Thermal station fault diagnosis method fusing physical mechanism simulation model and large model

Abstract

The application discloses a kind of heat station fault diagnosis methods of fusing physical mechanism simulation model and big model, comprising: when carrying out routine operation maintenance, input heat station routine operation maintenance voice or text first request information, after being parsed by big language model, together with real-time operation monitoring data, input into heat station fault diagnosis big model and carry out heat station operation diagnosis;When heat station operation diagnosis result is fault, call heat station whole process physical mechanism simulation model, according to heat station fault equipment and fault mode, analyze fault root cause and influence range;When carrying out diagnosis analysis of the fault phenomenon that has occurred, input heat station fault phenomenon voice or text second request information, after being recognized by big language model fault entity, fault intention, obtain fault entity equipment, professional fault feature, then call heat station whole process physical mechanism simulation model, carry out virtual reproduction to fault phenomenon, analyze fault root cause and influence range.

Description

Technical Field

[0001] This invention belongs to the field of thermal power station fault diagnosis technology, specifically involving a thermal power station fault diagnosis method that integrates physical mechanism simulation model and large model. Background Technology

[0002] The safe operation of heating stations is a critical link in the heating system. Once a malfunction or outage occurs, it directly affects the normal operation of residents and industrial and commercial establishments. For example, an outage can cause a sudden drop in indoor temperature, leading to an increase in resident complaints. In addition, some industrial parks rely on a continuous and stable supply of steam or hot water. If a heating station malfunctions, it can lead to production interruptions, order delays, and direct economic losses.

[0003] Heating stations comprise heat exchange equipment, pumps and valves, measurement and control instruments, and auxiliary equipment. Each type of equipment has different functions and potential failure modes, and the characteristics of their operating parameter changes vary under different failure modes. Therefore, diagnosing and analyzing the corresponding failure modes for each type of equipment is crucial. However, using large-scale models for heating station equipment fault diagnosis heavily relies on the coverage of historical fault data. This is problematic because data on different equipment and failure modes in heating stations suffers from imbalanced samples and scarcity of coupled fault data. Furthermore, large-scale models primarily output fault results through feature matching but cannot specifically explain the root causes of faults, the physical logic of fault evolution, or quantify the impact of faults on other equipment and the heating area. Additionally, while pure physical mechanism simulation models can reproduce equipment fault phenomena and analyze the root causes and evolution of faults layer by layer, dynamic simulation analysis is computationally intensive and sensitive to boundary conditions, making real-time diagnosis impossible.

[0004] Based on the above technical problems, a new method for diagnosing thermal power station faults is needed, which integrates physical mechanism simulation models and large-scale models. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a method for diagnosing thermal power station faults by integrating physical mechanism simulation models and large-scale models.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: This invention provides a method for diagnosing faults in thermal power stations that integrates physical mechanism simulation models and large-scale models, comprising: S1. After dynamically coupling the physical mechanism models of each core device in the heating station, a full-process physical mechanism simulation model of the heating station is formed. By inputting multi-fault scenario data of different fault modes of different core devices, the fault evolution process is simulated to generate virtual data of the heating station's full fault scenario. Combined with the historical normal operation data and fault operation data of the heating station, a full-scenario hybrid dataset of the heating station is constructed. S2. Utilize the full-scenario hybrid dataset of the heating station to pre-train a large model, learn the mapping relationship between the changes in equipment characteristic parameters and fault evolution knowledge under normal and fault modes of different core equipment, generate a large model for fault diagnosis of the heating station, and use it as a proxy model of the heating station to access the operation monitoring data of the heating station in real time to diagnose faulty equipment and fault modes of the heating station. S3. When the heating station personnel are performing routine operation and maintenance, they input the first request information in voice or text. The preset large language model parses the first request information and inputs it along with the real-time operation monitoring data of the heating station into the heating station fault diagnosis model for operation diagnosis. When the operation diagnosis result of the heating station is a fault, the full-process physical mechanism simulation model of the heating station is called to analyze the root cause and scope of impact of the fault based on the faulty equipment and fault mode of the heating station. When heating station personnel conduct diagnostic analysis of existing fault phenomena, they can input the second request information of the heating station fault phenomenon in voice or text. After the preset large language model identifies the fault entity and fault intent in the second request information, it obtains the fault entity equipment and professional fault characteristics. Then, it calls the full-process physical mechanism simulation model of the heating station to virtually reproduce the fault phenomenon and analyze the root cause and scope of impact of the fault. S4. Based on the root cause of the fault, the scope of impact, and historical fault handling data, train a large-scale fault handling model for the heating station. After outputting preliminary suggestions for actionable handling corresponding to the fault, call the full-process physical mechanism simulation model of the heating station to verify the rationality of the preliminary suggestions and output the final handling plan.

[0007] Furthermore, in S1, the core equipment includes heat exchange equipment, pump and valve equipment, measurement and control and instrumentation equipment, and auxiliary equipment; the core of the heat exchange equipment is the heat exchanger, which is equipped with seals, dirt separators, and heat exchange medium guides. The existing failure modes include heat exchange efficiency decay, structural and sealing failure, local heat transfer abnormality, and dirt separator failure. The physical mechanism models set include a fouling thermal resistance calculation model, a heat exchange model, and a seal aging failure model. The pump and valve equipment includes primary and secondary network circulating pumps, makeup water pumps, electrically adjustable valves, safety valves, pressure reducing valves, and steam traps. The existing failure modes include pump performance degradation, pump malfunction, pump mechanical failure, valve regulation failure, and valve safety protection failure. The physical mechanism models set include pump performance curve models, rotating machinery vibration models, pump net positive suction head (NPSH) calculation models, valve flow coefficient models, safety valve opening pressure calculation models, and steam trap flow calculation models. The measurement and control and instrumentation equipment includes sensors, controllers, communication modules and interlocking protection devices. The existing fault modes include measurement distortion, control failure, and communication and interlocking abnormalities. The physical mechanism models set include sensor error model, control adjustment model and signal transmission error model. The auxiliary equipment includes constant pressure water supply devices, water treatment equipment, air vents, venting valves, pipes and supports. The existing fault modes include constant pressure and water supply failure, air venting failure, and pipe and support failure. The physical mechanism models set include constant pressure water supply pressure calculation model, water hammer effect model, and pipe corrosion model.

[0008] Furthermore, the physical mechanism models of each core device in the heating station are dynamically coupled to form a full-process physical mechanism simulation model of the heating station, including: Based on the energy flow transmission and processing of each core device in the heating station, the topological connection relationship between the devices is established. Based on the rules of mass conservation, energy conservation, signal interaction, and data transmission, the dynamic coupling logic between the devices is established. Input and output interfaces and simulation time steps of the unified physical mechanism model of each device are defined to form a preliminary simulation model of the physical mechanism of the entire heating station process. Input the normal operating parameters of the heating station and compare the simulation output with the actual operating data to verify the normal operating conditions; input the parameters of a typical single fault mode and compare the simulation output with the actual fault data to verify that the fault feature evolution trend is consistent with reality; input the parameters of a multi-fault coupling scenario to verify the model's simulation capability for the evolution of cascading faults. After verification, a simulation model of the physical mechanism of the entire process of the heating station was formed.

[0009] Furthermore, in S2, a large model is pre-trained using a mixed dataset of the entire heating station scenario to learn the mapping relationship between changes in equipment characteristic parameters and fault evolution knowledge under normal and fault modes of different core equipment, generating a large-scale fault diagnosis model for the heating station, including: The parameters of multiple devices in the mixed dataset of the entire scenario of the heating station are aligned with time data to generate device-level time-series feature sequences. Numerical operating parameters and event-type fault labels are merged, and each data point is labeled with faulty device and fault mode labels to form a standardized dataset. The selected large-scale model of the base is pre-trained using the dataset to learn the dynamic correlation between parameters under normal and fault modes of different core equipment, the fault reasoning logic of the simulation model of the physical mechanism of the whole process of the heating station, and to embed the fault evolution rules to generate a large-scale model for fault diagnosis of the heating station.

[0010] Furthermore, in S3, when heating station personnel perform routine operation and maintenance, they input a first request message (voice or text) for routine operation and maintenance of the heating station. The first request message is then parsed by a preset large language model and input along with the real-time operation monitoring data of the heating station into the heating station fault diagnosis large model for heating station operation diagnosis, including: When heating station personnel perform routine operation and maintenance, they can input the first request information for routine operation and maintenance via voice or text through the input terminal. The preset voice recognition model will then convert the voice information into text information, and the text information will be preprocessed to form the core text of the routine operation and maintenance request. The selected general language model is fine-tuned with small samples using a pre-set corpus of thermal power station operation and maintenance dialogues. At the same time, a lexicon of thermal power station domain and a device-parameter association graph are embedded to form a large language model. After using the first major language model to identify the core intent and extract the relevant physical devices from the core text of daily operation and maintenance requests, real-time operation monitoring data of the relevant physical devices and coupled devices are obtained. By calling the large-scale fault diagnosis model of the heating station, based on the pre-trained full-scenario fault knowledge of the heating station, targeted feature extraction and fault matching are performed. If it is a normal mode, it indicates that the relevant equipment of the heating station is operating normally; if it is a fault mode, it indicates that the relevant equipment of the heating station is faulty. A preliminary fault diagnosis report containing the faulty equipment, fault mode, fault occurrence time and fault parameter deviation value is output.

[0011] Furthermore, in S3, when the diagnostic result of the heating station operation is a fault, the full-process physical mechanism simulation model of the heating station is invoked. Based on the faulty equipment and fault mode of the heating station, the root cause and scope of impact of the fault are analyzed, including: The large-scale fault diagnosis model for heating stations maps the core parameters in the preliminary fault diagnosis report to the preset boundary conditions and fault initialization parameters of the full-process physical mechanism simulation model of the heating station. Then, it calls the full-process physical mechanism simulation model of the heating station to dynamically reproduce the occurrence and evolution of the fault. If it is a single fault scenario, it directly simulates the parameter changes and fault evolution process of the faulty equipment before and after the fault. If it is a multi-equipment coupled fault scenario, it simulates the chain triggering and fault evolution process before and after the fault. Perform parameter back-inference and causal analysis on the fault evolution process, locate the root cause of the fault, and output a root cause analysis report, including the basis for root cause determination, the change curve of key mechanism parameters, and the causal relationship between the root cause and the fault. Using a full-process physical mechanism simulation model of a heating station, the chain reaction of a fault from the core equipment to other equipment in the heating station is simulated, and the scope and degree of the impact are quantitatively analyzed, outputting a quantitative analysis report on the scope of impact.

[0012] Furthermore, in S3, when heating station personnel are diagnosing and analyzing existing faults, they input a second request message (voice or text) regarding the fault phenomenon. A pre-set large language model then identifies the fault entity and fault intent in the second request message, obtaining the faulty equipment and its specific characteristics. Finally, a full-process physical mechanism simulation model of the heating station is invoked to virtually reproduce the fault phenomenon and analyze its root causes and scope of impact, including: When heating station personnel conduct diagnostic analysis of existing faults, they can input the second request information of the heating station fault symptoms via voice or text. The preset large language model will then extract and classify entities, locate the core physical equipment corresponding to the fault symptoms, and transform the fault symptoms into professional fault characteristics that can be quantified and described. The simulation model of the physical mechanism of the heating station is called, and the normal operating parameters of the heating station before the failure are loaded from the historical operation database as the basic operating parameters of the simulation model. Then, based on the preset professional fault feature-simulation physical parameter mapping library, the professional fault features are converted into corresponding simulation physical parameters. The initialization parameters and simulation boundary conditions corresponding to the fault phenomenon are set, and the simulation reproduction of the entire process of fault phenomenon evolution is started, including the initial triggering of the fault, the manifestation of the fault phenomenon, and the steady-state operation of the fault. The simulation-reproduced fault phenomena are quantitatively matched with the actual fault phenomena described by the maintenance personnel. If the matching degree is greater than the threshold, the root cause analysis and impact range analysis stages are entered. If the matching degree is less than the threshold, the fault initialization parameters of the simulation model are adjusted and the simulation is repeated until the matching degree meets the standard. Perform parameter back-inference and causal analysis on the entire process of fault phenomenon evolution, locate the root cause of the fault, and output a root cause analysis report, including the basis for root cause determination, the change curve of key mechanism parameters, and the causal relationship between the root cause and the fault. Using a full-process physical mechanism simulation model of a heating station, the chain reaction of a fault from the core equipment to other equipment in the heating station is simulated, and the scope and degree of the impact are quantitatively analyzed, outputting a quantitative analysis report on the scope of impact.

[0013] Furthermore, S4 includes: Data on fault causes, fault impact range, and historical fault handling are obtained. The selected basic large language model is pre-trained to allow the model to learn the mapping relationship between fault information and handling measures. At the same time, the model is fine-tuned and optimized by embedding handling phenomenon constraints and high-risk handling measures to form a large-scale model for handling faults in heating stations. Based on the current root cause and scope of impact of the heating station fault handling model, the current fault information is matched with historical faults for similarity. Priority is given to matching historical cases with high consistency in fault equipment and fault mode. Then, effective handling measures without secondary faults are selected. In addition, the selected handling measures are optimized and adjusted according to the scope of impact of the current fault to adapt to the on-site working conditions and output preliminary fault handling suggestions. Using a pre-defined handling suggestion-simulation parameter mapping library, the initial fault handling suggestions are analyzed, and the pre-handling operating parameters, post-handling operating parameters, and dynamic parameters of the handling process are extracted. The full-process physical mechanism simulation model of the heating station is then called to perform full-process simulation before, during, and after the handling. The steady-state operating conditions before and after the handling are compared to evaluate the fault elimination effect and heating compliance. The dynamic parameter changes during the handling are simulated to evaluate the feasibility, stability, and whether secondary faults will be caused by the handling operation. The time-series curves of parameter changes for each process and the final handling plan are output.

[0014] Furthermore, the fault cause data includes the fault cause, faulty equipment, fault mode, and deviation parameters; the fault impact range includes the impact on the operation of related equipment, the impact on the heating range, the impact on heating quality, and secondary faults; the historical fault handling data includes the handling measures and handling effects for different fault modes.

[0015] The beneficial effects of this invention are: (1) This invention uses a full-process physical mechanism simulation model of a heat station to dynamically couple the core equipment mechanism, simulate the evolution process of multiple faults, generate virtual data of full fault scenarios, cover low-probability accidents, extreme coupled faults, and rare operating condition faults in the actual operation and maintenance of heat stations, and make up for the lack of such scenarios in real fault data. (2) This invention utilizes a full-scenario hybrid dataset to pre-train a large-scale model for fault diagnosis of heat stations, allowing the large model to simultaneously learn the parameter mapping relationship between the normal mode and the fault mode of the equipment, as well as the fault evolution knowledge at the mechanism level. Even when faced with fault scenarios that have never occurred in the actual operation and maintenance of heat stations, it can still make accurate diagnoses. (3) For daily operation and maintenance request information, the present invention automatically parses the intention of the operation and maintenance personnel by the large language model, and after calling the large model for fault diagnosis of the heating station to output the fault equipment and fault mode, it calls the mechanism simulation model to analyze the root cause and impact range of the fault; for request information of fault phenomena that have occurred, the large language model is used to identify the fault entity and fault intention, accurately extract the fault entity equipment and professional fault characteristics, and call the mechanism simulation model to virtually reproduce the fault that has occurred, so that the operation and maintenance personnel can intuitively see the occurrence and evolution process of the fault, and realize accurate root cause tracing and impact range quantification; (4) This invention utilizes a large fault handling model to learn the actual situation of on-site faults and historical operation and maintenance experience, which solves the problems of low efficiency and incomplete consideration when manually formulating handling plans; and calls the mechanism simulation model to verify the preliminary handling suggestions, verifying the feasibility, effectiveness and safety of the handling suggestions from the perspective of physical laws; (5) The mechanism simulation model of this invention provides physically compliant training data and physical-level verification and analysis support for the large-scale fault diagnosis model and fault handling model of the heating station. The large-scale model provides efficient reasoning ability and low-threshold interaction entry for the mechanism simulation model. The integration of the two realizes the accuracy and interpretability of diagnosis and handling, solves the problems of black box diagnosis, misjudgment of data association and lack of physical support in the pure large-scale model, and also solves the problems of computation time consumption and inability to diagnose in real time in the pure mechanism simulation model. In addition, from dataset construction, fault diagnosis, root cause and impact analysis, to the generation and verification of handling plan, automation and intelligence are realized, which greatly improves the overall efficiency of heating station operation and maintenance.

[0016] Other features and advantages will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a thermal power station fault diagnosis method that integrates physical mechanism simulation model and large model according to the present invention; Figure 2 This is a flowchart of the heating station operation diagnosis method for heating station personnel during routine operation and maintenance, as described in this invention. Figure 3 This is a flowchart of the fault analysis method for personnel performing routine operation and maintenance at a heating station, as described in this invention. Figure 4 This invention provides a flowchart of a diagnostic analysis method for personnel at heating stations to perform analysis of existing malfunctions. Figure 5 This is a flowchart of the method for generating a fault handling scheme for a heating station according to the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] like Figure 1 As shown in the figure, this embodiment provides a method for diagnosing faults in thermal power stations that integrates physical mechanism simulation models and large-scale models. It includes: S1. After dynamically coupling the physical mechanism models of each core device in the heating station, a full-process physical mechanism simulation model of the heating station is formed. By inputting multi-fault scenario data of different fault modes of different core devices, the fault evolution process is simulated to generate virtual data of the heating station's full fault scenario. Combined with the historical normal operation data and fault operation data of the heating station, a full-scenario hybrid dataset of the heating station is constructed. S2. Utilize the full-scenario hybrid dataset of the heating station to pre-train a large model, learn the mapping relationship between the changes in equipment characteristic parameters and fault evolution knowledge under normal and fault modes of different core equipment, generate a large model for fault diagnosis of the heating station, and use it as a proxy model of the heating station to access the operation monitoring data of the heating station in real time to diagnose faulty equipment and fault modes of the heating station. S3. When the heating station personnel are performing routine operation and maintenance, they input the first request information in voice or text. The preset large language model parses the first request information and inputs it along with the real-time operation monitoring data of the heating station into the heating station fault diagnosis model for operation diagnosis. When the operation diagnosis result of the heating station is a fault, the full-process physical mechanism simulation model of the heating station is called to analyze the root cause and scope of impact of the fault based on the faulty equipment and fault mode of the heating station. When heating station personnel conduct diagnostic analysis of existing fault phenomena, they can input the second request information of the heating station fault phenomenon in voice or text. After the preset large language model identifies the fault entity and fault intent in the second request information, it obtains the fault entity equipment and professional fault characteristics. Then, it calls the full-process physical mechanism simulation model of the heating station to virtually reproduce the fault phenomenon and analyze the root cause and scope of impact of the fault. S4. Based on the root cause of the fault, the scope of impact, and historical fault handling data, train a large-scale fault handling model for the heating station. After outputting preliminary suggestions for actionable handling corresponding to the fault, call the full-process physical mechanism simulation model of the heating station to verify the rationality of the preliminary suggestions and output the final handling plan.

[0022] In this embodiment, in S1, the core equipment includes heat exchange equipment, pump and valve equipment, measurement and control and instrumentation equipment, and auxiliary equipment; the core of the heat exchange equipment is the heat exchanger, which is equipped with seals, dirt separators, and heat exchange medium guides. The existing failure modes include heat exchange efficiency decay, structural and sealing failure, local heat transfer abnormality, and dirt separator failure. The physical mechanism models set include a fouling thermal resistance calculation model, a heat exchange model, and a seal aging failure model. The pump and valve equipment includes primary and secondary network circulating pumps, makeup water pumps, electrically adjustable valves, safety valves, pressure reducing valves, and steam traps. The existing failure modes include pump performance degradation, pump malfunction, pump mechanical failure, valve regulation failure, and valve safety protection failure. The physical mechanism models set include pump performance curve models, rotating machinery vibration models, pump net positive suction head (NPSH) calculation models, valve flow coefficient models, safety valve opening pressure calculation models, and steam trap flow calculation models. The measurement and control and instrumentation equipment includes sensors, controllers, communication modules and interlocking protection devices. The existing fault modes include measurement distortion, control failure, and communication and interlocking abnormalities. The physical mechanism models set include sensor error model, control adjustment model and signal transmission error model. The auxiliary equipment includes constant pressure water supply devices, water treatment equipment, air vents, venting valves, pipes and supports. The existing fault modes include constant pressure and water supply failure, air venting failure, and pipe and support failure. The physical mechanism models set include constant pressure water supply pressure calculation model, water hammer effect model, and pipe corrosion model.

[0023] Heat exchange efficiency decline: scaling, slag accumulation and fouling of heat exchange tubes and heat exchange surfaces; structural and sealing failures: aging and damage of plate gaskets, corrosion and cracking of heat exchange tubes, and leakage of flange sealing surfaces; local heat transfer abnormalities: local freezing of heat exchange surfaces and media deviation; dirt separator failure: filter screen blockage, damage, and impurities entering the heat exchange surface, exacerbating fouling.

[0024] The fouling thermal resistance calculation model quantifies the increase in thermal resistance caused by fouling on the heat exchanger surface. It is the core model for judging heat exchange efficiency degradation faults, and is expressed as: ; The primary heat transfer coefficient is the convective heat transfer coefficient on the grid side. The thickness of the heat exchanger wall surface; The thermal conductivity of the heat exchange wall material; The secondary network side convective heat transfer coefficient; Total fouling thermal resistance; The heat transfer model quantifies the total heat transfer and local heat transfer of a heat exchanger. It is the core calculation model for heat transfer efficiency decay and local heat transfer anomalies. By analyzing heat transfer deviations and local heat transfer flux anomalies, the degree and location of the fault are determined. The overall heat transfer model is expressed as: ; Total heat exchange under scaling conditions; The effective heat exchange area of ​​the heat exchanger; The logarithmic mean temperature difference; Local heat transfer flux is used to determine local heat transfer anomalies and quantify the heat transfer capacity of a local area of ​​a heat exchanger, expressed as: ; The heat transfer coefficient of the medium; This refers to the temperature difference between the local medium and the heat exchange wall surface; The purpose of the seal aging model is to consider the dual aging effects of operating temperature and operating time, quantify the aging coefficient, and express it as: ; Performance after aging; For the initial performance of the seal; This is a correction factor for the aging coefficient; The aging coefficient of the seal; The essence of seal failure determination is that the sealing specific pressure generated by the seal under compression is greater than the working pressure of the medium. As the seal ages, its elasticity decreases, and insufficient compression leads to a drop in the sealing specific pressure. When the specific pressure is lower than the medium pressure, a seal leakage failure occurs. The sealing specific pressure is calculated as follows: ; The compressive force exerted on the seal; This refers to the sealing contact area of ​​the seal. Dynamic coupling calculation of heat exchange equipment: Increased fouling thermal resistance leads to a decrease in the overall heat transfer coefficient and a reduction in heat exchange capacity; increased aging of seals leads to a decrease in sealing specific pressure, seal leakage, and loss of medium flow, which further leads to a decrease in the overall heat transfer coefficient and a reduction in heat exchange capacity, which is consistent with the chain evolution law of actual faults.

[0025] The pump performance curve model characterizes the inherent characteristics of centrifugal pumps in heating stations, including flow rate-head, flow rate-shaft power, and flow rate-efficiency. It is the core model for diagnosing pump performance degradation faults, quantifying the degree of head, efficiency, and power degradation by analyzing the deviation between the actual performance curve and the rated curve. The core formula for flow rate-head is expressed as: ; For actual head; Rated head; Performance curve fitting coefficients; This refers to the actual operating flow rate of the pump. The head deviation is caused by the malfunction; The core formula for flow rate-shaft power is expressed as: ; This represents the actual shaft power. Rated shaft power; These are the coefficients for fitting the performance curve; The power deviation is caused by the fault; The core formula for flow-efficiency is expressed as: ; For actual efficiency; For maximum efficiency; These are the coefficients for fitting the performance curve; The flow rate at which the pump achieves optimal efficiency; Efficiency deviation caused by malfunction; The purpose of rotating machinery vibration models is to quantify the vibration intensity of centrifugal pump rotor systems, identify abnormal pump operation faults, and provide fault characteristic evidence for pump mechanical failures. The core formula for vibration intensity is expressed as: ; Effective vibration intensity; Vibration sampling time; Instantaneous vibration velocity; The purpose of the pump cavitation margin calculation model is to determine whether cavitation has occurred in a centrifugal pump. Cavitation directly leads to pump performance degradation, excessive vibration, and ultimately mechanical failure. The core formula for effective cavitation margin is expressed as: ; This is the absolute pressure at the pump inlet; The average flow velocity at the pump inlet; This corresponds to the water vaporization pressure at the specified water temperature. The valve flow coefficient model quantifies the valve's flow regulation capability and is the core model for determining valve regulation failure. It is represented as: ; The deviation rate of the flow coefficient; This is the actual flow coefficient; The rated flow coefficient; The safety valve trip pressure calculation model quantifies the trip pressure of the safety valve and is the core judgment model for valve safety protection failure. It is expressed as: ; Set the opening pressure for the safety valve; For the spring stiffness of the safety valve; This refers to the pre-compression amount of the safety valve spring. The sealing pressure ratio of the safety valve sealing surface; This refers to the back pressure at the safety valve outlet. , These are the pressure-bearing areas on the inlet side and the pressure-bearing areas on the outlet side of the safety valve disc, respectively. The function of the steam trap flow calculation model is to quantify the condensate discharge capacity of the steam trap in the heating station. It is the core judgment model for valve safety protection failure. By the deviation between the actual water delivery and the rated condensate discharge, it can determine whether the steam trap is not draining properly or is leaking steam.

[0026] The sensor error model quantifies the measurement errors of sensors such as temperature, pressure, and flow rate in a heating station. It is the core model for judging measurement distortion faults. The error rate quantification formula is expressed as: ; For measurement error rate; This is the actual value indicated by the sensor; The true value of the measured physical quantity; The control and regulation model quantifies the PID proportional-integral-derivative regulation process of the heating station controller. It is the core judgment model for control failure. By calculating indicators such as regulation deviation, overshoot, and settling time, it can determine problems such as controller parameter mismatch, actuator lag, and control logic failure.

[0027] The signal transmission error model quantifies the transmission error of analog and digital signals between sensors, controllers, and actuators in a thermal power station. It is the core judgment model for communication and interlocking anomalies. By calculating the amplitude attenuation, noise interference, and delay lag of the transmitted signal, it can determine problems such as aging of communication lines and electromagnetic interference.

[0028] The constant pressure water supply calculation model quantifies the system constant pressure point pressure, the working pressure of the pressure tank, and the water supply volume of the constant pressure water supply device in the heating station. It is the core model for judging constant pressure and water supply faults, quantifying fault characteristics such as excessively high or low constant pressure and insufficient water supply capacity through constant pressure deviation and insufficient water supply rate. The calculated working pressure of the pressure tank is expressed as follows: ; Absolute pressure of gas inside the pressure tank after adding water; This is the absolute pressure of the gas inside the pressure tank after the water is drained. The volume of gas inside the pressure tank after adding water; This represents the volume of gas inside the pressure tank after the water has been drained. The water replenishment amount is calculated as follows: ; For safety factor; For system circulation flow; The system leakage rate; This refers to the volume change caused by variations in system water temperature. The water hammer effect model is used to quantify the sudden rise and fall of pressure caused by abrupt changes in flow velocity within the pipeline of a heating station. It is the core model for pipeline and support failures, and the risk of failures such as pipeline vibration, support loosening, and pipeline rupture is determined by the water hammer pressure amplitude.

[0029] In this embodiment, the physical mechanism models of each core device in the heating station are dynamically coupled to form a full-process physical mechanism simulation model of the heating station, including: Based on the energy flow transmission and processing of each core device in the heating station, the topological connection relationship between the devices is established. Based on the rules of mass conservation, energy conservation, signal interaction, and data transmission, the dynamic coupling logic between the devices is established. Input and output interfaces and simulation time steps of the unified physical mechanism model of each device are defined to form a preliminary simulation model of the physical mechanism of the entire heating station process. Input the normal operating parameters of the heating station and compare the simulation output with the actual operating data to verify the normal operating conditions; input the parameters of a typical single fault mode and compare the simulation output with the actual fault data to verify that the fault feature evolution trend is consistent with reality; input the parameters of a multi-fault coupling scenario to verify the model's simulation capability for the evolution of cascading faults. After verification, a simulation model of the physical mechanism of the entire process of the heating station was formed.

[0030] In this embodiment, in step S2, a large model is pre-trained using a mixed dataset of the entire heating station scenario to learn the mapping relationship between changes in equipment characteristic parameters and fault evolution knowledge under normal and fault modes of different core equipment, thereby generating a large-scale fault diagnosis model for the heating station, including: The parameters of multiple devices in the mixed dataset of the entire scenario of the heating station are aligned with time data to generate device-level time-series feature sequences. Numerical operating parameters and event-type fault labels are merged, and each data point is labeled with faulty device and fault mode labels to form a standardized dataset. The selected large-scale model of the base is pre-trained using the dataset to learn the dynamic correlation between parameters under normal and fault modes of different core equipment, the fault reasoning logic of the simulation model of the physical mechanism of the whole process of the heating station, and to embed the fault evolution rules to generate a large-scale model for fault diagnosis of the heating station.

[0031] In practical applications, a two-level labeling system is established: first, the faulty equipment is divided into primary labels according to the core equipment of the heating station, and then the fault mode is divided into secondary labels according to the typical fault modes of each equipment. A unique label ID is assigned to each label (to facilitate large model coding). The labeling system needs to cover all fault types in the full-scenario hybrid dataset.

[0032] Labeling rules include: Single device, single fault: Label the faulty device ID and fault mode ID with a unique identifier; Multi-device coupled failure: Label the primary failure device ID, primary failure mode ID, secondary failure device ID, and secondary failure mode ID (classify primary and secondary according to the cause-and-effect relationship of the failure, such as heat exchanger scaling causing pump cavitation, in which case the heat exchanger is the primary failure device). Normal operating conditions: Mark faulty equipment ID=0, fault mode ID=0.

[0033] Since the core of fault diagnosis in heating stations is time-series data modeling and fault mechanism reasoning, the base model should prioritize models that support time-series modeling, can integrate knowledge rules, and have lightweight reasoning capabilities. The selected base models include time-series dedicated models, multimodal models, and knowledge-enhanced models. During parameter dynamic correlation learning, the large model learns the time-series dynamic correlation rules between parameters within the same equipment and between parameters of different equipment (e.g., an increase in the primary network temperature of the heat exchanger will lead to a synchronous increase in the secondary network temperature, and a decrease in the secondary network flow will lead to a decrease in the inlet pressure of the circulating pump), and grasps the parameter change logic under normal / fault conditions of the heating station. During the learning of mechanistic simulation reasoning logic, the large model learns the fault reasoning logic of the full-process physical mechanism simulation model of the thermal power station. This enables the large model to think like the mechanistic simulation model: calculating intermediate mechanistic features from equipment operating parameters, and then determining faulty equipment and fault modes. Specifically, the process is as follows: Teacher Model (Mechanism Simulation Model): Inputs the standardized dataset's equipment operating parameters into the full-process physical mechanism simulation model of the thermal power station, outputting intermediate mechanistic features and fault determination soft labels; Student Model (Base Large Model): Adds a mechanistic feature embedding layer to the large model, allowing it to learn and extract virtual mechanistic features from equipment operating parameters (with dimensions consistent with the intermediate mechanistic features of the teacher model); Distillation Training: Allows the virtual mechanistic features of the student model to approximate the real mechanistic features of the teacher model, while simultaneously allowing the fault prediction probability of the student model to approximate the fault determination soft labels of the teacher model, achieving transfer learning of mechanistic reasoning logic. Fault evolution rule embedding involves deeply embedding the thermal power station fault evolution rules into the model weights, ensuring that the model's fault diagnosis results conform to the causal logic of fault evolution. The large-scale fault diagnosis model for heating stations can learn the time-series dynamic correlation of parameters of multiple devices in the heating station, perform fault mechanism reasoning like a mechanism simulation model, and determine faulty devices and fault modes according to fault evolution rules.

[0034] like Figure 2 As shown, in this embodiment, in step S3, when the heating station personnel perform routine operation and maintenance, they input a first request message in voice or text. The first request message is then parsed by a preset large language model and input along with the heating station's real-time operation monitoring data into the heating station fault diagnosis large model for heating station operation diagnosis, including: When heating station personnel perform routine operation and maintenance, they can input the first request information for routine operation and maintenance via voice or text through the input terminal. The preset voice recognition model will then convert the voice information into text information, and the text information will be preprocessed to form the core text of the routine operation and maintenance request. The selected general language model is fine-tuned with small samples using a pre-set corpus of thermal power station operation and maintenance dialogues. At the same time, a lexicon of thermal power station domain and a device-parameter association graph are embedded to form a large language model. After using the first major language model to identify the core intent and extract the relevant physical devices from the core text of daily operation and maintenance requests, real-time operation monitoring data of the relevant physical devices and coupled devices are obtained. By calling the large-scale fault diagnosis model of the heating station, based on the pre-trained full-scenario fault knowledge of the heating station, targeted feature extraction and fault matching are performed. If it is a normal mode, it indicates that the relevant equipment of the heating station is operating normally; if it is a fault mode, it indicates that the relevant equipment of the heating station is faulty. A preliminary fault diagnosis report containing the faulty equipment, fault mode, fault occurrence time and fault parameter deviation value is output.

[0035] It should be noted that routine operation and maintenance refers to the daily maintenance work performed by the heating station personnel. At fixed times each day, the heating station's fault diagnosis model performs the day's operational diagnostics, such as checking the operating status of the secondary network circulation pumps and whether the plate heat exchangers are operating normally. The heating station operation and maintenance dialogue corpus collects and organizes question-and-answer pairs from real-world daily operation and maintenance scenarios at heating stations, forming a small-sample fine-tuning dataset covering core scenarios such as equipment status queries, parameter monitoring, fault diagnosis, and operation and maintenance consultation. Each data entry is formatted as input question (operation and maintenance request) and output result (intent, entity equipment). The heating station domain lexicon organizes heating station-specific equipment vocabulary, parameter vocabulary, fault vocabulary, and operation and maintenance vocabulary, forming a structured lexicon to improve the word segmentation accuracy and domain understanding capabilities of the large language model. The equipment-parameter association graph organizes the relationships between equipment, its category, monitored parameters, associated coupling devices, and fault modes, allowing the large language model to automatically deduce relevant parameters and coupling devices based on entity equipment.

[0036] Using an operations and maintenance (O&M) dialogue corpus as training data, only the attention layer of the general-purpose large language model is fine-tuned, while the weights of other layers in the large language model are fixed. The core task is to enable the large language model to learn the mapping relationship between heat station O&M requests, intents, and entity devices. Fine-tuning tasks: Two supervised tasks are constructed: one for intent classification and the other for entity extraction. Optimization goals: Improve the accuracy of intent recognition and the recall rate of entity extraction. Fine-tuning can be stopped when the validation set accuracy is ≥95%. Domain lexicon and device-parameter association graph are embedded into the fine-tuned model to achieve knowledge enhancement: 1) Lexicon embedding: The domain lexicon is added to the word segmentation vocabulary of the model, and the word segmenter is retrained to ensure that the model can accurately identify domain words such as cavitation margin and constant pressure water supply device, avoiding word segmentation errors; 2) Graph embedding: The device-parameter association graph is transformed into triple knowledge and embedded into the large language model through the knowledge injection layer, so that the large language model can automatically call the association information (such as coupled devices and core parameters) in the graph when parsing entity devices.

[0037] After receiving the data, the fault diagnosis big data model performs targeted feature extraction, fault mode matching, and result determination based on pre-trained full-scenario fault knowledge. Step 1): Targeted Feature Extraction For physical devices and coupled devices, two types of core features are extracted: temporal dynamic features: the temporal variation trend of parameters; physical mechanism features: intermediate mechanism parameters are calculated through the built-in physical mechanism model. Step 2): Fault mode matching The extracted core features are compared and matched with the normal mode feature library and the fault mode feature library pre-trained by the large model: Step 1: Match with the normal mode feature library to determine whether it conforms to the characteristic patterns of normal equipment operation; Step 2: If it does not conform to the normal pattern, it is matched with the fault mode feature library (including single device fault and multi-device coupled fault). The most matching fault mode is found by calculating the feature similarity, and the primary fault device and secondary fault device are determined at the same time (according to the fault causal relationship). Step 3): Result Determination Based on the matching results, output one of two modes: Normal mode: If the extracted features have a similarity of ≥95% with the normal mode feature library, the device is determined to be operating normally; Fault mode: If the extracted features have a similarity of ≥85% with a certain fault mode feature library, the equipment is determined to be faulty, and fault-related details (fault occurrence time, parameter deviation value) are extracted.

[0038] like Figure 3 As shown, in this embodiment, in step S3, when the diagnostic result of the heating station operation is a fault, the full-process physical mechanism simulation model of the heating station is invoked. Based on the faulty equipment and fault mode of the heating station, the root cause and scope of influence of the fault are analyzed, including: The large-scale fault diagnosis model for heating stations maps the core parameters in the preliminary fault diagnosis report to the preset boundary conditions and fault initialization parameters of the full-process physical mechanism simulation model of the heating station. Then, it calls the full-process physical mechanism simulation model of the heating station to dynamically reproduce the occurrence and evolution of the fault. If it is a single fault scenario, it directly simulates the parameter changes and fault evolution process of the faulty equipment before and after the fault. If it is a multi-equipment coupled fault scenario, it simulates the chain triggering and fault evolution process before and after the fault. Perform parameter back-inference and causal analysis on the fault evolution process, locate the root cause of the fault, and output a root cause analysis report, including the basis for root cause determination, the change curve of key mechanism parameters, and the causal relationship between the root cause and the fault. Using a full-process physical mechanism simulation model of a heating station, the chain reaction of a fault from the core equipment to other equipment in the heating station is simulated, and the scope and degree of the impact are quantitatively analyzed, outputting a quantitative analysis report on the scope of impact.

[0039] In practical applications, constructing a standardized fault parameter-simulation mechanism parameter mapping library is crucial for parameter conversion. Based on the physical mechanism models of four main categories of equipment in the heating station—heat exchangers, pumps / valve systems, measurement and control systems, and auxiliary systems—the library establishes the correspondence between monitoring parameters in the fault diagnosis report and physical mechanism parameters in the simulation model, categorizing them by faulty equipment and fault mode. Fault initialization parameter conversion: The deviation and actual values ​​of monitoring parameters are converted into initial values ​​of the mechanism parameters of the faulty equipment in the simulation model (e.g., converting a 25% decrease in heat exchanger efficiency into a specific value for fouling thermal resistance), serving as the initial fault triggering condition for the simulation. Preset boundary condition conversion: The current operating conditions of the heating station and the associated system parameters of the faulty equipment are converted into global boundary conditions of the simulation model, ensuring that the simulation environment is consistent with the actual field conditions.

[0040] Single-device, single-fault scenario: Focuses on the independent evolution simulation of the faulty device, suitable for scenarios where only a single device fails without triggering abnormal parameters in other devices (e.g., performance degradation of a single water pump, false tripping of a single safety valve, or measurement distortion of a single temperature sensor). The simulation logic focuses on the faulty device itself and does not need to consider the cascading triggering of multiple devices. Core steps: Simulate normal operating conditions before failure: Based on the boundary conditions of the configuration package and the rated parameters of the equipment, simulate the normal operating state of the faulty equipment and related systems before the fault is triggered, and output the time series curves of all parameters (mechanistic parameters + monitoring parameters) in this stage as a fault comparison benchmark (such as fouling thermal resistance, heat exchange efficiency, and supply and return water temperature difference when the heat exchanger is operating normally). Simulation of fault triggering process: At the set fault occurrence time node, input fault initialization parameters (such as adjusting the heat exchanger fouling thermal resistance from the rated value to the fault value) into the simulation model, simulate the moment of fault triggering, and output the parameter mutation characteristics at the time of fault triggering (such as sudden increase in differential pressure and sudden decrease in heat exchange efficiency). Simulation of fault evolution and steady-state process: Based on the physical mechanism model and the characteristics of the equipment itself, the simulation of the entire process of fault from triggering to gradual development to fault steady state is performed (e.g., after fouling of the heat exchanger, the fouling thermal resistance increases slowly over time, and the heat exchange efficiency continues to decline until it reaches a stable value). The continuous change curve of parameters in this stage is output, focusing on capturing the key nodes of fault evolution (e.g., the node where the parameter decay rate changes from fast to slow, and the node where the parameters reach steady state). Fault reproduction result output: Integrate the full time-series simulation data of the fault pre-fault, fault trigger, fault evolution, and fault steady state, output the parameter change comparison table of the faulty equipment before and after the fault and the time-series change curve of the mechanism parameter / monitoring parameter, and generate a visual animation of fault evolution (such as the dynamic process of heat exchanger scaling and the impeller cavitation process of pump cavitation).

[0041] Multi-device coupled failure scenario: Simulation based on causal chain triggering and evolution, suitable for scenarios where a core device failure leads to subsequent failures in other devices. The core simulation logic is to simulate the chain triggering and sequential evolution of the failure from the core faulty device to related coupled devices according to the fault causal chain. Core steps: Identify the cause-and-effect chain of the fault: Based on the preliminary fault diagnosis report and the equipment-parameter correlation diagram of the heating station, identify the core fault equipment (fault initiating equipment), secondary fault equipment (equipment triggered by the core equipment), the causal relationship of the fault, and the chain triggering sequence, thus forming a clear cause-and-effect chain of the fault. Simulates fault cascading triggering step-by-step according to the causal chain: Starting with the fault triggering of the core faulty device, the fault triggering process of each coupled node is simulated step-by-step according to the time sequence. Each simulation step is based on the fault evolution result of the previous device, ensuring the logic and timing of the cascading triggering. Step 1: Simulate the fault triggering and initial evolution of the core faulty equipment, and output the impact of its fault parameters on the related systems (such as heat exchanger fouling leading to persistently low secondary network temperature). Step 2: Use the failure impact parameters of the core equipment as the failure triggering conditions of the secondary equipment, and simulate the failure triggering process of the secondary equipment (such as the low temperature of the secondary network causing the electric regulating valve to open wide and the flow rate to increase sharply). Step 3: Simulate the fault triggering and evolution of all subsequent secondary devices in sequence until the entire coupled fault system reaches a steady state (i.e., the fault parameters of all coupled devices no longer change significantly with time). Simulation of the interaction process of coupled faults: Some coupled faults interact with each other (such as pump cavitation leading to a further decrease in secondary network flow, which in turn aggravates fouling in the heat exchanger). The simulation needs to consider this positive and negative feedback effect and simulate the mutual influence of parameters to ensure the authenticity of the simulation results. Output of coupled fault reproduction results: Integrate the parameter change curves before and after the fault of the core equipment and all secondary equipment, output the coupled fault chain triggering timing table, mechanism parameters of each equipment, full time-series change curves of monitoring parameters, and coupled fault evolution visualization animation (showing the process of fault propagation from the core equipment to other equipment according to the causal chain), highlighting the trigger time, triggering conditions, and parameter mutation characteristics of each fault node.

[0042] The core of root cause analysis is to first find the first parameter anomaly point before the fault is triggered by reverse reasoning, and then verify the causal relationship between the anomaly point and the subsequent fault through a physical mechanism model to ensure the accuracy and uniqueness of root cause location.

[0043] Based on the reverse parameter back-reasoning of simulation data, the anomaly point of the fault initiation parameter is located. Taking the steady-state parameter data of the fault in the fault reproduction simulation as the endpoint and the normal operating condition parameter data before the fault triggering as the starting point, the core mechanism parameters are traced back and back-reasoned in reverse chronological order. The core steps are as follows: Determine the core mechanism parameters for back-inference: For the failure mode of the faulty equipment, select the mechanism parameters that best reflect the essence of the failure as the core parameters for back-inference (e.g., select fouling thermal resistance for scaling failure, select cavitation margin for cavitation failure, and select flow coefficient for valve regulation failure). Tracing the parameter change trajectory in reverse: View the time series change curves of the core mechanism parameters in reverse order, find the time node when the parameter first deviates abnormally from the rated value (i.e., the fault initiation time node), and extract all related parameters of that node (such as water hardness, heat exchanger cleaning time, and secondary network flow rate when the fouling thermal resistance first becomes abnormal). Locating the source of abnormal parameters: Analyze the related parameters at the initial time point to find the source parameter that caused the first abnormality of the core mechanism parameter (e.g., the first abnormality of the fouling thermal resistance is because the water hardness increased from the rated ≤0.03mmol / L to 0.15mmol / L, and more than 3 months have passed since the last plate cleaning). The abnormality of the source parameter is the direct manifestation of the root cause of the failure. Exclude non-root cause parameter anomalies: For other parameter anomalies found during the reverse investigation process, exclude non-root causes by chronological order and mechanistic logic (such as abnormal vibration acceleration after pump cavitation, which is the result of fault evolution rather than the root cause), to ensure the uniqueness of root cause location.

[0044] Based on the physical mechanism model, a forward causal analysis verifies the logical chain between the root cause and the fault. For the abnormal parameter source found through reverse engineering, a forward causal verification is performed using the physical mechanism model of the heating station, proving that this abnormal source is the sole root cause of the fault. Core steps: Constructing the mechanistic logic chain of root causes and failures: Based on physical mechanism formulas, sorting out the complete mechanistic logic chain from root cause to direct cause and then to failure phenomenon; Forward simulation verification: In the simulation model, only the abnormal parameters corresponding to the root cause are input (such as adjusting the water hardness to the excessive value, while other parameters are at the rated value), start the simulation, and verify whether the entire subsequent fault process can be reproduced (from the abnormal core mechanism parameters to the occurrence of the direct cause, and then to the final fault phenomenon). Root cause uniqueness verification: In the simulation model, eliminate the root cause individually (e.g., restore the water hardness to the rated value, or simulate periodic cleaning of the plates), start the simulation, and verify whether the fault will not occur at all; if the fault does not occur after eliminating the root cause, it can be proved that the root cause is the only fundamental cause of the fault.

[0045] like Figure 4As shown, in this embodiment, in step S3, when the heating station personnel are diagnosing and analyzing a fault, they input a second request message (voice or text) regarding the fault. A preset large language model then identifies the fault entity and fault intent in the second request message, obtaining the faulty equipment and its specific characteristics. Finally, the heating station's full-process physical mechanism simulation model is invoked to virtually reproduce the fault phenomenon and analyze its root cause and impact range, including: When heating station personnel conduct diagnostic analysis of existing faults, they can input the second request information of the heating station fault symptoms via voice or text. The preset large language model will then extract and classify entities, locate the core physical equipment corresponding to the fault symptoms, and transform the fault symptoms into professional fault characteristics that can be quantified and described. The simulation model of the physical mechanism of the heating station is called, and the normal operating parameters of the heating station before the failure are loaded from the historical operation database as the basic operating parameters of the simulation model. Then, based on the preset professional fault feature-simulation physical parameter mapping library, the professional fault features are converted into corresponding simulation physical parameters. The initialization parameters and simulation boundary conditions corresponding to the fault phenomenon are set, and the simulation reproduction of the entire process of fault phenomenon evolution is started, including the initial triggering of the fault, the manifestation of the fault phenomenon, and the steady-state operation of the fault. The simulation-reproduced fault phenomena are quantitatively matched with the actual fault phenomena described by the maintenance personnel. If the matching degree is greater than the threshold, the root cause analysis and impact range analysis stages are entered. If the matching degree is less than the threshold, the fault initialization parameters of the simulation model are adjusted and the simulation is repeated until the matching degree meets the standard. Perform parameter back-inference and causal analysis on the entire process of fault phenomenon evolution, locate the root cause of the fault, and output a root cause analysis report, including the basis for root cause determination, the change curve of key mechanism parameters, and the causal relationship between the root cause and the fault. Using a full-process physical mechanism simulation model of a heating station, the chain reaction of a fault from the core equipment to other equipment in the heating station is simulated, and the scope and degree of the impact are quantitatively analyzed, outputting a quantitative analysis report on the scope of impact.

[0046] It's important to clarify that the "already occurred fault scenario" is defined as a fault phenomenon that has been clearly identified by the heating station personnel. This requires subsequent explanations of why the fault occurred, how it evolved, and its impact. Since heating station personnel possess certain operational and maintenance knowledge, they can quickly identify fault characteristics when they discover anomalies during operation monitoring. Therefore, maintenance personnel only need to rely on the dynamic evolution, parameter back-calculation, and causal verification capabilities of the physical mechanism simulation model for fault reproduction, root cause analysis, and impact assessment, rather than the pattern matching capabilities of a large-scale fault diagnosis model. However, maintenance personnel may describe surface-level fault phenomena, which could correspond to multiple fault modes. Furthermore, large-scale fault diagnosis models can only perform probability matching and cannot verify whether the fault phenomenon is truly caused by a specific fault. In contrast, the physical mechanism simulation model can virtually reproduce the entire fault evolution process and quantitatively verify the matching degree between the reproduced results and the actual phenomena.

[0047] like Figure 5 As shown, in this embodiment, S4 includes: Data on fault causes, fault impact range, and historical fault handling are obtained. The selected basic large language model is pre-trained to allow the model to learn the mapping relationship between fault information and handling measures. At the same time, the model is fine-tuned and optimized by embedding handling phenomenon constraints and high-risk handling measures to form a large-scale model for handling faults in heating stations. Based on the current root cause and scope of impact of the heating station fault handling model, the current fault information is matched with historical faults for similarity. Priority is given to matching historical cases with high consistency in fault equipment and fault mode. Then, effective handling measures without secondary faults are selected. In addition, the selected handling measures are optimized and adjusted according to the scope of impact of the current fault to adapt to the on-site working conditions and output preliminary fault handling suggestions. Using a pre-defined handling suggestion-simulation parameter mapping library, the initial fault handling suggestions are analyzed, and the pre-handling operating parameters, post-handling operating parameters, and dynamic parameters of the handling process are extracted. The full-process physical mechanism simulation model of the heating station is then called to perform full-process simulation before, during, and after the handling. The steady-state operating conditions before and after the handling are compared to evaluate the fault elimination effect and heating compliance. The dynamic parameter changes during the handling are simulated to evaluate the feasibility, stability, and whether secondary faults will be caused by the handling operation. The time-series curves of parameter changes for each process and the final handling plan are output.

[0048] In this embodiment, the fault cause data includes the fault cause, faulty equipment, fault mode, and deviation parameters; the fault impact range includes the impact on the operation of related equipment, the impact on the heating range, the impact on heating quality, and secondary faults; the historical fault handling data includes the handling measures and handling effects for different fault modes.

[0049] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0050] Furthermore, the functional modules in the various embodiments of this invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part. If the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0051] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for diagnosing faults in thermal power stations that integrates physical mechanism simulation models and large-scale models, characterized in that, It includes: S1. After dynamically coupling the physical mechanism models of each core device in the heating station, a full-process physical mechanism simulation model of the heating station is formed. By inputting multi-fault scenario data of different fault modes of different core devices, the fault evolution process is simulated to generate virtual data of the heating station's full fault scenario. Combined with the historical normal operation data and fault operation data of the heating station, a full-scenario hybrid dataset of the heating station is constructed. S2. Utilize the full-scenario hybrid dataset of the heating station to pre-train a large model, learn the mapping relationship between the changes in equipment characteristic parameters and fault evolution knowledge under normal and fault modes of different core equipment, generate a large model for fault diagnosis of the heating station, and use it as a proxy model of the heating station to access the operation monitoring data of the heating station in real time to diagnose faulty equipment and fault modes of the heating station. S3. When the heating station personnel are performing routine operation and maintenance, they input the first request information in voice or text. The preset large language model parses the first request information and inputs it along with the real-time operation monitoring data of the heating station into the heating station fault diagnosis model for operation diagnosis. When the operation diagnosis result of the heating station is a fault, the full-process physical mechanism simulation model of the heating station is called to analyze the root cause and scope of impact of the fault based on the faulty equipment and fault mode of the heating station. When heating station personnel conduct diagnostic analysis of existing fault phenomena, they can input the second request information of the heating station fault phenomenon in voice or text. After the preset large language model identifies the fault entity and fault intent in the second request information, it obtains the fault entity equipment and professional fault characteristics. Then, it calls the full-process physical mechanism simulation model of the heating station to virtually reproduce the fault phenomenon and analyze the root cause and scope of impact of the fault. S4. Based on the root cause of the fault, the scope of impact, and historical fault handling data, train a large-scale fault handling model for the heating station. After outputting preliminary suggestions for actionable handling corresponding to the fault, call the full-process physical mechanism simulation model of the heating station to verify the rationality of the preliminary suggestions and output the final handling plan.

2. The method for diagnosing faults in a heating station according to claim 1, characterized in that, In S1, the core equipment includes heat exchange equipment, pump and valve equipment, measurement and control and instrumentation equipment, and auxiliary equipment. The core of the heat exchange equipment is the heat exchanger, which is equipped with seals, dirt separators, and heat exchange medium guides. The existing failure modes include heat exchange efficiency decay, structural and sealing failure, local heat transfer abnormality, and dirt separator failure. The physical mechanism models set include a fouling thermal resistance calculation model, a heat exchange model, and a seal aging failure model. The pump and valve equipment includes primary and secondary network circulating pumps, makeup water pumps, electrically adjustable valves, safety valves, pressure reducing valves, and steam traps. The existing failure modes include pump performance degradation, pump malfunction, pump mechanical failure, valve regulation failure, and valve safety protection failure. The physical mechanism models set include pump performance curve models, rotating machinery vibration models, pump net positive suction head (NPSH) calculation models, valve flow coefficient models, safety valve opening pressure calculation models, and steam trap flow calculation models. The measurement and control and instrumentation equipment includes sensors, controllers, communication modules and interlocking protection devices. The existing fault modes include measurement distortion, control failure, and communication and interlocking abnormalities. The physical mechanism models set include sensor error model, control adjustment model and signal transmission error model. The auxiliary equipment includes constant pressure water supply devices, water treatment equipment, air vents, venting valves, pipes and supports. The existing fault modes include constant pressure and water supply failure, air venting failure, and pipe and support failure. The physical mechanism models set include constant pressure water supply pressure calculation model, water hammer effect model, and pipe corrosion model.

3. The method for diagnosing faults in a heating station according to claim 2, characterized in that, In step S1, the physical mechanism models of each core device in the heating station are dynamically coupled to form a full-process physical mechanism simulation model of the heating station, including: Based on the energy flow transmission and processing of each core device in the heating station, the topological connection relationship between the devices is established. Based on the rules of mass conservation, energy conservation, signal interaction, and data transmission, the dynamic coupling logic between the devices is established. Input and output interfaces and simulation time steps of the unified physical mechanism model of each device are defined to form a preliminary simulation model of the physical mechanism of the entire heating station process. Input the normal operating parameters of the heating station and compare the simulation output with the actual operating data to verify the normal operating conditions; input the parameters of a typical single fault mode and compare the simulation output with the actual fault data to verify that the fault feature evolution trend is consistent with reality; input the parameters of a multi-fault coupling scenario to verify the model's simulation capability for the evolution of cascading faults. After verification, a simulation model of the physical mechanism of the entire process of the heating station was formed.

4. The method for diagnosing faults in a heating station according to claim 1, characterized in that, In step S2, a large model is pre-trained using a mixed dataset of the entire heating station scenario to learn the mapping relationship between changes in equipment characteristic parameters and fault evolution knowledge under normal and fault modes of different core equipment, generating a large-scale fault diagnosis model for the heating station, including: The parameters of multiple devices in the mixed dataset of the entire scenario of the heating station are aligned with time data to generate device-level time-series feature sequences. Numerical operating parameters and event-type fault labels are merged, and each data point is labeled with faulty device and fault mode labels to form a standardized dataset. The selected large-scale model of the base is pre-trained using the dataset to learn the dynamic correlation between parameters under normal and fault modes of different core equipment, the fault reasoning logic of the simulation model of the physical mechanism of the whole process of the heating station, and to embed the fault evolution rules to generate a large-scale model for fault diagnosis of the heating station.

5. The method for diagnosing faults in a heating station according to claim 1, characterized in that, In step S3, when heating station personnel perform routine operation and maintenance, they input a first request message (voice or text) for routine operation and maintenance. This first request message is then parsed by a pre-defined large language model and input along with real-time operation monitoring data of the heating station into the heating station fault diagnosis large model for heating station operation diagnosis, including: When heating station personnel perform routine operation and maintenance, they can input the first request information for routine operation and maintenance via voice or text through the input terminal. The preset voice recognition model will then convert the voice information into text information, and the text information will be preprocessed to form the core text of the routine operation and maintenance request. The selected general language model is fine-tuned with small samples using a pre-set corpus of thermal power station operation and maintenance dialogues. At the same time, a lexicon of thermal power station domain and a device-parameter association graph are embedded to form a large language model. After using the first major language model to identify the core intent and extract the relevant physical devices from the core text of daily operation and maintenance requests, real-time operation monitoring data of the relevant physical devices and coupled devices are obtained. By calling the large-scale fault diagnosis model of the heating station, based on the pre-trained full-scenario fault knowledge of the heating station, targeted feature extraction and fault matching are performed. If it is a normal mode, it indicates that the relevant equipment of the heating station is operating normally; if it is a fault mode, it indicates that the relevant equipment of the heating station is faulty. A preliminary fault diagnosis report containing the faulty equipment, fault mode, fault occurrence time and fault parameter deviation value is output.

6. The method for diagnosing faults in a heating station according to claim 5, characterized in that, In S3, when the diagnostic result of the heating station operation is a fault, the full-process physical mechanism simulation model of the heating station is invoked. Based on the faulty equipment and fault mode of the heating station, the root cause and scope of impact of the fault are analyzed, including: The large-scale fault diagnosis model for heating stations maps the core parameters in the preliminary fault diagnosis report to the preset boundary conditions and fault initialization parameters of the full-process physical mechanism simulation model of the heating station. Then, it calls the full-process physical mechanism simulation model of the heating station to dynamically reproduce the occurrence and evolution of the fault. If it is a single fault scenario, it directly simulates the parameter changes and fault evolution process of the faulty equipment before and after the fault. If it is a multi-equipment coupled fault scenario, it simulates the chain triggering and fault evolution process before and after the fault. Perform parameter back-inference and causal analysis on the fault evolution process, locate the root cause of the fault, and output a root cause analysis report, including the basis for root cause determination, the change curve of key mechanism parameters, and the causal relationship between the root cause and the fault. Using a full-process physical mechanism simulation model of a heating station, the chain reaction of a fault from the core equipment to other equipment in the heating station is simulated, and the scope and degree of the impact are quantitatively analyzed, outputting a quantitative analysis report on the scope of impact.

7. The method for diagnosing faults in a heating station according to claim 1, characterized in that, In step S3, when heating station personnel are diagnosing and analyzing a fault, they input a second request message (voice or text) regarding the fault. A pre-defined large language model then identifies the fault entity and fault intent in the second request message, obtaining the faulty equipment and its specific characteristics. Finally, a full-process physical mechanism simulation model of the heating station is invoked to virtually reproduce the fault phenomenon and analyze its root causes and impact scope, including: When heating station personnel conduct diagnostic analysis of existing faults, they can input the second request information of the heating station fault symptoms via voice or text. The preset large language model will then extract and classify entities, locate the core physical equipment corresponding to the fault symptoms, and transform the fault symptoms into professional fault characteristics that can be quantified and described. The simulation model of the physical mechanism of the heating station is called, and the normal operating parameters of the heating station before the failure are loaded from the historical operation database as the basic operating parameters of the simulation model. Then, based on the preset professional fault feature-simulation physical parameter mapping library, the professional fault features are converted into corresponding simulation physical parameters. The initialization parameters and simulation boundary conditions corresponding to the fault phenomenon are set, and the simulation reproduction of the entire process of fault phenomenon evolution is started, including the initial triggering of the fault, the manifestation of the fault phenomenon, and the steady-state operation of the fault. The simulation-reproduced fault phenomena are quantitatively matched with the actual fault phenomena described by the maintenance personnel. If the matching degree is greater than the threshold, the root cause analysis and impact range analysis stages are entered. If the matching degree is less than the threshold, the fault initialization parameters of the simulation model are adjusted and the simulation is repeated until the matching degree meets the standard. Perform parameter back-inference and causal analysis on the entire process of fault phenomenon evolution, locate the root cause of the fault, and output a root cause analysis report, including the basis for root cause determination, the change curve of key mechanism parameters, and the causal relationship between the root cause and the fault. Using a full-process physical mechanism simulation model of a heating station, the chain reaction of a fault from the core equipment to other equipment in the heating station is simulated, and the scope and degree of the impact are quantitatively analyzed, outputting a quantitative analysis report on the scope of impact.

8. The method for diagnosing faults in a heating station according to claim 1, characterized in that, S4 includes: Data on fault causes, fault impact range, and historical fault handling are obtained. The selected basic large language model is pre-trained to allow the model to learn the mapping relationship between fault information and handling measures. At the same time, the model is fine-tuned and optimized by embedding handling phenomenon constraints and high-risk handling measures to form a large-scale model for handling faults in heating stations. Based on the current root cause and scope of impact of the heating station fault handling model, the current fault information is matched with historical faults for similarity. Priority is given to matching historical cases with high consistency in fault equipment and fault mode. Then, effective handling measures without secondary faults are selected. In addition, the selected handling measures are optimized and adjusted according to the scope of impact of the current fault to adapt to the on-site working conditions and output preliminary fault handling suggestions. Using a pre-defined handling suggestion-simulation parameter mapping library, the initial fault handling suggestions are analyzed, and the pre-handling operating parameters, post-handling operating parameters, and dynamic parameters of the handling process are extracted. The full-process physical mechanism simulation model of the heating station is then called to perform full-process simulation before, during, and after the handling. The steady-state operating conditions before and after the handling are compared to evaluate the fault elimination effect and heating compliance. The dynamic parameter changes during the handling are simulated to evaluate the feasibility, stability, and whether secondary faults will be caused by the handling operation. The time-series curves of parameter changes for each process and the final handling plan are output.

9. The method for diagnosing faults in a heating station according to claim 8, characterized in that, The fault cause data includes the fault cause, faulty equipment, fault mode, and deviation parameters; the fault impact range includes the impact on the operation of related equipment, the impact on the heating range, the impact on heating quality, and secondary faults; the historical fault handling data includes the handling measures and handling effects for different fault modes.

Images