A fault intelligent early warning and diagnosis system for a gas-steam combined cycle unit

Abstract

The present disclosure provides a kind of gas steam combined cycle unit fault intelligent early warning and diagnosis system, the system includes: unit operation system main end, for unified scheduling each end collaborative work;Operation data acquisition end, for collecting unit operation data, unit operation data is synchronized calibration and pre-processing, unit operation data is sent to early warning analysis diagnosis end;Early warning analysis diagnosis end, for identifying the multi-modal non-stationary dynamic characteristics of unit based on unit operation data and mining the associated information between data, output fault early warning signal and fault diagnosis result;Working condition self-adaptive optimization end, for evaluating the model performance and working condition adaptability of early warning analysis diagnosis end, and in the case where model performance and working condition adaptability do not meet preset conditions, early warning analysis diagnosis end is self-adaptively optimized.By the present disclosure, fault timely early warning and accurate positioning and identification of fault can be realized.

Description

Technical Field

[0001] This disclosure relates to the field of gas-fired combined cycle power plant operation technology, and in particular to an intelligent early warning and diagnosis system for gas-fired combined cycle unit faults. Background Technology

[0002] Currently, gas-fired steam combined cycle units have become one of the core equipment in the power production field due to their high overall cycle thermal efficiency, low environmental pollution, low unit investment under the same conditions, good peak-shaving performance, quick start-up and shutdown, small footprint, low water consumption, short construction period and phased commissioning, high degree of automation and few operating personnel.

[0003] The stable operation of gas turbine combined cycle units is directly related to the power generation efficiency and production safety of power plants. Currently, the condition monitoring and fault diagnosis of these units mostly adopt traditional single data monitoring and fixed model analysis methods, which are difficult to adapt to the multimodal non-stationary dynamic characteristics of gas turbine power generation processes. At the same time, traditional methods lack in-depth mining of static and dynamic correlation information between process data, making it difficult to effectively predict the development trend of faults. After a fault occurs, it is impossible to quickly give a clear fault direction, which increases equipment maintenance time and costs and restricts the safe and stable operation of the unit. Summary of the Invention

[0004] This disclosure provides an intelligent early warning and diagnosis system for gas-fired steam combined cycle units to address the problems of untimely and inaccurate status monitoring and fault diagnosis schemes for existing gas-fired steam combined cycle units.

[0005] Based on the above problems, in the first aspect, a gas-fired steam combined cycle unit fault intelligent early warning and diagnosis system is provided, the system including: a unit operation system main terminal, an operation data acquisition terminal, an early warning analysis and diagnosis terminal, and an operating condition adaptive optimization terminal; The main terminal of the unit operation system is used to uniformly schedule the collaborative work of the operation data acquisition terminal, the early warning analysis and diagnosis terminal, and the operating condition adaptive optimization terminal, and to receive fault early warning information and fault diagnosis results sent by the early warning analysis and diagnosis terminal, as well as optimization information from the operating condition adaptive optimization terminal. The operation data acquisition terminal is used to collect unit operation data, perform synchronous calibration and preprocessing on the unit operation data, and send the unit operation data to the early warning analysis and diagnosis terminal. The early warning analysis and diagnosis terminal is used to identify the multimodal non-stationary dynamic characteristics of the unit based on the unit operation data and to mine the correlation information between the data, and to output fault early warning signals and fault diagnosis results. The adaptive optimization terminal is used to evaluate the model performance and working condition adaptability of the early warning analysis and diagnosis terminal, and to perform adaptive optimization of the early warning analysis and diagnosis terminal when the model performance and working condition adaptability do not meet preset conditions.

[0006] In conjunction with the first aspect, in one possible implementation, the operating data acquisition terminal includes: a unit parameter acquisition module, an operating condition monitoring module, a data preprocessing module, a data synchronization calibration module, and a data transmission module; The unit parameter acquisition module is used to collect unit operating parameters; The operating condition monitoring module is used to track the operating status of the unit. The data synchronization and calibration module is used to achieve time and dimension synchronization of multi-parameter acquired data; The data preprocessing module is used to perform noise reduction, completion, and standardization processing on the collected raw data; The data transmission module is used to transmit unit operation data to the early warning analysis and diagnosis terminal. The unit operation data includes unit operation parameters and unit operation status.

[0007] In conjunction with the first aspect, in one possible implementation, the early warning analysis and diagnosis terminal includes a data feedback module, a multimodal state analysis module, an operation process data mining module, a fault early warning core module, and a fault diagnosis module; The data feedback module is used to receive unit operation data sent by the operation data acquisition terminal, the unit operation data including unit operation parameters and unit operation status; The multimodal analysis module is used to perform continuous wavelet transform on the unit operating parameters using a sliding window time-frequency analysis algorithm, extract time-frequency features under different operating conditions, and identify the multimodal non-stationary dynamic characteristics of the unit based on the time-frequency features under different operating conditions. The operation process data mining module is used to extract the correlation information between the unit operation data through Granger causality test and FP-Growth association rule mining algorithm; The fault early warning core module is used to suppress residual contamination of the unit operating status characteristics, predict fault trends based on the residual sequence after contamination suppression, and output a fault early warning signal to the main terminal of the unit operating system based on the fault trend; the unit operating status characteristics include at least one of the following: unit operating data, multimodal non-stationary dynamic characteristics, and correlation information between unit operating data. The fault diagnosis module is used to construct a fault feature matrix by reconstructing the deviation of the unit operating parameters, obtain a fault diagnosis result based on the fault feature matrix, and send the fault diagnosis result to the main terminal of the unit operation system. The fault diagnosis result includes the fault type and the fault location.

[0008] In conjunction with the first aspect, in one possible implementation, the fault early warning core module includes: a residual contamination suppression module; the residual contamination suppression module is used for: Calculate the original residual sequence of the unit operating state characteristics, where the original residual of the unit operating state characteristics is the difference between the model prediction value and the current measurement value of the unit operating state characteristics; Based on the impact factor function Time-weighted summation is applied to the residual sequence, where... The time interval between the time of data acquisition for the unit's operation and the current time. The preset attenuation coefficient; Based on the sensor health status of the collected unit operation data, the reliability coefficient of the unit operation data is determined, and the time-weighted residual sequence is corrected based on the reliability coefficient. Abnormal residuals exceeding a preset threshold are removed from the corrected residual sequence to obtain the residual sequence after contamination suppression.

[0009] In conjunction with the first aspect, in one possible implementation, the fault early warning core module includes: a fault trend prediction module and an early warning signal output module; The fault trend prediction module is used to take the residual sequence after pollution suppression as input, and use a sliding window and LSTM time series prediction model to output the fault development probability sequence for future periods. The early warning signal output module is used to send a fault early warning signal to the main terminal of the unit operation system based on the fault development probability sequence of the future time period.

[0010] In conjunction with the first aspect, in one possible implementation, the fault diagnosis module includes: a fault feature construction module; the fault feature construction module is configured to: The time window length is dynamically set according to the current operating conditions of the unit. Principal component analysis was used to reduce the dimensionality of the high-dimensional unit operating parameters, and a reconstructed model under normal operating conditions was built based on the operating parameters under normal operating conditions. Based on the reconfiguration model, the reconfiguration deviation sequence of the current unit operating parameters is calculated, and the reconfiguration deviation sequence is concatenated into a fault feature matrix.

[0011] In conjunction with the first aspect, in one possible implementation, the fault diagnosis module includes: a deep learning diagnosis module and a fault feedback module; The deep learning diagnostic module is used to analyze the fault feature matrix and obtain fault diagnosis results by using a model that combines a deep stacked autoencoder network and a convolutional neural network. The fault feedback module is used to send the fault diagnosis results to the main terminal of the unit operation system.

[0012] In conjunction with the first aspect, in one possible implementation, the deep learning diagnostic module is used for: Feature extraction and noise reduction are performed on the fault feature matrix using a deep stacked autoencoder network; The extracted features are input into a convolutional neural network, which outputs fault diagnosis results, including fault type and fault location.

[0013] In conjunction with the first aspect, in one possible implementation, the working condition adaptive optimization terminal includes: a working condition model performance evaluation module, a modal adaptation monitoring module, and a working condition model adaptive optimization module; The operating condition model performance evaluation module is used to calculate the current operating performance parameters of the early warning analysis and diagnosis terminal, and the operating performance parameters include at least one of the early warning accuracy rate, false alarm rate, and missed alarm rate. The modal adaptation monitoring module calculates the cosine similarity between the current unit operating condition feature vector and the model training set operating condition feature vector of the early warning analysis and diagnosis terminal, and determines the matching degree between the early warning analysis and diagnosis terminal and the current operating condition based on the cosine similarity. The adaptive optimization module for the operating condition model is used to update the model parameters of the early warning analysis and diagnosis terminal when the current operating performance parameters or the degree of matching with the current operating condition do not meet the preset requirements.

[0014] In conjunction with the first aspect, in one possible implementation, the operating condition model adaptive optimization module is used to fine-tune and update the model parameters of the early warning analysis and diagnosis terminal using the current unit operating data in an incremental learning manner.

[0015] The beneficial effects of the embodiments disclosed herein include: 1. This disclosure achieves full-dimensional data acquisition and preprocessing of multiple parameters and multiple operating conditions of gas-fired steam combined cycle units through the operation data acquisition terminal. Combined with the operation process data mining module, it deeply extracts static and dynamic latent variables and highly cross-correlation information between data, adapts to the multi-modal non-stationary process characteristics of gas turbine power generation, provides comprehensive and accurate data support for fault early warning and diagnosis, and solves the problem of insufficient data mining in traditional methods. 2. This disclosure introduces an influence factor function and a reliability coefficient into the model bias calculation through a residual contamination suppression module, which effectively reduces the adverse effects of abnormal data, increases the weight of important data, solves the "residual contamination" problem from the root, reduces the false alarm rate of fault warning, and realizes accurate early warning of faults through a fault trend prediction module, which buys time for unit operation and maintenance and effectively increases the efficiency of unit operation and maintenance. 3. This disclosure constructs a high-dimensional fault feature matrix through a fault feature construction module, and combines it with a deep learning diagnosis module using deep stacked autoencoders and convolutional neural network technology to achieve accurate fault location and identification. This can quickly provide a clear fault direction, significantly improve fault diagnosis accuracy, and reduce equipment repair time and maintenance costs. 4. This invention includes an adaptive optimization terminal for operating conditions, which enables real-time performance evaluation, operating condition modal adaptation monitoring, and adaptive optimization updates of the early warning and diagnostic model. This effectively solves the problems of model stability and long-term effectiveness, ensuring that the model always adapts to changes in the unit's operating status and effectively guarantees the long-term stable operation of the early warning and diagnostic system. 5. This disclosure realizes the integrated functions of intelligent monitoring of the operating status of gas-fired steam combined cycle units, accurate early warning of faults, rapid location diagnosis and adaptive updating of operating condition models, and intelligent operation throughout the process, which reduces the workload of operation and maintenance personnel and improves the safety, stability and power generation efficiency of unit operation. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of an intelligent early warning and diagnosis system for faults in a gas-fired steam combined cycle unit provided in an embodiment of this disclosure; Figure 2 This is a schematic diagram of the structure of a data acquisition terminal provided in an embodiment of the present disclosure; Figure 3 This is a schematic diagram of the structure of an early warning analysis and diagnosis terminal provided in an embodiment of the present disclosure; Figure 4 This is a schematic diagram of the structure of a fault early warning core module provided in an embodiment of the present disclosure; Figure 5 This is a schematic diagram of the structure of a fault diagnosis module provided in an embodiment of the present disclosure; Figure 6 This is a schematic diagram of the structure of an adaptive optimization terminal for operating conditions provided in an embodiment of this disclosure. Detailed Implementation

[0017] This disclosure provides an intelligent early warning and diagnosis system for faults in a gas-fired steam combined cycle unit. The preferred embodiments of this disclosure are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustrative and explanatory purposes only and are not intended to limit the scope of this disclosure. Furthermore, the embodiments and features described herein can be combined with each other unless otherwise specified.

[0018] This disclosure provides an intelligent early warning and diagnosis system for faults in a gas-fired steam combined cycle unit. Figure 1 This is a schematic diagram of the structure of an intelligent early warning and diagnosis system for faults in a gas-fired steam combined cycle unit provided in an embodiment of this disclosure, as shown below. Figure 1 As shown, the system includes: a main terminal 101 for unit operation system, an operation data acquisition terminal 102, an early warning analysis and diagnosis terminal 103, and an adaptive optimization terminal 104 for operating conditions; The main terminal 101 of the unit operation system is used to coordinate the operation data acquisition terminal 102, the early warning analysis and diagnosis terminal 103 and the operating condition adaptive optimization terminal 104 to work together, and to receive fault early warning information and fault diagnosis results sent by the early warning analysis and diagnosis terminal 103 as well as optimization information from the operating condition adaptive optimization terminal 104. The operation data acquisition terminal 102 is used to collect unit operation data, perform synchronous calibration and preprocessing on the unit operation data, and send the unit operation data to the early warning analysis and diagnosis terminal 103; The early warning analysis and diagnosis terminal 103 is used to identify the multimodal non-stationary dynamic characteristics of the unit based on the unit operation data and to mine the correlation information between the data, and output fault early warning signals and fault diagnosis results. The adaptive optimization terminal 104 is used to evaluate the model performance and working condition adaptability of the early warning analysis and diagnosis terminal 103, and to perform adaptive optimization on the early warning analysis and diagnosis terminal 103 when the model performance and working condition adaptability do not meet the preset conditions.

[0019] In this embodiment of the disclosure, the main terminal 101 of the unit operation system is connected to the operation data acquisition terminal 102, the early warning analysis and diagnosis terminal 103, and the operating condition adaptive optimization terminal 104 via control signals. The early warning analysis and diagnosis terminal 103 is connected to the operation data acquisition terminal 102 and the operating condition adaptive optimization terminal 104 via data signals. The early warning analysis and diagnosis terminal 103 is also connected to the main terminal 101 of the unit operation system via bidirectional data signals.

[0020] The main terminal 101 of the unit operation system serves as the core control and display unit of the system. It is responsible for the unified scheduling of the collaborative work of all terminals, receiving early warning, diagnosis and optimization information in real time, and completing human-machine interaction to ensure that the unit's operating status is visible, manageable and controllable throughout the entire process.

[0021] In yet another embodiment provided in this disclosure, Figure 2This is a schematic diagram of the structure of a data acquisition terminal provided in an embodiment of the present disclosure, as shown below. Figure 2 As shown, the operation data acquisition terminal 102 includes: a unit parameter acquisition module 1021, an operating condition monitoring module 1022, a data preprocessing module 1023, a data synchronization calibration module 1024, and a data transmission module 1025; Unit parameter acquisition module 1021 is used to collect unit operating parameters; The operating condition monitoring module 1022 is used to track the operating status of the unit. The data synchronization calibration module 1023 is used to realize the time and dimension synchronization of multi-parameter acquired data; The data preprocessing module 1024 is used to perform noise reduction, completion, and standardization processing on the collected raw data; The data transmission module 1025 is used to transmit the unit operation data to the early warning analysis and diagnosis terminal 103. The unit operation data includes the unit operation parameters and the unit operation status.

[0022] In this embodiment of the disclosure, the operation data acquisition terminal 102 includes a unit parameter acquisition module 1021 electrically connected to the main terminal 101 of the unit operation system. The unit parameter acquisition module 1021 is connected to a data preprocessing module 1024 via a data signal. The data preprocessing module 1024 is connected to the early warning analysis and diagnosis terminal 103 via a data transmission module 1025. The operation data acquisition terminal 102 also includes a condition monitoring module 1022 and a data synchronization calibration module 1023 electrically connected to the main terminal 101 of the unit operation system.

[0023] The unit parameter acquisition module 1021 and the operating condition monitoring module 1022 of the data acquisition terminal 102 collect data from the unit in all dimensions. After calibration by the data synchronization calibration module 1023 and preprocessing by the data preprocessing module 1024, the data is sent to the early warning analysis and diagnosis terminal 103 through the data transmission module 1025.

[0024] Among them, the unit parameter acquisition module 1021 comprehensively collects key operating parameters of the core equipment of the unit, such as the operating parameters of the gas turbine, steam turbine and generator in the unit. The operating parameters include temperature, pressure, speed, liquid level, water flow and valve opening data.

[0025] The operating status monitoring module 1022 tracks the unit's operating status in real time (operating status includes: cold start, warm start, hot start, load increase, rated load operation, load decrease, and shutdown). The data synchronization calibration module 1023 ensures the consistency of time and dimensions of multi-source data. The data preprocessing module 1024 completes noise reduction, completion, and standardization processing, which greatly improves data quality and usability.

[0026] In yet another embodiment provided in this disclosure, Figure 3 This is a schematic diagram of the structure of an early warning analysis and diagnosis terminal provided in an embodiment of this disclosure, such as... Figure 3 As shown, the early warning analysis and diagnosis terminal 103 includes a data feedback module 1031, a multimodal state analysis module 1032, an operation process data mining module 1033, a fault early warning core module 1034, and a fault diagnosis module 1035. The data feedback module 1031 is used to receive unit operation data sent by the operation data acquisition terminal 102. The unit operation data includes unit operation parameters and unit operation status. The multimodal analysis module 1032 is used to perform continuous wavelet transform on the unit operating parameters using a sliding window time-frequency analysis algorithm, extract time-frequency features under different operating conditions, and identify the multimodal non-stationary dynamic characteristics of the unit based on the time-frequency features under different operating conditions. The process data mining module 1033 is used to extract the correlation information between unit operation data through Granger causality test and FP-Growth association rule mining algorithm; The fault early warning core module 1034 is used to suppress residual contamination of the unit's operating status characteristics, predict fault trends based on the residual sequence after contamination suppression, and output fault early warning signals to the main terminal 101 of the unit operation system based on the fault trends. The unit's operating status characteristics include at least one of the following: unit operating data, multimodal non-stationary dynamic characteristics, and correlation information between unit operating data. The fault diagnosis module 1035 is used to construct a fault feature matrix by reconstructing the deviation of the unit operating parameters, obtain the fault diagnosis result based on the fault feature matrix, and send the fault diagnosis result to the main terminal 101 of the unit operation system. The fault diagnosis result includes the fault type and the fault location.

[0027] In this embodiment, the early warning analysis and diagnosis terminal 103 includes a multimodal state analysis module 1032, a fault early warning core module 1034, and a fault diagnosis module 1035, which are connected to the main terminal 101 of the unit operation system via data signals. The multimodal state analysis module 1032 is connected to the operation process data mining module 1033 via data signals. The operation process data mining module 1033 is connected to the fault early warning core module 1034 via data signals. The fault early warning core module 1034 is connected to the fault diagnosis module 1035 and the operating condition adaptive optimization terminal 104 via data signals. The early warning analysis and diagnosis terminal 103 also includes a data feedback module 1031, which is connected to the data transmission module 1025 via data signals. The data feedback module 1031 is connected to the operation process data mining module 1033 via data signals.

[0028] The multimodal state analysis module 1032 can accurately identify the multimodal non-stationary dynamic characteristics of the unit. The multimodal non-stationary dynamic characteristics of the unit refer to the characteristics of the statistical characteristics such as the mean, variance, and frequency of the operating parameters that change dynamically over time during start-up, shutdown, load change, and fault development, and do not meet the stationarity assumption. The multimodal state analysis module 1032 adopts a sliding window time-frequency analysis algorithm to perform continuous wavelet transform on the unit's operating parameters, extract time-frequency features under different operating conditions, and achieve accurate identification of non-stationary dynamic characteristics.

[0029] The operational process data mining module 1033 deeply extracts static and dynamic correlation information from the data. Static correlation information includes all correlation coefficients and mutual information between different parameters, while dynamic correlation information mainly includes the temporal causal relationship between parameters, the correlation rules of operating condition transitions, and the temporal evolution law of fault parameters. The operational process data mining module 1033 achieves the deep extraction of the above-mentioned correlation information by using Granger causality test and FP-Growth association rule mining algorithm, providing accurate data support for fault early warning and diagnosis, overcoming the shortcomings of insufficient data mining and poor operating condition adaptation of traditional methods. Furthermore, the data feedback module 1035 realizes the stable reception and accurate forwarding of collected data, ensuring smooth data flow without loss, and providing a reliable data entry point for subsequent in-depth data mining and analysis.

[0030] Existing fault diagnosis models are susceptible to "residual contamination" caused by abnormal data, resulting in a high false alarm rate and insufficient diagnostic accuracy. After long-term operation, the models are prone to performance degradation and cannot achieve early and accurate fault warning and rapid fault location.

[0031] In yet another embodiment provided in this disclosure, Figure 4 This is a schematic diagram of the structure of a fault early warning core module provided in an embodiment of this disclosure, as shown below. Figure 4 As shown, the fault early warning core module 1034 includes: a residual contamination suppression module 1034A; the residual contamination suppression module 1034A is used for: The original residual sequence of the unit's operating status characteristics; the original residual of the unit's operating status characteristics is the difference between the model prediction value and the current measurement value of the unit's operating status characteristics. Based on the impact factor function Time-weighted summation is applied to the residual sequence, where... This is the time interval between the moment the unit's operational data was collected and the current moment. The preset attenuation coefficient; preset attenuation coefficient The value range can be 0.02-0.12; Based on the sensor health status of the collected unit operation data, the reliability coefficient of the unit operation data is determined. Based on the reliability coefficient, the time-weighted residual sequence is corrected. The value range of the reliability coefficient can be [0,1]. Abnormal residuals exceeding a preset threshold are removed from the corrected residual sequence to obtain the residual sequence after contamination suppression.

[0032] In this embodiment of the disclosure, the residual contamination suppression module 1034A is connected to the operation process data mining module 1033 via data signals.

[0033] The residual contamination suppression module 1034A introduces an influence factor function and a reliability coefficient to suppress abnormal data interference at its source, solve the "residual contamination" problem, and significantly reduce the false alarm rate of early warnings.

[0034] In yet another embodiment provided in this disclosure, such as Figure 4 As shown, the fault early warning core module 1034 includes: a fault trend prediction module 1034B and an early warning signal output module 1034C; The fault trend prediction module 1034B is used to take the residual sequence after pollution suppression as input, and use a sliding window and LSTM time series prediction model to output the fault development probability sequence for future periods; for example, the fault development probability sequence for the next 20-60 time steps. The early warning signal output module 1034C is used to send a fault early warning signal to the main terminal 101 of the unit operation system based on the fault development probability sequence of future time periods.

[0035] In this embodiment of the disclosure, the residual pollution suppression module 1034A is connected to the fault trend prediction module 1034B via data signals. The fault trend prediction module 1034B is connected to the early warning signal output module 1034C and the fault diagnosis module 1035 via data signals. The early warning signal output module 1034C is connected to the main terminal 101 of the unit operation system via data signals.

[0036] The fault trend prediction module 1034B outputs the fault development probability sequence for future periods, enabling accurate prediction of fault trends and reserving time for maintenance and operation. The early warning signal output module 1034C sends early warning prompts to the main terminal 101 of the unit operation system in real time, enabling early detection of faults.

[0037] In yet another embodiment provided in this disclosure, Figure 5 This is a schematic diagram of the structure of a fault diagnosis module provided in an embodiment of this disclosure, as shown below. Figure 5 As shown, the fault diagnosis module 1035 includes: a fault feature construction module 1035A; the fault feature construction module 1035A is used for: The time window length is dynamically set according to the current operating conditions of the unit; in some embodiments, the time window length ranges from 20 to 60 minutes, and the window is shorter when the unit load is higher. Principal Component Analysis (PCA) is used to reduce the dimensionality of high-dimensional unit operating parameters, and a reconstructed model under normal operating conditions is built based on the operating parameters under normal operating conditions. Based on the reconfiguration model, the reconfiguration deviation sequence of the current unit operating parameters is calculated, and the reconfiguration deviation sequence is concatenated into a fault feature matrix; for example, the fault feature matrix. , where N is the number of samples and M is the feature dimension.

[0038] In this embodiment, the fault feature construction module 1035A constructs a fault feature matrix through high-dimensional reconstruction deviation within a variable time region. This matrix can adapt to the non-stationary operating conditions of equipment dynamic operation, accurately extract weak fault features from the original monitoring data, and structure and organize multi-source, high-dimensional, and discrete operating deviation data. This provides accurate and reliable input for subsequent fault diagnosis, location, and early warning models, effectively improving the accuracy of fault identification and early warning capabilities, and ensuring the reliability of equipment operating status monitoring.

[0039] In yet another embodiment provided in this disclosure, such as Figure 5 As shown, the fault diagnosis module 1035 includes: a deep learning diagnosis module 1035B and a fault feedback module 1035C; The 1035B deep learning diagnostic module is used to analyze the fault feature matrix and obtain fault diagnosis results by using a model that combines a deep stacked autoencoder network and a convolutional neural network. The fault feedback module 1035C is used to send the fault diagnosis results to the main terminal 101 of the unit operation system.

[0040] In this embodiment of the disclosure, the fault feature construction module 1035A is connected to the deep learning diagnosis module 1035B via data signal, the deep learning diagnosis module 1035B is connected to the fault feedback module 1035C via data signal, and the fault feedback module is connected to the main terminal 101 of the unit operation system via data signal.

[0041] The 1035C fault feedback module quickly transmits diagnostic results, effectively shortening repair time and reducing maintenance costs.

[0042] In yet another embodiment provided in this disclosure, the deep learning diagnostic module 1035B is used for: Feature extraction and noise reduction of the fault feature matrix are performed using a deep stacked autoencoder network. The extracted features are input into a convolutional neural network, which outputs fault diagnosis results, including fault type and fault location.

[0043] For example, the deep learning diagnostic module 1035B uses a deep stacked autoencoder network with four hidden layers, containing 256, 128, 64, and 8 neurons respectively, and ReLU activation function to perform preliminary feature extraction and noise reduction on the fault feature matrix. The extracted features are then input into a one-dimensional CNN (Convolutional Neural Network) with a kernel size of 4×4 and a pooling layer stride of 2. Finally, the fault type label and fault location information are output to achieve accurate fault location and identification.

[0044] In yet another embodiment provided in this disclosure, Figure 6 This is a schematic diagram of the structure of an adaptive optimization terminal provided in an embodiment of this disclosure, as shown below. Figure 6 As shown, the adaptive optimization terminal 104 includes: a working condition model performance evaluation module 1041, a modal adaptation monitoring module 1042, and a working condition model adaptive optimization module 1043. The operating condition model performance evaluation module 1041 is used to calculate the current operating performance parameters of the early warning analysis and diagnosis terminal 103. The operating performance parameters include at least one of the early warning accuracy rate, false alarm rate, and false negative rate. For example, when the false alarm rate is higher than 4%, the model performance is determined to have degraded. The modal adaptation monitoring module 1042 calculates the cosine similarity between the current unit operating condition feature vector and the model training set operating condition feature vector of the early warning analysis and diagnosis terminal 103, and determines the matching degree between the early warning analysis and diagnosis terminal 103 and the current operating condition based on the cosine similarity; for example, when the similarity is less than 0.8, it is determined that the model is not compatible with the current operating condition. The adaptive optimization module 1043 is used to update the model parameters of the early warning analysis and diagnosis terminal 103 when the current operating performance parameters or the degree of matching with the current operating conditions do not meet the preset requirements.

[0045] In this embodiment of the disclosure, such as Figure 3 and Figure 6 As shown, the operating condition model performance evaluation module 1041 is connected to the fault early warning core module 1034 via data signals, the operating condition model performance evaluation module 1041 is connected to the modal adaptation monitoring module 1042 via data signals, the modal adaptation monitoring module 1042 is connected to the operating condition model adaptive optimization module 1043 via data signals, and the operating condition model adaptive optimization module 1041 is connected to the multimodal state analysis module 1032 and the operation process data mining module 1033 via data signals.

[0046] The operating condition model performance evaluation module 1041 is used to evaluate the operating performance of the early warning and diagnostic model in real time. The modal adaptation monitoring module 1042 is used to monitor the adaptability of the model to the current operating mode of the unit. The operating condition model adaptive optimization module 1043 is used to adaptively optimize and update the model or algorithm of the multimodal state analysis module 1032 and the operation process data mining module 1033 based on the evaluation and monitoring results.

[0047] In another embodiment provided in this disclosure, the operating condition model adaptive optimization module 1043 is used to fine-tune and update the model parameters of the early warning analysis and diagnosis terminal 103 using the current unit operating data in an incremental learning manner, so as to realize the adaptive optimization and continuous iteration of the model and ensure the effectiveness of long-term operation.

[0048] The workflow of the system provided in this disclosure is as follows: The main terminal 101 of the unit operation system issues a start command to control the operation data acquisition terminal 102, the early warning analysis and diagnosis terminal 103 and the operating condition adaptive optimization terminal 104 to complete the signal connection and initialization, and enter the working state. Subsequently, the data acquisition terminal 102 collects data from the unit through the unit parameter acquisition module 1021 and the operating condition monitoring module 1022. After calibration by the data synchronization calibration module 1023 and preprocessing by the data preprocessing module 1024, the data is sent to the data feedback module 1031 of the early warning analysis and diagnosis terminal 103 through the data transmission module 1025. The data feedback module 1031 forwards the data to the operation process data mining module 1033. The multimodal state analysis module 1032 analyzes the current operation mode of the unit. The operation process data mining module 1033 combines the modal information to extract static, dynamic latent variables and highly correlated information from the data and sends the processed data to the fault early warning core module 1034. Subsequently, the residual contamination suppression module 1034A processes the data to suppress residual contamination caused by abnormal data. The fault trend prediction module 1034B performs fault trend analysis based on the processed data. If a fault risk is predicted, the early warning signal output module 1034C immediately sends an early warning signal to the main terminal 101 of the unit operation system. The fault trend prediction module 1034B sends fault-related data to the fault diagnosis module 1035, the fault feature construction module 1035A constructs a fault feature matrix, the deep learning diagnosis module 1035B locates and identifies faults using deep learning technology, and the fault feedback module 1035C feeds back the diagnosis results to the main terminal 101 of the unit operation system. The operating condition adaptive optimization terminal 104 evaluates the performance of the early warning and diagnostic model in real time through the operating condition model performance evaluation module 1041, and the modal adaptation monitoring module 1042 monitors the adaptability of the model to the unit operating conditions. If the model performance is found to be degraded or the adaptability is insufficient, the operating condition model adaptive optimization module 1043 performs adaptive optimization and updates to the model / algorithm of the multimodal state analysis module 1032 and the operation process data mining module 1033 to ensure that the system is always in the best working state.

[0049] This system enables comprehensive data collection and in-depth analysis of operating data from gas-fired combined cycle units, effectively solving residual contamination and model adaptability issues. It achieves accurate early warning and rapid fault location and diagnosis, enabling fully intelligent operation, improving the stability and reliability of unit operation, reducing maintenance costs, and effectively enhancing the user experience.

[0050] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes in the drawings are not necessarily essential for implementing this disclosure.

[0051] The sequence numbers of the embodiments disclosed above are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0052] Obviously, those skilled in the art can make various modifications and variations to this disclosure without departing from its spirit and scope. Therefore, if such modifications and variations fall within the scope of the claims of this disclosure and their equivalents, this disclosure is also intended to include such modifications and variations.

Claims

1. A fault intelligent early warning and diagnosis system for a gas-fired steam combined cycle unit, characterized in that, The system includes: a main terminal for unit operation system, an operation data acquisition terminal, an early warning analysis and diagnosis terminal, and an adaptive optimization terminal for operating conditions; The main terminal of the unit operation system is used to uniformly schedule the collaborative work of the operation data acquisition terminal, the early warning analysis and diagnosis terminal, and the operating condition adaptive optimization terminal, and to receive fault early warning information and fault diagnosis results sent by the early warning analysis and diagnosis terminal, as well as optimization information from the operating condition adaptive optimization terminal. The operation data acquisition terminal is used to collect unit operation data, perform synchronous calibration and preprocessing on the unit operation data, and send the unit operation data to the early warning analysis and diagnosis terminal. The early warning analysis and diagnosis terminal is used to identify the multimodal non-stationary dynamic characteristics of the unit based on the unit operation data and to mine the correlation information between the data, and to output fault early warning signals and fault diagnosis results. The adaptive optimization terminal is used to evaluate the model performance and working condition adaptability of the early warning analysis and diagnosis terminal, and to perform adaptive optimization of the early warning analysis and diagnosis terminal when the model performance and working condition adaptability do not meet preset conditions.

2. The system according to claim 1, characterized in that, The operational data acquisition terminal includes: a unit parameter acquisition module, an operating condition monitoring module, a data preprocessing module, a data synchronization and calibration module, and a data transmission module; The unit parameter acquisition module is used to collect unit operating parameters; The operating condition monitoring module is used to track the operating status of the unit. The data synchronization and calibration module is used to achieve time and dimension synchronization of multi-parameter acquired data; The data preprocessing module is used to perform noise reduction, completion, and standardization processing on the collected raw data; The data transmission module is used to transmit unit operation data to the early warning analysis and diagnosis terminal. The unit operation data includes unit operation parameters and unit operation status.

3. The system according to claim 1, characterized in that, The early warning analysis and diagnosis terminal includes a data feedback module, a multimodal state analysis module, an operation process data mining module, a fault early warning core module, and a fault diagnosis module. The data feedback module is used to receive unit operation data sent by the operation data acquisition terminal, the unit operation data including unit operation parameters and unit operation status; The multimodal analysis module is used to perform continuous wavelet transform on the unit operating parameters using a sliding window time-frequency analysis algorithm, extract time-frequency features under different operating conditions, and identify the multimodal non-stationary dynamic characteristics of the unit based on the time-frequency features under different operating conditions. The operation process data mining module is used to extract the correlation information between the unit operation data through Granger causality test and FP-Growth association rule mining algorithm; The fault early warning core module is used to suppress residual contamination of the unit operating status characteristics, predict fault trends based on the residual sequence after contamination suppression, and output a fault early warning signal to the main terminal of the unit operating system based on the fault trend; the unit operating status characteristics include at least one of the following: unit operating data, multimodal non-stationary dynamic characteristics, and correlation information between unit operating data. The fault diagnosis module is used to construct a fault feature matrix by reconstructing the deviation of the unit operating parameters, obtain a fault diagnosis result based on the fault feature matrix, and send the fault diagnosis result to the main terminal of the unit operation system. The fault diagnosis result includes the fault type and the fault location.

4. The system according to claim 3, characterized in that, The fault early warning core module includes: a residual contamination suppression module; the residual contamination suppression module is used for: Calculate the original residual sequence of the unit operating state characteristics, where the original residual of the unit operating state characteristics is the difference between the model prediction value and the current measurement value of the unit operating state characteristics; Based on the impact factor function Time-weighted summation is applied to the residual sequence, where... The time interval between the time of data acquisition for the unit's operation and the current time. The preset attenuation coefficient; Based on the sensor health status of the collected unit operation data, the reliability coefficient of the unit operation data is determined, and the time-weighted residual sequence is corrected based on the reliability coefficient. Abnormal residuals exceeding a preset threshold are removed from the corrected residual sequence to obtain the residual sequence after contamination suppression.

5. The system according to claim 3, characterized in that, The core module for fault early warning includes: a fault trend prediction module and a warning signal output module; The fault trend prediction module is used to take the residual sequence after pollution suppression as input, and use a sliding window and LSTM time series prediction model to output the fault development probability sequence for future periods. The early warning signal output module is used to send a fault early warning signal to the main terminal of the unit operation system based on the fault development probability sequence of the future time period.

6. The system according to claim 3, characterized in that, The fault diagnosis module includes: a fault feature construction module; the fault feature construction module is used for: The time window length is dynamically set according to the current operating conditions of the unit. Principal component analysis was used to reduce the dimensionality of the high-dimensional unit operating parameters, and a reconstructed model under normal operating conditions was built based on the operating parameters under normal operating conditions. Based on the reconfiguration model, the reconfiguration deviation sequence of the current unit operating parameters is calculated, and the reconfiguration deviation sequence is concatenated into a fault feature matrix.

7. The system according to claim 3, characterized in that, The fault diagnosis module includes: a deep learning diagnosis module and a fault feedback module; The deep learning diagnostic module is used to analyze the fault feature matrix and obtain fault diagnosis results by using a model that combines a deep stacked autoencoder network and a convolutional neural network. The fault feedback module is used to send the fault diagnosis results to the main terminal of the unit operation system.

8. The system according to claim 7, characterized in that, The deep learning diagnostic module is used for: Feature extraction and noise reduction are performed on the fault feature matrix using a deep stacked autoencoder network; The extracted features are input into a convolutional neural network, which outputs fault diagnosis results, including fault type and fault location.

9. The system according to claim 1, characterized in that, The adaptive optimization terminal for operating conditions includes: an operating condition model performance evaluation module, a modal adaptation monitoring module, and an adaptive optimization module for operating conditions. The operating condition model performance evaluation module is used to calculate the current operating performance parameters of the early warning analysis and diagnosis terminal, and the operating performance parameters include at least one of the early warning accuracy rate, false alarm rate, and missed alarm rate. The modal adaptation monitoring module calculates the cosine similarity between the current unit operating condition feature vector and the model training set operating condition feature vector of the early warning analysis and diagnosis terminal, and determines the matching degree between the early warning analysis and diagnosis terminal and the current operating condition based on the cosine similarity. The adaptive optimization module for the operating condition model is used to update the model parameters of the early warning analysis and diagnosis terminal when the current operating performance parameters or the degree of matching with the current operating condition do not meet the preset requirements.

10. The system according to claim 9, characterized in that, The adaptive optimization module of the operating condition model is used to fine-tune and update the model parameters of the early warning analysis and diagnosis terminal using the current unit operating data in an incremental learning manner.

Images