A combined cycle unit state prediction and fault diagnosis method

Abstract

The application provides a combined cycle unit state prediction and fault diagnosis method, belonging to the intelligent operation and maintenance technical field of combined cycle units, which can significantly alleviate or solve the problems of disconnection between offline simulation and actual working conditions, insufficient state prediction accuracy, lagging fault early warning, and poor closed-loop regulation in the prior art. The method comprises: constructing a digital twin model comprising a three-dimensional physical model, a mathematical model, and a simulation model; collecting and preprocessing multiple source operation data of the unit to form a standardized real-time data set; dynamically updating and bias checking the digital twin model based on the real-time data set; combining long short-term memory network and digital twin simulation for state prediction, and performing fault diagnosis based on parameter bias and fault feature matching; generating control instructions and maintenance plans according to the prediction results and diagnosis results to realize closed-loop regulation. The method can improve the accuracy of state prediction and fault diagnosis, realize early fault warning, and improve the safety and reliability of unit operation.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance technology for combined cycle units, and specifically relates to a method for predicting the status and diagnosing faults of combined cycle units. Background Technology

[0002] Gas-steam combined cycle (GSCB) units are a type of important power generation equipment characterized by both high efficiency and low emissions, widely used in power generation, industrial park energy supply, and integrated energy utilization. As energy systems evolve towards higher efficiency, cleaner operation, and smarter systems, GSCB units are playing an increasingly vital role in grid peak shaving, rapid load response, and improving energy efficiency. Against this backdrop, accurately monitoring the unit's health status under complex operating conditions, promptly identifying abnormal signs, and mitigating the risk of unplanned outages have become critical technical challenges that urgently need to be addressed in the field of intelligent operation and maintenance of power equipment.

[0003] For gas-steam combined cycle units, core components such as the gas turbine, waste heat boiler, and steam turbine are highly coupled and their operation is closely interrelated. Parameters such as temperature, pressure, flow rate, speed, and vibration collectively determine the unit's power generation efficiency, operational safety, and operating costs. As the unit ages, problems such as blade wear, boiler scaling, and abnormal rotor vibration tend to accumulate. If potential faults are not detected early and targeted measures are not taken, it often leads to decreased equipment efficiency, and in severe cases, even shutdowns for maintenance, resulting in significant economic losses. Therefore, establishing a method that combines unit operating mechanisms and real-time operating data to predict unit status and provide early warnings of potential faults has significant engineering application value.

[0004] In existing technologies, two main approaches are typically used for state prediction and fault diagnosis of combined cycle power units: one is based on real-time data monitoring, and the other is based on offline simulation. The former relies primarily on field-collected operational data for threshold judgment, trend monitoring, or empirical analysis. While it offers some real-time capability, it often lacks a deep understanding of the unit's operating patterns and equipment coupling mechanisms, making it difficult to predict potential faults proactively. In many cases, problems can only be identified after anomalies have become apparent or even after a fault has occurred, resulting in delayed response. The latter analyzes equipment operating status through mathematical or simulation models, reflecting the unit's mechanistic characteristics to some extent. However, traditional offline simulations are often disconnected from the real-time operating conditions of the actual unit, failing to reflect the impact of operating environment, load changes, and equipment aging in a timely manner, thus resulting in poor timeliness and accuracy of simulation results. In recent years, digital twin technology, as an emerging technology integrating physical modeling, mathematical simulation, and data-driven approaches, has provided a new path for the intelligent operation and maintenance of complex industrial equipment. This technology can construct virtual models corresponding to physical equipment and achieve synchronous updates between the virtual model and the physical equipment through data interaction. However, current solutions for applying digital twin technology to gas-steam combined cycle units generally suffer from problems such as insufficient model accuracy, inadequate integration of real-time data and mechanistic models, insufficient utilization of fault characteristics, and weak fault prediction capabilities. They are still unable to balance real-time performance, accuracy, and feasibility, and cannot fully meet the needs of intelligent operation and maintenance throughout the entire life cycle of combined cycle units.

[0005] To address this, a method for combined cycle unit condition prediction and fault diagnosis is proposed. Summary of the Invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art, and to provide a method for predicting the condition and diagnosing faults of combined cycle units.

[0007] This invention provides a method for predicting the condition and diagnosing faults in combined cycle power units, comprising the following steps: S1: Collect multi-source operating data of the gas-steam combined cycle unit, preprocess the multi-source operating data, and generate a standardized real-time dataset aligned with timestamps; S2: Input the standardized real-time dataset into the digital twin model of the gas-steam combined cycle unit. The digital twin model includes a three-dimensional physical model, a mathematical model, and a simulation model. Based on the standardized real-time dataset, dynamically update the parameters of the mathematical model in the digital twin model, and periodically verify the deviation between the simulation data output by the digital twin model and the real-time operating data of the gas-steam combined cycle unit. When the deviation value exceeds a preset threshold, trigger the model calibration process to ensure that the digital twin model and the gas-steam combined cycle unit maintain a synchronous mapping relationship. S3: Based on the digital twin model after the synchronization mapping is completed, the unit status is predicted to obtain the unit status prediction result; and, based on the parameter deviation between the simulation data and the real-time operating data of the gas-steam combined cycle unit, fault diagnosis is performed and the fault diagnosis result is output. S4. Generate unit operation and maintenance optimization suggestions based on the state prediction results and the fault diagnosis results, and convert the unit operating parameter adjustment suggestions in the unit operation and maintenance optimization suggestions into control commands and send them to the control system of the gas-steam combined cycle unit to adjust the unit operating parameters; and feed back the adjusted unit operating data to the digital twin model to realize closed-loop update of unit state prediction, fault diagnosis and operation control.

[0008] Further, in step S1, the digital twin model is constructed in the following manner, including: constructing a three-dimensional physical model of the gas turbine, waste heat boiler, steam turbine, and auxiliary equipment based on the design drawings, equipment parameters, and connection relationships of the gas-steam combined cycle unit; establishing thermodynamic equations, kinetic equations, and heat and mass transfer equations for the gas turbine, waste heat boiler, and steam turbine respectively, and coupling them to form a mathematical model of the entire unit; and coupling the three-dimensional physical model with the mathematical model of the entire unit, and constructing a simulation model in combination with preset operating boundary conditions.

[0009] Specifically, in step S1, the multi-source operating data includes at least sensor data, operating log data, and environmental parameter data; the sensor data includes temperature, pressure, flow rate, speed, and vibration parameters of the gas turbine, waste heat boiler, steam turbine, and auxiliary equipment; the operating log data includes start-up and shutdown records, load adjustment records, and operating condition switching records; and the environmental parameter data includes ambient temperature, humidity, and atmospheric pressure. The multi-source operational data is preprocessed, including outlier removal, missing value imputation and normalization, and sorted and integrated according to timestamps to generate the standardized real-time dataset.

[0010] Specifically, in step S2, the dynamic updating of the model parameters and the deviation verification of the mathematical model include: Input the standardized real-time dataset into the digital twin model; Based on the standardized real-time dataset, the parameters of the mathematical model are dynamically adjusted using incremental learning, and the running status of the three-dimensional physical model is updated synchronously; and The simulation data output by the digital twin model is periodically compared with the real-time operating data of the gas-steam combined cycle unit to calculate the deviation value. When the deviation value is greater than a preset threshold, the model calibration process is triggered to adjust the model parameters in the digital twin model.

[0011] Preferably, in step S3, the unit status prediction includes: Historical operating data is input into a pre-trained long short-term memory network model to obtain the predicted values ​​of the target operating parameters of the gas-steam combined cycle unit within the target time interval; Based on the digital twin model after synchronization mapping, the operating state of the gas-steam combined cycle unit is simulated within the target time interval, and the predicted values ​​of simulation parameters within the target time interval are obtained; and The predicted values ​​of the target operating parameters and the predicted values ​​of the simulation parameters are fused to obtain the state prediction results of the gas-steam combined cycle unit within the target time interval.

[0012] Specifically, in step S3, the fault diagnosis includes: Obtain the parameter deviation values ​​corresponding to each operating parameter of the gas-steam combined cycle unit; When the deviation value of any of the operating parameters exceeds the corresponding deviation threshold, the parameter deviation value is input into a pre-built fault feature library and matched with the fault feature parameters, fault characterization information, and fault evolution patterns stored in the fault feature library to determine the fault type and fault location; and Based on the determined fault type and fault location, the fault evolution process is simulated using the digital twin model to obtain fault diagnosis results including fault type, fault location, fault evolution result, and warning level.

[0013] Furthermore, the fault feature database stores characteristic parameters, fault characterization information, and fault evolution patterns, including those related to gas turbine blade wear, waste heat boiler scaling, and abnormal steam turbine rotor vibration. The fault feature database employs a dynamic update mechanism, periodically updating the fault feature parameters based on historical fault data, maintenance records, and newly discovered fault cases of the gas-steam combined cycle unit.

[0014] Furthermore, in step S4, the generated unit operation and maintenance optimization suggestions include fault handling plans, unit operating parameter adjustment suggestions, and maintenance plans.

[0015] Furthermore, the prediction and fault diagnosis method also includes a digital twin display step based on the three-dimensional physical model. The digital twin display step is used to display the unit's operating status, status prediction results, and fault diagnosis results in real time, and supports fault location, historical data query, and simulation process playback.

[0016] Specifically, the digital twin model also includes a life prediction sub-model, which is used to predict the remaining life of unit components and the whole unit based on unit operation data, fault diagnosis results and simulation data, and to provide a basis for maintenance plan formulation based on the prediction results.

[0017] The beneficial effects of this invention are as follows: This invention achieves real-time synchronization between the virtual model and the physical unit of a combined cycle unit by constructing a high-precision digital twin model and combining it with multi-source real-time data; it enables high-precision prediction of unit operating status by combining digital twin simulation with long short-term memory networks; it achieves advanced prediction and graded early warning of potential faults through a diagnostic mechanism based on deviation identification, fault feature matching, and fault evolution simulation; it achieves closed-loop optimization of unit operation control by feeding the prediction and diagnostic results back to the control system; and it improves the safety, reliability, economy, and operation and maintenance efficiency of combined cycle units by combining visualization and scalable life prediction capabilities. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the steps of a combined cycle unit condition prediction and fault diagnosis method according to a specific embodiment of the present invention. Figure 2 This is an overall flowchart of a combined cycle unit condition prediction and fault diagnosis method according to a specific embodiment of the present invention. Figure 3 This is a schematic diagram illustrating the construction process and structure of a digital twin model for a combined cycle unit condition prediction and fault diagnosis method according to a specific embodiment of the present invention. Figure 4 This is a schematic diagram of the multi-source real-time data acquisition and preprocessing process of a combined cycle unit condition prediction and fault diagnosis method according to a specific embodiment of the present invention. Figure 5 This is a schematic diagram of the fault diagnosis system of the combined cycle unit condition prediction and fault diagnosis method according to a specific embodiment of the present invention. Figure 6This is a schematic diagram of a digital twin visualization interface for a combined cycle unit condition prediction and fault diagnosis method according to a specific embodiment of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] like Figure 1 As shown in the figure, a method for predicting the condition and diagnosing faults of a combined cycle power unit provided by a specific embodiment of the present invention includes the following steps: S1: Collect multi-source operating data of the gas-steam combined cycle unit, preprocess the multi-source operating data, and generate a standardized real-time dataset aligned with timestamps; S2: Input the standardized real-time dataset into the digital twin model of the gas-steam combined cycle unit. The digital twin model includes a three-dimensional physical model, a mathematical model, and a simulation model. Based on the standardized real-time dataset, dynamically update the parameters of the mathematical model in the digital twin model, and periodically check the deviation between the simulation data output by the digital twin model and the real-time operating data of the gas-steam combined cycle unit. When the deviation value exceeds a preset threshold, trigger the model calibration process to ensure that the digital twin model and the gas-steam combined cycle unit maintain a synchronous mapping relationship. S3: Based on the digital twin model after synchronization mapping, perform unit status prediction to obtain unit status prediction results; and, based on the parameter deviation between simulation data and real-time operating data of the gas-steam combined cycle unit, perform fault diagnosis and output fault diagnosis results. S4. Generate unit operation and maintenance optimization suggestions based on the state prediction results and the fault diagnosis results, and convert the unit operating parameter adjustment suggestions in the unit operation and maintenance optimization suggestions into control commands and send them to the control system of the gas-steam combined cycle unit to adjust the unit operating parameters; and feed back the adjusted unit operating data to the digital twin model to realize closed-loop update of unit state prediction, fault diagnosis and operation control.

[0021] In one implementation, incremental learning is used to update the mathematical model parameters in the digital twin model. These updated parameters include coefficients of the thermodynamic equations, boundary conditions of the kinetic equations, and model parameters related to heat and mass transfer processes. Specifically, for a gas turbine mathematical model, the correlation coefficients in the combustion equations of the combustion chamber, the compression equations of the compressor, and the turbine expansion equations can be updated; for a waste heat boiler mathematical model, the heat transfer coefficient, working fluid flow parameters, and parameters related to heat loss can be updated; for a steam turbine mathematical model, the turbine expansion parameters, thermodynamic process parameters, and kinetic parameters related to vibration characteristics can be updated. Through incremental updates based on real-time data, model parameters can be gradually corrected without completely rebuilding the mathematical model, allowing the digital twin model to continuously track changes in the actual operating conditions of the unit. The combined cycle unit state prediction and fault diagnosis method takes a gas-steam combined cycle unit as the object. By constructing a digital twin model corresponding to the unit entity and combining it with multi-source real-time data during unit operation, it performs real-time mapping, state prediction, and fault diagnosis of the unit's operating state, thereby achieving dynamic monitoring of the unit's operating state, early identification of potential faults, and closed-loop control of operating parameters. This method improves the real-time performance of unit status awareness, the accuracy of fault diagnosis, and the targeted nature of operation and maintenance decisions. The incremental learning approach can be implemented using an online parameter update method based on a sliding time window. Specifically, using the latest standardized real-time dataset of a preset time length as the input for the current batch, key operating parameters such as temperature, pressure, flow rate, speed, vibration, and load are extracted. These key operating parameters are then compared with the corresponding simulation parameters output by the digital twin model to construct a parameter update objective function. This objective function characterizes the comprehensive deviation between the simulated and measured values ​​of each key operating parameter, and different operating parameters can be assigned different weights based on their impact on unit safety, economy, and stability. After the deviation calculation for the current batch of data is completed, the combustion coefficient, compressor compression coefficient, and turbine expansion coefficient in the gas turbine mathematical model, the heat transfer coefficient, flow resistance coefficient, and heat loss coefficient in the waste heat boiler mathematical model, and the thermodynamic process parameters, expansion parameters, and vibration dynamics parameters in the steam turbine mathematical model are corrected online. The corrected parameters are then re-inputted into the digital twin model to generate simulation data for the next time step. By continuously executing the cyclical process of "data input—deviation calculation—parameter correction—simulation output," the digital twin model can continuously track the actual operating status of the unit under conditions of load fluctuations, environmental changes, and component performance degradation.

[0022] Preferably, the digital twin model is constructed by coupling a three-dimensional physical model, a mathematical model, and a simulation model. The three-dimensional physical model is used to characterize the structural composition and connection relationship of the gas-steam combined cycle unit, the mathematical model is used to characterize the thermodynamics, dynamics, and heat and mass transfer laws of each piece of equipment in the unit under different operating conditions, and the simulation model is used to perform numerical simulation of the unit's operating state and output the results, so as to improve the digital twin model's ability to characterize the actual operating state of the unit.

[0023] Specifically, the multi-source operational data includes at least sensor data, operational log data, and environmental parameter data. The sensor data is used to characterize the temperature, pressure, flow rate, speed, and vibration status of the gas turbine, waste heat boiler, steam turbine, and auxiliary equipment. The operational log data is used to record the unit start-up and shutdown, load adjustment, and operating condition switching processes. The environmental parameter data is used to characterize external operating conditions such as ambient temperature, humidity, and atmospheric pressure, providing a data foundation for the real-time updating and status prediction of the digital twin model.

[0024] Furthermore, the unit status prediction and fault diagnosis methods form a closed-loop management process; during operation, the digital twin model is continuously updated based on real-time collected data, and generates corresponding operation and maintenance suggestions and control commands based on the prediction and diagnosis results, so that the unit is always in a safe, stable and efficient operating state.

[0025] Based on the above basic implementation method, in step S1, the digital twin model is constructed in the following ways: a three-dimensional physical model of the gas turbine, waste heat boiler, steam turbine, and auxiliary equipment is constructed based on the design drawings, equipment parameters, and connection relationships of the gas-steam combined cycle unit; thermodynamic equations, kinetic equations, and heat and mass transfer equations are established for the gas turbine, waste heat boiler, and steam turbine respectively, and coupled to form a mathematical model of the entire unit; and the three-dimensional physical model is coupled with the mathematical model of the entire unit, and a simulation model is constructed in combination with preset operating boundary conditions to realize the operation simulation under the scenarios of unit start-up and shutdown, load adjustment, and operating condition switching.

[0026] Furthermore, when constructing a three-dimensional physical model, the gas turbine, waste heat boiler, steam turbine, and auxiliary equipment can be modeled in three dimensions based on the unit's structural dimensions, material parameters, and equipment connection relationships. The equipment location, parameter type, and connection interface can be marked in the three-dimensional model to facilitate the subsequent visualization of the equipment's operating status and data interaction with the mathematical model.

[0027] In one specific implementation, in step S1, the multi-source operating data includes at least sensor data, operating log data, and environmental parameter data; the sensor data includes temperature, pressure, flow rate, speed, and vibration parameters of the gas turbine, waste heat boiler, steam turbine, and auxiliary equipment; the operating log data includes start-up and shutdown records, load adjustment records, and operating condition switching records; the environmental parameter data includes ambient temperature, humidity, and atmospheric pressure. The multi-source operational data is preprocessed, including outlier removal, missing value imputation and normalization, and sorted and integrated according to timestamps to generate the standardized real-time dataset.

[0028] In this embodiment, outlier removal can be performed using the 3σ criterion to detect the collected sensor data, identifying and removing data that exceeds the normal operating range or undergoes abrupt changes as outliers; missing value imputation can be performed using linear interpolation or mean imputation to complete the missing data; normalization can be performed using the min-max normalization algorithm to map data of different dimensions such as temperature, pressure, flow rate, rotational speed, and vibration to a unified data range, thereby reducing the impact of dimensional differences on subsequent model training and simulation calculations.

[0029] Furthermore, after preprocessing, multi-source operational data can be sorted and integrated according to a unified timestamp to form a standardized real-time dataset. This allows sensor data, operational log data, and environmental parameter data to correspond under the same time reference, facilitating the dynamic updating of the digital twin model and the analysis of discrepancies between simulation data and real-time operational data.

[0030] In another specific embodiment, step S2, the dynamic updating of the model parameters and the deviation verification of the mathematical model include: A standardized real-time dataset is input into the digital twin model. Based on the standardized real-time dataset, the parameters of the mathematical model are dynamically adjusted using incremental learning, and the operating status of the 3D physical model is updated synchronously. The simulation data output by the digital twin model is periodically compared with the real-time operating data of the gas-steam combined cycle unit to calculate the deviation value. When the deviation value exceeds a preset threshold, a model calibration process is triggered to adjust the model parameters in the digital twin model. The mechanism for triggering model calibration based on the deviation threshold includes: comparing the simulation data output by the digital twin model with the real-time operating data of the gas-steam combined cycle unit item by item according to a preset time period, and calculating the deviation value corresponding to each operating parameter; when the deviation value of any operating parameter exceeds the preset deviation threshold corresponding to that operating parameter, it is determined that there is a mapping deviation between the current digital twin model and the actual operating status of the unit, and the model calibration process is triggered. The model calibration process is used to correct the mathematical model parameters in the digital twin model according to the deviation value, so as to reduce the difference between the simulation output and the real-time operating data, and restore the synchronous mapping relationship between the digital twin model and the actual operating conditions of the unit. Preferably, the deviation verification is performed at a fixed time period, for example, once every 10 minutes. In one example, the deviation threshold can be set to 5%, or different thresholds can be set for different parameters, wherein the temperature parameter deviation threshold can be set to ±3℃, and the vibration parameter deviation threshold can be set to ±0.5mm / s. The model calibration process includes the following sub-steps: First, based on the current deviation values ​​of each operating parameter, determine the target device and target parameter whose deviation exceeds the limit; second, corresponding to the target device, select the set of parameters to be calibrated related to the target parameter from the digital twin model; then, with the goal of reducing the deviation between the simulated value and the measured value of the target parameter, perform local correction on the set of parameters to be calibrated; finally, re-substitute the corrected parameters into the digital twin model for simulation verification. When the calibrated deviation value falls back to the preset threshold range, the current parameter is kept as the valid model parameter for subsequent operating cycles; when the calibrated deviation value is still higher than the preset threshold, the next round of parameter correction is performed, or the pre-stored initial model parameters are called as a reference for recalibration; to avoid frequent calibration triggered by a single abnormal data, the model calibration process can be set with continuous triggering conditions. Specifically, a model is only considered to have a stability mapping deviation and is calibrated when the same operating parameter exceeds the corresponding deviation threshold for multiple consecutive calibration cycles. If the parameter exceeds the limit only briefly within a single calibration cycle, the abnormal event is simply recorded and monitoring continues. This approach reduces the impact of instantaneous disturbances, communication fluctuations, or occasional measurement errors on the stability of model calibration.

[0031] In this embodiment, the standardized real-time dataset can be input into the digital twin model in real time via a data transmission channel to achieve synchronous updates between the digital twin model and the operating status of the gas-steam combined cycle unit; the data transmission channel can adopt a 5G communication link, industrial Ethernet, or other communication methods suitable for real-time data transmission in industrial sites.

[0032] Specifically, the incremental learning method is used to gradually correct the parameters in the mathematical model based on standardized real-time data collected in real time, without completely reconstructing the model, so that the digital twin model can continuously track the changes in the actual operating conditions of the unit; at the same time, the equipment temperature distribution, vibration status and operating condition display parameters in the three-dimensional physical model are also updated accordingly to reflect the current operating status of the unit.

[0033] Furthermore, deviation verification can be performed according to a preset time period. When the deviation between the simulation data and the real-time operating data exceeds a preset threshold, the system initiates the model calibration process to correct the mathematical model parameters or boundary conditions, thereby improving the mapping accuracy and synchronization between the digital twin model and the gas-steam combined cycle unit.

[0034] In another specific embodiment, in step S3, the unit state prediction includes: Historical operating data is input into a pre-trained Long Short-Term Memory (LSTM) network model to obtain predicted values ​​of target operating parameters for the gas-steam combined cycle unit within a target time interval. Based on the digital twin model after synchronization mapping, the operating state of the gas-steam combined cycle unit within the target time interval is simulated to obtain predicted values ​​of simulation parameters within the target time interval. The predicted values ​​of target operating parameters and simulation parameters are then fused to obtain the state prediction result of the gas-steam combined cycle unit within the target time interval. The fusion rule between the predicted values ​​of target operating parameters and simulation predictions includes: for the same target operating parameter within the same target time interval, the predicted values ​​of target operating parameters output by the LSM network model and the simulation parameters output by the digital twin simulation model are obtained separately, and the two are weighted and fused to obtain the final state prediction result. Preferably, the weighted fusion can be expressed as: Final predicted value = α × Target operating parameter predicted value + β × Simulation predicted value Wherein, α and β are fusion weights, and α + β = 1. Further, the fusion weights can be dynamically set according to the current deviation between the digital twin model and the real-time operating data of the unit: when the current mapping deviation is small, the weight of the simulation prediction value is increased; when the real-time data fluctuates greatly or the operating conditions change frequently, the weight of the predicted value of the target operating parameter is increased. In this way, the sensitivity of the model to temporal changes and the constraint effect of the digital twin simulation model on the consistency of physical mechanisms can be considered. The input to the Long Short-Term Memory (LSTM) network model can be a sequence of multivariate operating parameters arranged in chronological order, which includes at least some or all of the parameters from unit load, power generation efficiency, temperature, pressure, flow rate, speed, vibration, and environmental parameters. The output of the LSM network model is the predicted value of one or more target operating parameters within the target time interval. To ensure the comparability of data-driven prediction results and digital twin simulation results, the predicted values ​​of target operating parameters and simulation parameters are uniformly processed using the same parameter definitions, time granularity, and dimensional standards before fusion processing. The fusion weights α and β can be dynamically set based on the current mapping deviation, and can also be differentiated according to the parameter type of the predicted object. Specifically, for parameters significantly affected by rapid fluctuations in operating conditions, the weight of the predicted target operating parameters output by the Long Short-Term Memory (LSTM) network model is increased; for parameters strongly constrained by equipment mechanisms and exhibiting clear physical laws, the weight of the predicted simulation parameters output by the digital twin simulation model is increased. By setting fusion weights separately for different parameters, the sensitivity of the data-driven model to temporal fluctuations and the constraint effect of the mechanism simulation model on physical consistency can be balanced, thereby improving the stability and reliability of the state prediction results.

[0035] Furthermore, the Long Short-Term Memory Network model is used to extract the temporal correlation features of parameter changes from the unit's historical and real-time operating data. The digital twin model is used to perform mechanistic simulation of the unit's operating state within the target time interval based on the current synchronous operating conditions. The fusion of the two can take into account both the temporal sensitivity of data-driven prediction and the physical consistency of mechanistic model simulation, thereby improving the reliability of the state prediction results.

[0036] Specifically, the target time interval can be any period from 1 hour to 24 hours in the future. The state prediction objects can include unit load, power generation efficiency, and temperature, pressure and vibration parameters of gas turbine blades, steam turbine rotors and waste heat boiler heat exchange components. The state prediction results can include parameter change trend curves and operation stability assessment results to provide a basis for operation and maintenance personnel to formulate control plans.

[0037] In another specific embodiment, in step S3, fault diagnosis includes: The system acquires the parameter deviation values ​​corresponding to each operating parameter of the gas-steam combined cycle unit. When the parameter deviation value of any operating parameter exceeds the corresponding deviation threshold, the parameter deviation value is input into a pre-built fault feature library and matched with the fault feature parameters, fault characterization information, and fault evolution laws stored in the fault feature library to determine the fault type and fault location. Based on the determined fault type and fault location, the fault evolution process is simulated using a digital twin model to obtain fault diagnosis results including fault type, fault location, fault evolution result, and warning level. The fault feature library stores information including gas turbine blade wear... The fault feature database includes characteristic parameters, fault characterization information, and fault evolution patterns of gas turbine blade wear, waste heat boiler scaling, and abnormal steam turbine rotor vibration. It employs a dynamic update mechanism, periodically updating fault characteristic parameters based on historical fault data, maintenance records, and newly discovered fault cases from the gas-steam combined cycle unit. The database establishment process includes: collecting historical fault data, maintenance records, and known fault cases from the gas-steam combined cycle unit; extracting characteristic parameters, fault characterization information, and fault evolution patterns corresponding to different fault types; and classifying and storing the database according to fault type, fault location, and characteristic parameter category. Preferably, the fault feature database includes at least common fault types such as gas turbine blade wear, waste heat boiler scaling, and abnormal steam turbine rotor vibration, and can be expanded to 15 common fault categories. Furthermore, the database employs a dynamic update mechanism, periodically updating based on newly discovered fault cases, the latest maintenance records, and newly added historical fault data, for example, quarterly, to improve the accuracy and adaptability of fault diagnosis. The criteria for fault feature matching include: parameter deviation values, fault characteristic parameters, fault characterization information, and fault evolution patterns. Specifically, the parameter deviation value is obtained by comparing the simulation data output by the digital twin model with the real-time operating data of the unit item by item; the fault characteristic parameters are used to characterize the parameter change characteristics corresponding to different fault types; the fault characterization information is used to characterize the manifestation of the fault at the equipment level; and the fault evolution law is used to characterize the development trend of the fault in the future time period. During fault matching, the correspondence between the parameter deviation value and the fault characteristic parameter is used as the basic basis, the fault characterization information is used as the auxiliary verification basis, and the fault evolution law is used as the reference basis for subsequent early warning level classification and fault trend simulation; each type of fault feature in the fault feature library can store at least the following information: fault type identifier, location of the fault-occurring equipment, key characteristic parameters, normal and abnormal ranges of characteristic parameters, parameter duration requirements, combined characteristic conditions, corresponding fault characterization information, fault evolution law, and suggested handling measures.The key feature parameters may include single parameter features or combined features composed of multiple operating parameters; the combined feature conditions are used to characterize the matching criteria when a certain type of fault needs to meet the deviation relationship of multiple parameters at the same time; when performing fault matching, the deviation value and deviation duration of each operating parameter are first calculated, and then the deviation value is compared with each type of fault feature in the fault feature library. When the deviation of a certain operating parameter falls within the preset fault characteristic range and the duration of the deviation reaches the corresponding minimum duration requirement, the operating parameter is determined to meet the first-level matching condition. When multiple operating parameters simultaneously meet the combined characteristics corresponding to the same fault type, the second-level matching condition is determined to be met. When both the first-level and second-level matching conditions are met, and the fault characterization information is consistent with the current equipment operating status, the corresponding fault type and fault location are determined. In one example, when the temperature deviation of the gas turbine blades continues to rise, accompanied by an increase in the deviation of the rotor vibration parameters, it can be used as a candidate feature for gas turbine blade wear faults. When the temperature difference deviation of the heat exchange components of the waste heat boiler increases and the flow parameters change abnormally, it can be used as a candidate feature for waste heat boiler scaling faults. When the vibration amplitude of the steam turbine rotor continuously exceeds the corresponding threshold and the speed fluctuation increases, it can be used as a candidate feature for abnormal steam turbine rotor vibration faults. Through the above methods, the fault matching rules can have clear judgment criteria.

[0038] Furthermore, parameter deviation values ​​can be obtained by comparing the simulation data output by the digital twin model with the real-time operating data of the gas-steam combined cycle unit item by item. Different operating parameters correspond to different deviation thresholds. When the deviation value of a certain operating parameter exceeds the corresponding threshold, it can be determined that the operating parameter is abnormal and trigger the subsequent fault feature matching process. The judgment rules for fault feature matching include: first, obtaining the parameter deviation value corresponding to each operating parameter; then, comparing the parameter deviation value with the feature parameters corresponding to various faults in the fault feature library item by item; when the parameter deviation value and the feature parameter corresponding to a certain fault type meet the preset matching conditions, it is determined that the current unit has an abnormality corresponding to that fault type. Preferably, the preset matching conditions include at least one of the following: the parameter deviation value falls within the feature parameter range corresponding to the fault type, multiple parameter deviation values ​​simultaneously meet the combined features corresponding to the fault type, and the parameter deviation duration meets the preset duration requirement. Furthermore, when multiple fault types meet the matching conditions, the matching priority can be determined according to the number of matching parameters, the magnitude of the deviation, and the parameter duration, and the fault type with the highest priority is selected as the diagnostic result.

[0039] Specifically, the fault feature database can pre-store characteristic parameters, fault characterization information, and fault evolution patterns of common faults in combined cycle units. During fault diagnosis, parameter deviation values ​​can be compared with corresponding features in the fault feature database to determine the corresponding fault type and its location in the equipment. Subsequently, a digital twin model is used to simulate the evolution of the fault during future operation, and warning levels are classified according to the degree of fault impact to output fault diagnosis results. The warning level classification can be determined by comprehensively considering the parameter deviation amplitude, deviation duration, the importance of the equipment affected by the fault, and the fault evolution simulation results. Specifically, when the fault only manifests as slight abnormalities in local parameters, and the fault evolution simulation results show that it will not have a significant impact on the overall operational safety of the unit in the short term, it is classified as a general warning; when the fault has caused multiple key parameters to remain abnormal, and the fault evolution simulation results show that it may have an adverse impact on the unit's efficiency, stability, or component life within a preset time period, it is classified as an important warning; when the fault has approached or reached the protection boundary, or the fault evolution simulation results show that continued operation may lead to shutdown, equipment damage, or safety risks, it is classified as an emergency warning; different warning levels correspond to different handling priorities. General early warnings correspond to enhanced monitoring and parameter fine-tuning; important early warnings correspond to load limiting, operating condition adjustment, or scheduling upcoming maintenance; and emergency early warnings correspond to implementing protective control measures, shutdown strategies, or initiating maintenance procedures. By linking early warning levels with handling priorities, fault diagnosis results can directly serve subsequent operation and maintenance decisions and closed-loop control.

[0040] In another specific embodiment, in step S4, the generated unit operation and maintenance optimization suggestions include fault handling plans, unit operating parameter adjustment suggestions, and maintenance plans; closed-loop control includes converting the unit operating parameter adjustment suggestions in the unit operation and maintenance optimization suggestions into control commands and sending them to the unit control system to adjust the unit operating parameters, and feeding back the adjusted unit operating data to the digital twin model to achieve closed-loop updates of state prediction, fault diagnosis, and operation control.

[0041] Furthermore, the fault handling plan may include emergency shutdown for maintenance, reduced load operation, component inspection, or switching of operating conditions; the unit operating parameter adjustment suggestions may include load adjustment, intake air temperature adjustment, valve opening adjustment, or other parameter optimization schemes that affect the unit's operating status; the maintenance plan may include component replacement time, maintenance items, and maintenance cycle arrangements.

[0042] Furthermore, after the control command is issued, the unit's operating status after the control command is executed can be continuously tracked, and the adjusted operating status can be simulated and verified using a digital twin model. If there is still a deviation between the simulation results and the real-time operating data, or if abnormal parameters are still detected, the status prediction, fault diagnosis, and operation control steps can be repeated to continuously optimize the unit's operating status.

[0043] In one specific implementation, the prediction and fault diagnosis method further includes a digital twin display step based on a three-dimensional physical model. This digital twin display step is used to display the unit's operating status, status prediction results, and fault diagnosis results in real time, and supports fault location, historical data query, and simulation process playback. The digital twin model also includes a life prediction sub-model, which is used to predict the remaining lifespan of unit components and the entire unit based on unit operating data, fault diagnosis results, and simulation data, and provides a basis for maintenance plan formulation based on the prediction results. The process of mapping diagnostic results into control commands includes: based on the fault type, fault location, fault evolution results, and warning level in the fault diagnosis results, calling pre-set handling rules to generate corresponding unit operation and maintenance optimization suggestions; then converting the unit operation and maintenance optimization suggestions into control commands recognizable by the unit control system, and sending them to the unit control system for execution via an interface machine. Preferably, when the warning level is a general warning, the control command includes reducing part of the operating load or adjusting the intake air temperature; when the warning level is a critical warning, the control command includes further limiting the load, adjusting the valve opening, or switching the operating condition; when the warning level is an emergency warning, the control command includes executing shutdown protection or entering the maintenance process. Further, after the control command is executed, the adjusted unit operating data is acquired and re-input into the digital twin model to achieve control effect verification and closed-loop optimization; when the unit operating parameter adjustment suggestions in the unit operation and maintenance optimization suggestions are converted into control commands, the process can follow the sequence of "suggestion generation—executability judgment—boundary verification—command issuance". Specifically, firstly, candidate adjustment suggestions are generated based on the state prediction results and fault diagnosis results. Then, it is determined whether the candidate adjustment suggestions belong to the type of operating parameter adjustment that the control system can directly execute. For directly executable adjustment suggestions, it is further determined whether they meet the unit's preset safety boundary, equipment protection boundary, and current operating condition constraints. When the constraints are met, the corresponding adjustment suggestion is converted into a control command that the unit's control system can recognize and sent for execution. When the constraints are not met, it is retained as a manual operation and maintenance suggestion or maintenance plan and is not directly issued to the control system. The control command may include one or more of the following: load setpoint adjustment command, intake air temperature adjustment command, valve opening adjustment command, operating condition switching command, and shutdown protection command. After the control command is executed, the operating data after execution is re-collected and fed back to the digital twin model to calculate the changes in the corresponding key parameters before and after execution. If the operating data after execution shows that the fault risk has decreased and the parameter deviation has fallen back, the current control strategy is maintained. If the operating data after execution shows that there is still an abnormal deviation, a new operation and maintenance optimization suggestion is generated and the next round of closed-loop control process is entered.

[0044] Specifically, the digital twin demonstration steps can display the temperature distribution, pressure status, vibration status, and fault highlighted locations of gas turbines, waste heat boilers, steam turbines, and auxiliary equipment through a graphical interface. At the same time, it can present status prediction curves, operational stability assessment results, and fault diagnosis results to improve the intuitive perception of the unit's operating status by operation and maintenance personnel.

[0045] Furthermore, the life prediction sub-model can estimate the remaining life of unit components and the entire unit under current operating conditions based on unit operating history, real-time operating data, fault diagnosis results, and simulation data. The remaining life prediction results serve as the basis for maintenance plan formulation, overhaul cycle optimization, and component replacement decisions. The inputs to the life prediction sub-model include: unit operating data, fault diagnosis results, and simulation data. Specifically, the unit operating data may include operating parameters such as temperature, pressure, flow rate, speed, and vibration; the fault diagnosis results may include fault type, fault location, fault evolution results, and warning level; the simulation data may include simulation results of temperature, pressure, speed, and vibration under different operating conditions output by the digital twin model. The outputs of the life prediction sub-model include: predicted remaining life values ​​of unit components, predicted remaining life values ​​of the entire unit, and the basis for maintenance plan formulation generated based on the predicted remaining life values. Further, the basis for maintenance plan formulation may include component replacement time, overhaul cycle, and priority maintenance targets; the life prediction sub-model can estimate the remaining life of unit components and the entire unit based on component degradation degree indicators. Specifically, based on the unit's operating history, real-time operating data, fault diagnosis results, and simulation data, characteristic quantities related to component lifespan are extracted. These characteristic quantities may include at least one or more of the following: temperature load, pressure load, vibration intensity, load fluctuation frequency, start-stop frequency, cumulative operating time, and fault occurrence frequency. Then, a component degradation degree index is constructed based on these characteristic quantities, and the current component degradation degree index is compared with a preset lifespan termination threshold to obtain the predicted remaining lifespan value for the corresponding component. The lifespan prediction sub-model can predict the lifespan of gas turbine blades, waste heat boiler heat exchange components, steam turbine rotors, and the entire system. When the predicted remaining lifespan value of a component is lower than a preset lifespan warning threshold, the component is marked as a priority for maintenance and included in the maintenance plan. By combining fault diagnosis results with lifespan prediction results, maintenance plans can be formulated not only based on the current fault state but also based on future degradation trends to schedule maintenance cycles and component replacement times in advance, thereby improving the foresight of the operation and maintenance plan.

[0046] In one specific implementation, a method for combined cycle unit status prediction and fault diagnosis is provided, applied to a 9F-class gas-steam combined cycle unit. This method includes steps such as digital twin model construction, multi-source real-time data acquisition and preprocessing, real-time mapping of the digital twin model to the unit, unit status prediction, proactive fault diagnosis, and closed-loop control and operation optimization. The method forms a closed-loop management process of "acquisition—modeling—prediction—diagnosis—control—feedback" to achieve dynamic monitoring of the combined cycle unit's operating status, anomaly identification, fault prediction, and operation optimization.

[0047] In this embodiment, the steps include: constructing a digital twin model of the gas-steam combined cycle unit; collecting and preprocessing multi-source operating data; establishing a real-time mapping between the digital twin model and the unit; performing unit status prediction; and conducting advanced diagnosis of potential unit faults and closed-loop control and operation and maintenance optimization.

[0048] Specifically, the digital twin model includes a three-dimensional physical model, a mathematical model, and a simulation model. The three-dimensional physical model is created using SolidWorks 3D modeling software. Based on the design drawings and equipment parameters of the 9F-class gas-steam combined cycle unit, a 1:1 model of the gas turbine, waste heat boiler, steam turbine, and auxiliary equipment is created, and the dimensions, materials, connection methods, and operating parameters of each piece of equipment are labeled, thus forming a three-dimensional physical model corresponding to the unit entity. The mathematical model includes a gas turbine mathematical model, a waste heat boiler mathematical model, and a steam turbine mathematical model. The gas turbine mathematical model includes combustion equations for the combustion chamber, compression equations for the compressor, and turbine expansion equations to describe the energy conversion process of the gas turbine. A rotor vibration dynamics equation is also established to characterize the rotor's vibration characteristics, with a vibration frequency range of 0-1000Hz. The waste heat boiler mathematical model includes heat and mass transfer equations and working fluid flow equations to describe the heat transfer process between flue gas and the working fluid, as well as the evaporation and superheating processes of water in the boiler. The steam turbine mathematical model includes turbine expansion equations, thermodynamic process equations, and dynamic equations to describe the steam turbine's expansion and work process, speed variation, and vibration characteristics, with a speed variation range of 3000-3600 r / min. These mathematical models are then coupled to form the overall unit mathematical model. The simulation model is constructed by coupling the three-dimensional physical model with the mathematical model. In this embodiment, ANSYS finite element analysis software is used to build a real-time simulation platform, and the unit's operating boundary conditions are set as follows: ambient temperature 25℃, atmospheric pressure 101.3 kPa, and rated load 300MW. The simulation model can simulate operating scenarios such as unit start-up and shutdown, load adjustment, and operating condition switching, and output simulation data such as temperature, pressure, speed, and vibration in real time. Preferably, the simulation accuracy error is no higher than 3%. Before collecting multi-source data, a multi-source data acquisition system is built, which includes an industrial-grade edge gateway, a wireless data acquisition terminal, and an interface device. The edge gateway connects to sensors on the gas turbine, waste heat boiler, steam turbine, and auxiliary equipment to collect sensor data such as temperature, pressure, flow rate, and vibration in real time. The temperature acquisition range is 0-1500℃, the pressure acquisition range is 0-20MPa, the flow rate acquisition range is 0-100t / h, and the vibration acquisition range is 0-5mm / s. The acquisition frequency is preferably 10Hz, and the edge gateway supports 16GB of local cache. The wireless data acquisition terminal is used to collect environmental parameter data such as ambient temperature, humidity, and atmospheric pressure. The interface unit connects to the existing DCS control system of the unit via RS485 or Ethernet interface to collect operation log data such as start-up and shutdown records, load adjustment records, and operating condition switching records. The collected raw data is preprocessed to generate a standardized real-time dataset. The preprocessing process includes: using the 3σ criterion to remove outliers, using linear interpolation to fill in missing data, and using the min-max normalization algorithm to transform data of different dimensions into standardized data within the range of [0,1].Subsequently, the preprocessed sensor data, operation log data, and environmental parameter data are sorted and integrated by timestamp to form a standardized real-time dataset for use in subsequent digital twin model updates and status analysis. The obtained standardized real-time dataset is input into the digital twin model in real time via a 5G high-speed data transmission channel, with a data transmission latency of no more than 50ms to ensure data real-time performance. During the model update phase, based on the standardized real-time dataset, an incremental learning algorithm is used to dynamically adjust the mathematical model parameters in the digital twin model. The updated parameters include thermodynamic equation coefficients and kinetic equation boundary conditions. Simultaneously, the equipment temperature distribution, vibration status, and other operational display information in the three-dimensional physical model are updated to ensure that the digital twin model operates synchronously with the combined cycle unit. Preferably, the synchronization error between the digital twin model and the unit is no more than 2%. During the mapping verification phase, the simulation data output by the digital twin model is compared with the real-time operating data of the unit every 10 minutes to calculate the deviation value of each operating parameter. When the deviation exceeds 5%, a model calibration process is triggered to adjust the model parameters in the digital twin model to restore the synchronous mapping relationship between the digital twin model and the actual operating state of the unit. State prediction employs a combination of a Long Short-Term Memory (LSTM) network and digital twin simulation. The LSTM model is trained based on the unit's historical operating data from the past year to uncover the temporal correlation between historical and real-time operating data. The digital twin simulation model simulates the unit's state changes under future operating conditions. During the prediction execution phase, a standardized real-time dataset is input into the state prediction model, with the prediction time interval set to the next 24 hours. Predicted parameters include unit load, power generation efficiency, gas turbine blade temperature, and steam turbine rotor vibration parameters. The LSTM model outputs the prediction results for these parameters over the next 24 hours, while the digital twin simulation model simultaneously simulates the operating state for the next 24 hours and outputs the simulation prediction results. The two types of prediction results are then fused to obtain the final state prediction result. Furthermore, the state prediction result can be output as trend curves of key unit parameters and an operational stability assessment report. Preferably, the state prediction accuracy is not less than 95%. In the deviation identification stage, the parameter deviation values ​​between the simulation data and the real-time operating data are calculated in real time, and different deviation thresholds are set for different operating parameters. Preferably, the temperature parameter deviation threshold is set to ±3℃, and the vibration parameter deviation threshold is set to ±0.5mm / s. When the deviation value of any operating parameter exceeds the corresponding deviation threshold, the fault diagnosis process is triggered. In the fault feature matching stage, a fault feature library is constructed. The fault feature library contains characteristic parameters, fault characterization information, and fault evolution results for 15 common faults, such as gas turbine blade wear, waste heat boiler scaling, and excessive steam turbine rotor vibration. The fault feature matching unit compares the deviation data with the characteristic parameters in the fault feature library to determine the corresponding fault type and fault location.Furthermore, the fault feature database adopts a dynamic update mechanism, updating fault feature parameters quarterly based on historical fault data, maintenance records, and newly discovered fault cases. During the fault prediction phase, fault evolution simulation is performed using a digital twin model based on fault feature matching results to predict fault development trends, and the warning levels are divided into three levels: general warning, important warning, and emergency warning. Preferably, the fault prediction lead time is 1-6 hours. The final output is a fault diagnosis result, including fault type, fault location, fault evolution result, warning level, and preliminary handling suggestions. In this method, operation and maintenance optimization suggestions are generated based on the state prediction results and fault diagnosis results. These suggestions include fault handling plans, unit operating parameter adjustment suggestions, and maintenance plans. Fault handling plans may include emergency shutdown for maintenance or parameter adjustment; unit operating parameter adjustment suggestions may include adjusting unit load, adjusting gas turbine inlet temperature, etc.; and maintenance plans may include component replacement time and maintenance items. During the closed-loop control phase, the operating parameter adjustment suggestions are converted into control commands and sent to the unit control system for execution via an interface device. For example, in this embodiment, the unit load can be adjusted to 80% of the rated load, and the gas turbine inlet temperature can be adjusted simultaneously to mitigate the adverse effects of related fault risks on unit operation. After the control command is executed, the adjusted unit operation data is fed back to the data acquisition system and re-input into the digital twin model.

[0049] To aid in a better understanding of the invention, a more comprehensive and specific embodiment of the invention is described, in which, as follows: Figures 2 to 6 As shown, the present invention provides a method for combined cycle unit condition prediction and fault diagnosis, comprising the following steps: Step 1: Construct a digital twin model of the gas-steam combined cycle unit (S1) The model is constructed using a three-pronged approach: physical model, mathematical model, and simulation model, as detailed below: Physical model construction (S11): Using SolidWorks 3D modeling software, based on the design drawings and equipment parameters of a 9F-class gas-steam combined cycle unit, the 3D structure of the gas turbine, waste heat boiler, steam turbine and auxiliary equipment is restored at a 1:1 scale. The dimensions, materials, connection methods and key parameters of each equipment are marked to generate a 3D physical model, ensuring that it is completely consistent with the physical equipment.

[0050] Mathematical Model Construction (S12): Gas Turbine Mathematical Model: Establish combustion equations for the combustion chamber, compression equations for the compressor, and expansion equations for the turbine to describe the energy conversion process of the gas turbine; establish rotor vibration dynamic equations to describe the vibration characteristics of the rotor, with a vibration frequency range of 0-1000Hz; Waste Heat Boiler Mathematical Model: Establish heat and mass transfer equations to describe the heat transfer process between flue gas and working fluid, with the heat transfer coefficient dynamically adjusted according to operating conditions; establish working fluid flow equations to describe the evaporation and superheating process of water in the boiler; Steam Turbine Mathematical Model: Establish turbine expansion equations and thermodynamic process equations to describe the expansion and work process of steam, and combine with dynamic equations to describe the speed variation (range 3000-3600 r / min) and vibration characteristics of the steam turbine; Unit Integration: Couple the mathematical models of each device, considering the heat transfer between the gas turbine exhaust and the waste heat boiler intake, and the linkage between the steam turbine and the generator, to form a complete unit mathematical model.

[0051] Simulation model construction (S13): The three-dimensional physical model is coupled with the mathematical model, and a real-time simulation platform is built using ANSYS finite element analysis software. The unit operation boundary conditions are set (ambient temperature 25℃, atmospheric pressure 101.3kPa, rated load 300MW) to realize real-time simulation of scenarios such as unit start-up and shutdown, load adjustment, and operating condition switching. The simulation accuracy error is ≤3%, and the simulation data such as temperature, pressure, speed, and vibration of each device are output in real time (D6).

[0052] Step 2: Multi-source real-time data acquisition and preprocessing (S2) Data Acquisition (S21): Build a multi-source data acquisition system, including an industrial-grade edge gateway, wireless data acquisition terminals, and interface units: The edge gateway connects to the sensors of each core device of the unit, and collects sensor data (D1) in real time, such as temperature (0-1500℃), pressure (0-20MPa), flow rate (0-100t / h), and vibration (0-5mm / s), with an acquisition frequency of 10Hz and support for local caching (16GB cache capacity); The wireless data acquisition terminal collects environmental parameter data (D3) such as ambient temperature, humidity, and atmospheric pressure, and uploads it using 4G wireless transmission; The interface unit connects to the unit's original DCS control system and collects operation log data (D2), including start-up and shutdown records, load adjustment records, etc.

[0053] Data preprocessing (S22): Outliers in sensor data are removed using the 3σ criterion, missing data are filled using linear interpolation, and data of different dimensions are transformed into standardized data in the range of [0,1] using the min-max normalization algorithm to eliminate the influence of dimensions.

[0054] Standardized dataset generation (S23): Integrate the preprocessed D1, D2, and D3 data, sort them by timestamp, and generate a standardized real-time dataset (D5) for subsequent real-time updates of the digital twin model.

[0055] Step 3: Real-time mapping of the digital twin model to the actual unit (S3) Data synchronization (S31): Using a 5G high-speed data transmission channel, the standardized real-time dataset (D5) is transmitted to the digital twin model in real time with a transmission delay of ≤50ms, ensuring the timeliness of the data.

[0056] Model Update (S32): Set up a model update module, based on D5 data, using an incremental learning algorithm to dynamically adjust the parameters of the mathematical model (such as the coefficients of the thermodynamic equation and the boundary conditions of the kinetic equation), and at the same time update the operating status of the three-dimensional physical model (such as the temperature distribution and vibration status of the equipment), so as to realize the synchronous operation of the digital twin model and the unit entity, with a synchronization error ≤2%.

[0057] Mapping verification (S33): Every 10 minutes, compare the simulation data (D6) of the digital twin model with the real-time operating data (D5) of the unit entity, calculate the deviation value, and when the deviation value exceeds 5%, trigger the model calibration process to adjust the model parameters and ensure the accuracy of the mapping.

[0058] Step 4: Real-time status prediction of the unit (S4) Prediction Model Construction (S41): A state prediction model is built by combining LSTM neural network and digital twin simulation. The LSTM model is trained based on the historical operating data of the unit over the past year to mine data correlations; the digital twin simulation model is used to verify the rationality of the prediction results.

[0059] Predictive execution (S42): Input the standardized real-time dataset (D5) into the prediction model, set the prediction period to 24 hours, and predict the unit load, power generation efficiency, gas turbine blade temperature, and steam turbine rotor vibration parameters; the LSTM model predicts the parameter change trend, the digital twin simulation model simulates the operating scenario for the next 24 hours, and outputs the simulation prediction results; the two prediction results are fused to obtain the final state prediction result (D7).

[0060] Prediction Results Output (S43): Outputs trend curves of key unit parameters and operational stability assessment report, clarifying the unit's operational status for the next 24 hours, with a prediction accuracy of ≥95%.

[0061] Step 5: Proactive diagnosis of potential unit faults (S5) Deviation identification (S51): The deviation identification unit calculates the deviation between the simulation data (D6) and the real-time running data (D5) in real time. The temperature parameter deviation threshold is set to ±3℃ and the vibration parameter deviation threshold is set to ±0.5mm / s. When the deviation value exceeds the threshold, the fault diagnosis process is triggered.

[0062] Fault Feature Matching (S52): Construct a fault feature library (S54) containing feature parameters, fault characterization information, and fault evolution results for 15 common faults, such as gas turbine blade wear, waste heat boiler scaling, and excessive steam turbine rotor vibration. The fault feature matching unit compares the deviation data with the feature parameters in the fault feature library to determine the type and location of potential faults.

[0063] Fault prediction (S53): Based on the matching results and combined with the simulation of the digital twin model, predict the development trend of the fault, classify the warning level (general warning, important warning, emergency warning), realize the advanced prediction of the fault, and the prediction time is 1-6 hours in advance; output the fault diagnosis result (D8), including fault type, fault location, fault evolution result, warning level and preliminary handling suggestions.

[0064] The fault feature database (S54) adopts a dynamic update mechanism, updating fault feature parameters quarterly based on historical fault data, maintenance records, and newly discovered fault cases to improve the accuracy of fault diagnosis.

[0065] Step 6: Closed-loop control and operation and maintenance optimization (S6) Operation and maintenance suggestion generation (S61): Combine the status prediction results (D7) and fault diagnosis results (D8) to generate targeted operation and maintenance optimization suggestions (D9), including fault handling plans, unit operating parameter adjustment suggestions and maintenance plans.

[0066] Closed-loop control (S62): The operating parameter adjustment suggestions are converted into control commands (D10) and sent to the unit control system through the interface machine to adjust the unit's operating parameters (such as adjusting the load to 80% of the rated load and adjusting the gas turbine inlet temperature); at the same time, the adjusted unit operating data is fed back to the data acquisition unit (S21) in step 2 and re-input into the digital twin model to realize closed-loop management.

[0067] Operation and maintenance tracking (S63): Track the implementation effect of operation and maintenance optimization suggestions in real time, verify the adjusted unit operation status through digital twin model simulation, and repeat steps 4-6 if there are still deviations or potential faults to continuously optimize the control scheme.

[0068] In summary, this embodiment has at least the following technical effects: By constructing a digital twin model that includes a three-dimensional physical model, a mathematical model, and a simulation model, and by establishing a real-time mapping relationship between the unit entity and the virtual model in conjunction with multi-source real-time data, the digital twin model can continuously track the actual operating status of the combined cycle unit, solving the problem of the disconnect between traditional offline simulation and the actual operating conditions of the unit, thereby providing a high-fidelity model foundation for condition prediction and fault diagnosis. By uniformly collecting, preprocessing, and aligning unit sensor data, operation log data, and environmental parameter data by timestamp, a standardized real-time dataset is formed. This helps to eliminate the differences in units and time between different data sources, improve data quality, and enhance the accuracy of subsequent model updates, status predictions, and fault diagnosis. By combining digital twin simulation with long short-term memory networks for unit status prediction, we can simultaneously leverage the data-driven approach to uncover the patterns of time-series changes and the mechanistic model to characterize the operating patterns of equipment. This improves the accuracy of predictions for unit load, power generation efficiency, and operating parameters such as temperature, pressure, and vibration, and overcomes the limitations of predictions that rely solely on historical data or simulation. By comparing simulation data with real-time operating data to obtain parameter deviations, and combining them with a fault feature library to identify fault types and fault locations, and further combining them with a digital twin model to simulate and analyze the fault evolution trend, it is possible to achieve advanced prediction and graded early warning of potential faults in combined cycle units, solving the problems of delayed fault detection and difficulty in timely handling in existing technologies. The fault feature library in this invention adopts a dynamic update mechanism, which can regularly update the fault feature parameters based on the unit's historical fault data, maintenance records, and newly discovered fault cases. This allows the fault diagnosis model to be continuously optimized as operating experience accumulates, further improving the adaptability and accuracy of fault identification and early warning.

[0069] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for condition prediction and fault diagnosis of a combined cycle power unit, characterized in that, Includes the following steps: S1: Collect multi-source operating data of the gas-steam combined cycle unit, preprocess the multi-source operating data, and generate a standardized real-time dataset aligned with timestamps; S2: Input the standardized real-time dataset into the digital twin model of the gas-steam combined cycle unit. The digital twin model includes a three-dimensional physical model, a mathematical model, and a simulation model. Based on the standardized real-time dataset, dynamically update the parameters of the mathematical model in the digital twin model, and periodically verify the deviation between the simulation data output by the digital twin model and the real-time operating data of the gas-steam combined cycle unit. When the deviation value exceeds a preset threshold, trigger the model calibration process to ensure that the digital twin model and the gas-steam combined cycle unit maintain a synchronous mapping relationship. S3: Based on the digital twin model after the synchronization mapping is completed, the unit status is predicted to obtain the unit status prediction result; and, based on the parameter deviation between the simulation data and the real-time operating data of the gas-steam combined cycle unit, fault diagnosis is performed and the fault diagnosis result is output. S4. Generate unit operation and maintenance optimization suggestions based on the state prediction results and the fault diagnosis results, and convert the unit operating parameter adjustment suggestions in the unit operation and maintenance optimization suggestions into control commands and send them to the control system of the gas-steam combined cycle unit to adjust the unit operating parameters; and feed back the adjusted unit operating data to the digital twin model to realize closed-loop update of unit state prediction, fault diagnosis and operation control.

2. The method for combined cycle unit condition prediction and fault diagnosis according to claim 1, characterized in that, In step S1, the digital twin model is constructed in the following manner: based on the design drawings, equipment parameters, and connection relationships of the gas-steam combined cycle unit, a three-dimensional physical model of the gas turbine, waste heat boiler, steam turbine, and auxiliary equipment is constructed; thermodynamic equations, kinetic equations, and heat and mass transfer equations are established for the gas turbine, waste heat boiler, and steam turbine, respectively, and coupled to form a mathematical model of the entire unit; and the three-dimensional physical model is coupled with the mathematical model of the entire unit, and a simulation model is constructed by combining preset operating boundary conditions.

3. The method for combined cycle unit condition prediction and fault diagnosis according to claim 1, characterized in that, In step S1, the multi-source operating data includes at least sensor data, operating log data, and environmental parameter data; the sensor data includes temperature, pressure, flow rate, speed, and vibration parameters of the gas turbine, waste heat boiler, steam turbine, and auxiliary equipment; the operating log data includes start-up and shutdown records, load adjustment records, and operating condition switching records. The environmental parameter data includes ambient temperature, humidity, and atmospheric pressure; The multi-source operational data is preprocessed, including outlier removal, missing value imputation and normalization, and sorted and integrated according to timestamps.

4. The method for combined cycle unit condition prediction and fault diagnosis according to claim 1, characterized in that, In step S2, the dynamic updating of the model parameters and the deviation verification of the mathematical model include: Input the standardized real-time dataset into the digital twin model; Based on the standardized real-time dataset, the parameters of the mathematical model are dynamically adjusted using incremental learning, and the running status of the three-dimensional physical model is updated synchronously; and The simulation data output by the digital twin model is periodically compared with the real-time operating data of the gas-steam combined cycle unit to calculate the deviation value. When the deviation value is greater than a preset threshold, the model calibration process is triggered to adjust the model parameters in the digital twin model.

5. The method for combined cycle unit condition prediction and fault diagnosis according to claim 1, characterized in that, In step S3, the unit status prediction includes: Historical operating data is input into a pre-trained long short-term memory network model to obtain the predicted values ​​of the target operating parameters of the gas-steam combined cycle unit within the target time interval; Based on the digital twin model after synchronization mapping, the operating state of the gas-steam combined cycle unit is simulated within the target time interval, and the predicted values ​​of simulation parameters within the target time interval are obtained; and The predicted values ​​of the target operating parameters and the predicted values ​​of the simulation parameters are fused to obtain the state prediction results of the gas-steam combined cycle unit within the target time interval.

6. The method for combined cycle unit condition prediction and fault diagnosis according to claim 1, characterized in that, In step S3, the fault diagnosis includes: Obtain the parameter deviation values ​​corresponding to each operating parameter of the gas-steam combined cycle unit; When the deviation value of any of the operating parameters exceeds the corresponding deviation threshold, the parameter deviation value is input into a pre-built fault feature library and matched with the fault feature parameters, fault characterization information, and fault evolution patterns stored in the fault feature library to determine the fault type and fault location; and Based on the determined fault type and fault location, the fault evolution process is simulated using the digital twin model to obtain fault diagnosis results including fault type, fault location, fault evolution result, and warning level.

7. The method for combined cycle unit condition prediction and fault diagnosis according to claim 6, characterized in that, The fault feature database stores characteristic parameters, fault characterization information, and fault evolution patterns, including those related to gas turbine blade wear, waste heat boiler scaling, and abnormal steam turbine rotor vibration. Furthermore, the fault feature database employs a dynamic update mechanism, periodically updating the fault feature parameters based on historical fault data, maintenance records, and newly discovered fault cases of the gas-steam combined cycle unit.

8. The method for combined cycle unit condition prediction and fault diagnosis according to claim 1, characterized in that, In step S4, the generated unit operation and maintenance optimization suggestions include fault handling plans, unit operating parameter adjustment suggestions, and maintenance plans.

9. The method for combined cycle unit condition prediction and fault diagnosis according to claim 1, characterized in that, The prediction and fault diagnosis method also includes a digital twin display step based on the three-dimensional physical model. The digital twin display step is used to display the unit's operating status, status prediction results, and fault diagnosis results in real time, and supports fault location, historical data query, and simulation process playback.

10. The method for predicting the condition and diagnosing faults of a combined cycle unit according to any one of claims 1 to 9, characterized in that: The digital twin model also includes a life prediction sub-model, which is used to predict the remaining life of unit components and the whole unit based on unit operation data, fault diagnosis results and simulation data.

Images