abnormalities of the generator.

[0043] Liquid level of the generator stator cooling water tank, which is used to monitor the liquid level of the generator stator cooling water tank. An abnormal decrease in the liquid level may indicate a leak in the cooling water system.

[0044] Alarm signal for liquid leakage in the middle shell of the generator: used to timely detect liquid leakage in the middle shell of the generator.

[0045] Temperature of the generator stator cooling water tank: used to monitor the temperature of the generator stator cooling water tank.

[0046] The supplementary measuring points include: Hydrogen leakage amounts of the three phases of the generator enclosed busbar: used to supplement the monitoring of hydrogen leakage in the three phases of the generator enclosed busbar and further improve hydrogen leakage monitoring.

[0047] Data displayed by the hydrogen dew point monitor of the generator: used to monitor the dew point temperature of the hydrogen in the generator. Abnormal dew point temperatures may indicate problems in the hydrogen system.

[0048] Monitoring of hydrogen pressure of the generator in front of the panel: used to monitor the hydrogen pressure of the generator. Abnormal pressure changes help to timely detect hydrogen system failures.

[0049] The other relevant measuring points include: The liquid level of the stator cooling water tank of a certain generator - used to detect possible water leakage; The excitation current of a certain group - monitor the status of the excitation system: used to monitor the status of the excitation system. Abnormal excitation current may affect the normal operation of the generator.

[0050] The three-phase current of the generator under test: used to monitor the three-phase current of the generator. Abnormal current may indicate that there is a fault in the generator.

[0051] Non-critical measuring points: The flue gas temperature at the inlet of the reactor rectifier: related to the excitation system, but not directly related in this fault analysis, and can be used as a non-critical measuring point.

[0052] The flue gas temperature at the inlet of the reactor rectifier on the A side - these are related to the excitation system, but not directly related in this fault analysis.

[0053] Comprehensive suggestions: In order to ensure comprehensive monitoring of the generator status and timely detection of potential problems, the following key measuring points can be focused on: the hydrogen leakage amount at the neutral point 1 of the generator enclosed bus; the hydrogen leakage amount at the neutral point 2 of the generator enclosed bus; the return water temperature of the generator stator cooling water; the liquid level of the generator stator cooling water tank; the hydrogen leakage amount of phase A of the generator enclosed bus; the hydrogen leakage amount of phase B of the generator enclosed bus; the hydrogen leakage amount of phase C of the generator enclosed bus; the excitation current of a certain unit; the three-phase current of the generator.

[0054] Through the above measuring point screening method and system, it is possible to effectively screen out the key measuring points related to the generator fault detection from numerous measuring points, improve the accuracy and efficiency of fault detection, and ensure the safe and stable operation of the generator.

[0055] Embodiment 3 As Figure 2 shown, this embodiment provides a measuring point screening system for fault diagnosis of thermal power plant equipment, which is used to implement the measuring point screening method for fault diagnosis of thermal power plant equipment in the above Embodiment 1. The system includes: A data acquisition module, which acquires the measuring point data of all production equipment in the thermal power plant when a fault occurs in the production equipment of the thermal power plant; A measuring point screening module, the measuring point screening module includes a measuring point screening model, and the measuring point screening model is used to screen out the measuring points related to the fault point according to the input measuring point data; use the screened measuring points to diagnose the faulty equipment.

[0056] Embodiment 4 In one embodiment of the present invention, a computer-readable storage medium is provided. This medium belongs to the memory device of a terminal device and is mainly used to store programs and data. The computer-readable storage medium includes both the storage medium built into the terminal and the extended storage medium supported by the terminal. Specifically, any tangible medium that can store a program and be used by an instruction execution system, apparatus, or device belongs to this category. This storage medium provides storage space for storing the terminal operating system and instructions (including one or more computer programs and their codes) that can be loaded and executed by a processor. Examples include electrically connected devices, portable disks, hard disks, RAM, ROM, EPROM / flash memory, optical fibers, CD-ROMs, optical storage devices, magnetic storage devices, etc., and combinations thereof.

[0057] In addition, the computer-readable storage medium also relates to data signals propagated in a baseband or as a carrier wave, which carry readable program codes and can be in the form of electromagnetic signals, optical signals, etc. The readable storage medium is not limited to the above types and also includes other media that can send, propagate, or transmit a program for use by an instruction execution system, apparatus, or device. The program code can be transmitted by wireless, wired, optical cable, RF, etc.

[0058] The program code can be written in various programming languages, such as object-oriented languages (Python, Java, C++, etc.) and procedural languages (C language, etc.). The code can be executed completely or partially on a user device, or can be used as an independent software package, or be executed partially / fully on a remote device. The remote device is connected to the user device through a LAN, WAN, or Internet service provider.

[0059] One or more instructions stored in the computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the method for screening fault diagnosis measurement points of thermal power plant equipment in the above embodiment; one or more instructions in the computer-readable storage medium are loaded and executed by a processor to perform the following steps: When a fault occurs in the production equipment of a thermal power plant, obtain the measurement point data of all production equipment in the thermal power plant; Input the measurement point data into a trained measurement point screening model, and use the measurement point screening model to screen out the measurement points related to the fault point; and generate measurement point screening suggestions and supplementary measurement point types, and use the screened measurement points, measurement point screening suggestions, and supplementary measurement point types to diagnose the faulty equipment; The measurement point screening model uses a large language model optimized by the fine-tuning or RAG method.

[0060] Embodiment 5 As Figure 3As shown, the terminal device in this embodiment is a computer device 60, which mainly includes a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and running on the processor 61. When the computer program 63 is executed by the processor 61, the method for screening fault diagnosis measurement points of thermal power plant equipment can be implemented, or the functions of each model / unit of the calculation system of the fault diagnosis model in Embodiment 1 can be implemented, which will not be elaborated here.

[0061] The types of computer devices 60 are relatively diverse, including desktop computers, notebooks, palm computers, cloud servers, etc. Its components are not limited to the processor 61 and the memory 62, but may also include input / output devices, network access devices, buses, etc. Figure 3 Only as an example, the number and types of components of the actual device may vary.

[0062] The processor 61 can be a central processing unit (CPU), or other general-purpose processors, graphics processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) and other programmable logic devices, discrete gates, transistor logic devices, or even data processing logic devices based on quantum computing, discrete hardware components, etc., or can also be a conventional microprocessor.

[0063] Regarding the memory 62, it can be either an internal storage unit of the computer device 60, such as a hard disk or memory, or an external storage device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD card), flash card (FlashCard), etc. The memory 62 can also include both an internal storage unit and an external storage device, for storing computer programs, other required programs and data, and temporarily storing the data that has been output or to be output.

[0064] In various embodiments of the present application, the mentioned memories, databases, or other media cover non-volatile and volatile memories. Non-volatile memories include read-only memories (ROMs), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random-access memories (ReRAMs), magnetoresistive random-access memories (MRAMs), ferroelectric memories (FRAMs), phase-change memories (PCMs), graphene memories, etc.; volatile memories include random-access memories (RAMs) or external cache memories, etc., and RAM can be further divided into various forms such as static random-access memories (SRAMs) and dynamic random-access memories (DRAMs).

Claims

1. A method for screening measuring points for equipment fault diagnosis in thermal power plants, characterized in that: include: When a failure occurs in a thermal power plant's production equipment, obtain the measurement point data of all production equipment in the thermal power plant; The measuring point data is input into a trained measuring point screening model, and the measuring point screening model is used to screen out the measuring points related to the fault point; and a measuring point screening suggestion and a supplementary measuring point type are generated, and the faulty equipment is diagnosed using the screened measuring points, the measuring point screening suggestion and the supplementary measuring point type; The test point screening model adopts a large language model optimized by fine-tuning or RAG method.

2. The method for screening measuring points for equipment fault diagnosis in thermal power plants according to claim 1, characterized in that: The data sets used to train the measurement point screening model include: historical failure cases in the thermal power field, expert analysis reports, maintenance manuals, equipment operation logs, power generation industry technical manuals, operating guides, safety regulations, engineering and technical papers, industry standards and specifications, and other corpus data from various sources.

3. The method for screening measuring points for equipment fault diagnosis in thermal power plants according to claim 2, characterized in that: The large language model adopts an open source large language model, and the open source large language model includes qwen2, qwen2.5, and DeepSeek.

4. The method for selecting measuring points for fault diagnosis of thermal power plant equipment according to claim 1, characterized in that: The method of fine-tuning the measuring point screening model adopts a low-load fine-tuning method; the low-load fine-tuning method includes a LOMO or LoRA method; The LOMO method specifically includes: using the LOMO optimizer to fuse gradient calculation and parameter update during the fine-tuning of the measurement point screening model; The LoRA method specifically includes: during the fine-tuning process of the measuring point screening model, adding a trainable low-rank matrix to the weight of the measuring point screening model to achieve fine-tuning.

5. The method for selecting measuring points for fault diagnosis of thermal power plant equipment according to claim 1, characterized in that: The RAG method comprises: By vectorizing and encoding expert knowledge data, the embedding model is used to convert the vectorized data into a high-dimensional vector representation and construct a vector database that can be efficiently retrieved; When a query statement for the measurement point screening model is entered, the system will also vectorize it, match semantically related paragraphs or rules from the vector library based on algorithms such as cosine similarity, and finally perform weighted sorting on the search results based on the logic of the domain knowledge base to generate accurate measurement point screening suggestions.

6. The method for selecting measuring points for fault diagnosis of thermal power plant equipment according to claim 5, characterized in that: In the step of generating the measuring point screening suggestion, the measuring point screening model further performs at least one of the following functions: Generate explanatory text of the measurement point type and name corresponding to the fault phenomenon; Integrate multi-source knowledge base data to infer the correlation between measurement point anomalies and fault types; Provide troubleshooting priority suggestions based on historical cases.

7. The method for screening measuring points for equipment fault diagnosis in thermal power plants according to claim 3 or 6, characterized in that: It also includes feedback optimization of the measurement point screening model; the feedback optimization specifically includes: The user inputs the feedback data into the measuring point screening model, and the measuring point association weights and reasoning logic are adjusted according to the feedback data, thereby dynamically adjusting the output measuring point screening suggestions and supplementary measuring point types; The feedback data includes the matching degree between the measurement point screening result and the actual fault.

8. A thermal power plant equipment fault diagnosis measuring point screening system, used to implement the thermal power plant equipment fault diagnosis measuring point screening method according to any one of claims 1 to 7, characterized in that: include: The data acquisition module obtains the measurement point names of all production equipment in the thermal power plant when a failure occurs in the production equipment of the thermal power plant; A measuring point screening module, the measuring point screening module includes a measuring point screening model, the measuring point screening model is used to screen out the measuring points related to the fault point according to the input measuring point name data, and then obtain the measuring point data and use the screened measuring point data to diagnose the faulty equipment; The test point screening model adopts a large language model optimized by fine-tuning or RAG method.

9. A computer-readable storage medium storing one or more programs, characterized in that: The one or more programs include instructions, which, when executed by a computing device, enable the computing device to execute the method for screening measuring points for equipment fault diagnosis in a thermal power plant according to any one of claims 1 to 7.

10. A computing device, characterized in that include: One or more processors, a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include steps for executing the method for screening fault diagnosis points of thermal power plant equipment according to any one of claims 1 to 7.

Images