Fault early warning system and method for thermal power generating unit

Abstract

The invention relates to the field of fault detection, in particular to a fault early warning system and method for a thermal power generating unit. The system comprises a data acquisition module used for acquiring operation parameters of a thermal power generating unit in real time; the data preprocessing module is used for preprocessing the received operation parameters to obtain preprocessed parameter data; the dynamic multivariate state estimation module is used for dynamically constructing a historical memory matrix based on the preprocessed parameter data and monitoring the running state of the thermal power generating unit by comparing the real-time data with the historical memory matrix; the intelligent early warning module is used for generating a corresponding early warning signal according to a preset fault mode and a threshold value when the dynamic multivariate state estimation module recognizes an abnormal state; and the fault diagnosis module is used for analyzing the early warning signal generated by the intelligent early warning module, determining a fault reason and providing a maintenance suggestion. According to the invention, the accuracy of fault early warning is improved, and abnormity can be found at the initial stage of fault occurrence.

Description

Technical Field

[0001] The present invention relates to the field of fault detection, and in particular to a fault early warning system and method for a thermal power unit. Background Art

[0002] As an important part of the modern power system, the stable operation of thermal power units is of great significance to power supply. However, thermal power units may fail during operation due to various reasons, such as equipment aging, operating errors, environmental changes, etc. If these failures are not discovered and handled in time, they may cause equipment damage, reduced production efficiency, and even safety accidents. Therefore, effective fault warning and diagnosis of thermal power units are crucial to ensure the stable operation of the power system and improve power generation efficiency.

[0003] Existing thermal power unit fault warning systems mostly use traditional monitoring methods, such as regular inspections and experience-based fault diagnosis, which often lack real-time performance and accuracy. With the development of technology, some modern monitoring systems have begun to integrate sensors and data acquisition technologies, which can collect real-time operating parameters of thermal power units, such as temperature, pressure, flow, and vibration. However, these systems usually rely on fixed thresholds to judge faults and cannot adapt to the changes of thermal power units under different operating conditions, resulting in frequent false alarms and missed alarms. In addition, existing technologies often lack continuous monitoring and prediction of the fault development process, making it difficult to achieve early warning of potential faults.

[0004] In the patent literature, there have been some technical solutions that attempt to solve the above problems. For example, patent CN117743742A proposes a method and system for early warning of power generation equipment failures. The method involves obtaining the observed values ​​of parameters of multiple dimensions of the equipment, and obtaining the first residual value corresponding to the parameters of each dimension based on the observed values ​​of the parameters of each dimension and the historical memory matrix to monitor the operating status of the thermal power unit. Another patent CN118470934A proposes a method and system for early warning of thermal power unit failures. The method includes obtaining a mapping relationship between the monitored object and the early warning method, and determining the target early warning method based on the monitored object to determine whether the monitored object meets the early warning conditions.

[0005] Although these technologies provide new monitoring and early warning methods, they still have limitations. For example, the system in CN117743742A relies on the construction of a historical memory matrix and may not be updated in time when parameters change rapidly, while the system in CN118470934A may have insufficient accuracy when dealing with multi-dimensional parameters and complex operating conditions. Therefore, in response to the above problems, a new type of fault early warning system and method for thermal power units is urgently needed. Summary of the invention

[0006] The object of the present invention is to provide a fault warning system and method for a thermal power unit, so as to solve the problem that the prior art lacks a dedicated fire extinguishing bomb mounting bracket and has low manual operation efficiency.

[0007] To achieve the above purpose, the following technical solution is adopted.

[0008] A fault early warning system for a thermal power unit, comprising:

[0009] Data acquisition module, used to collect real-time operating parameters of thermal power units, including but not limited to temperature, pressure, flow and vibration;

[0010] The data preprocessing module is used to filter, denoise and normalize the received operating parameters to obtain preprocessed parameter data;

[0011] Dynamic multivariate state estimation module, which is used to dynamically build a historical memory matrix based on preprocessed parameter data and monitor the operating status of the thermal power unit by comparing real-time data with the historical memory matrix;

[0012] An intelligent early warning module, used to generate a corresponding early warning signal according to a preset fault mode and threshold when the dynamic multivariate state estimation module identifies an abnormal state;

[0013] The fault diagnosis module is used to analyze the warning signal generated by the intelligent warning module, determine the cause of the fault, and provide maintenance suggestions.

[0014] Optionally, a human-computer interaction module is also included, which is connected to the fault diagnosis module and is used to display warning information, fault diagnosis results and maintenance suggestions to the operator, and to receive control instructions from the operator.

[0015] Optionally, the data acquisition module specifically includes:

[0016] A plurality of distributedly deployed sensor units, each of which includes but is not limited to a temperature sensor, a pressure sensor, a flow sensor and a vibration sensor, and the sensor unit is configured with a wireless communication module for sending the collected data to the data preprocessing module in real time;

[0017] A central data collector establishes a communication connection with the wireless communication modules of all the sensor units to collect the data sent by each sensor unit and perform preliminary data synchronization and verification;

[0018] A data calibration unit, used for periodically calibrating the sensor unit;

[0019] Data quality monitoring unit, used to monitor data loss, errors or anomalies that may occur during data collection and trigger alarms for inspection or replacement;

[0020] The intelligent data sampling strategy unit is used to dynamically adjust the data collection frequency and collection parameters according to the operating status and historical fault data of the thermal power unit, and optimize the efficiency and coverage of data collection.

[0021] Optionally, the data preprocessing module includes:

[0022] A signal enhancement unit, used to receive the original signal from the data acquisition module, enhance the strength of the signal, improve the signal-to-noise ratio, and output the enhanced signal;

[0023] A feature extraction unit, connected to the signal enhancement unit, for processing the enhanced signal and extracting key features therefrom, including but not limited to statistical features, frequency domain features and time-frequency domain features, to obtain a key feature set of each part of the thermal power unit;

[0024] A data fusion unit, connected to the feature extraction unit, for synthesizing the key feature sets of various parts of the thermal power unit to form a comprehensive feature set of the equipment status;

[0025] The data transmission module is used to input the key feature sets of each part of the thermal power unit and the comprehensive feature sets of the equipment status as pre-processed parameter data into the dynamic multivariate state estimation module.

[0026] Optionally, the dynamic multivariate state estimation module specifically includes:

[0027] A plurality of separate state estimation units, each of which corresponds to a part of a thermal power unit, including a boiler, a steam turbine and a generator, and each of which is used to dynamically construct a corresponding separate history memory matrix based on a key feature set preprocessed for the corresponding part;

[0028] A comprehensive state estimation unit, used to integrate the historical memory matrix and the comprehensive feature set of each of the individual state estimation units to dynamically construct a comprehensive historical memory matrix;

[0029] A plurality of individual anomaly detection units, each of which is connected to a corresponding individual state estimation unit, and uses a pre-trained machine learning algorithm to compare real-time data with an individual historical memory matrix to identify individual anomaly patterns;

[0030] a comprehensive anomaly detection unit, connected to the comprehensive state estimation unit and all the anomaly detection units, for comparing the real-time data with the comprehensive historical memory matrix to identify the overall anomaly pattern;

[0031] The fault evolution prediction unit is used to receive the overall abnormal pattern and the individual abnormal pattern to perform comprehensive fault evolution and prediction to obtain a prediction result.

[0032] Optionally, the intelligent early warning module includes:

[0033] A threshold setting unit, used to set an abnormal mode detection threshold of each part of the thermal power unit according to historical fault data and operating characteristics of each part of the thermal power unit, and dynamically adjust the abnormal mode detection threshold based on the operating status of each part of the thermal power unit;

[0034] a decision logic unit, connected to the threshold setting unit, the plurality of individual anomaly detection units, the comprehensive anomaly detection unit and the fault evolution prediction unit, for generating an early warning signal based on the prediction result, the set dynamic threshold, the identified individual anomaly mode and the overall anomaly mode, and triggering an early warning of a corresponding level;

[0035] A feedback adjustment unit is connected to the decision logic unit and is used to adjust the parameters of the individual state estimation unit, the individual anomaly detection unit and the comprehensive anomaly detection unit according to the early warning signal and the feedback of the actual fault occurrence, so as to optimize the accuracy and response speed of the early warning system.

[0036] Optionally, the state evaluation unit dynamically constructs a history memory matrix using a dynamic multivariate state estimation technique, including the following steps:

[0037] Acquire multivariate parameter data of each part of the thermal power unit under normal operating conditions to form an initial historical memory matrix of each part of the thermal power unit;

[0038] Normalize the real-time operating parameter data of each part of the thermal power unit to obtain the normalized real-time observation vector of each part of the thermal power unit;

[0039] Using the normalized real-time observation vector, the similarity between the real-time observation vector and the historical observation vector in the historical memory matrix is ​​estimated through the Euclidean distance;

[0040] Several historical observation vectors with the highest similarity to the real-time observation vector are selected to form a dynamic historical memory matrix of each part of the thermal power unit.

[0041] Optionally, the fault evolution prediction unit specifically includes:

[0042] Fault correlation analysis engine, used to analyze the potential connection between the individual abnormal modes and the comprehensive abnormal modes of each part of the thermal power unit, evaluate the impact that the abnormal state of one part may have on other parts, and obtain the fault correlation analysis results;

[0043] The evolution path simulation engine is used to simulate the propagation path of abnormal states. By simulating different failure scenarios, it predicts how a single abnormal mode may cause or aggravate abnormalities in other parts, and how the combined abnormal mode affects the stability of the entire system, and obtains the evolution path simulation results.

[0044] Fault propagation model library, which is used to store various fault propagation models based on historical data and expert knowledge, and is used to represent the propagation mechanism and evolution law between different abnormal modes;

[0045] The prediction result analysis unit is used to conduct comprehensive comparative analysis of fault correlation analysis, evolution path simulation and fault propagation model library to form a comprehensive prediction result.

[0046] Optionally, the decision logic unit specifically includes:

[0047] A data analysis subunit is used to receive and analyze the individual abnormal mode data and the overall abnormal mode data, as well as the fault evolution prediction data provided by the fault evolution prediction unit, to obtain an analysis result;

[0048] A threshold application subunit, used to apply the dynamic threshold provided by the threshold setting unit to the analysis result of the data analysis subunit to determine whether the abnormal mode exceeds the normal operating parameter range;

[0049] A risk assessment subunit, used to assess the risk level of the abnormal mode to the operation of the thermal power unit, including short-term risk and long-term evolution risk, according to the judgment result of the threshold application subunit;

[0050] The early warning signal decision subunit determines whether it is necessary to generate an early warning signal, as well as the urgency and type of the signal based on the assessment result of the risk assessment subunit.

[0051] A fault early warning method for a thermal power unit comprises the following steps:

[0052] Use the data acquisition module to collect the operating parameters of the thermal power unit in real time, including but not limited to temperature, pressure, flow and vibration;

[0053] The received operating parameters are filtered, denoised and normalized by the data preprocessing module to obtain preprocessed parameter data;

[0054] Use the dynamic multivariate state estimation module to dynamically build a historical memory matrix based on preprocessed parameter data, and monitor the operating status of the thermal power unit by comparing real-time data with the historical memory matrix;

[0055] When the dynamic multivariate state estimation module identifies an abnormal state, the intelligent early warning module generates a corresponding early warning signal according to the preset fault mode and threshold;

[0056] The fault diagnosis module analyzes the warning signal generated by the intelligent warning module, determines the cause of the fault, and provides maintenance suggestions;

[0057] The human-computer interaction module displays warning information, fault diagnosis results and maintenance suggestions to the operator, and receives control instructions from the operator.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] Improve the accuracy and real-time performance of fault warning. The present invention collects the operating parameters of the thermal power unit in real time and uses the dynamic multivariate state estimation module to dynamically build a historical memory matrix based on the preprocessed parameter data, thereby achieving real-time monitoring of the operating status of the thermal power unit. This method not only improves the accuracy of fault warning, but also can detect abnormalities at the early stage of the fault, thus buying valuable time for fault handling.

[0060] Reduce false alarms and missed alarms. The intelligent early warning module of the present invention can generate corresponding early warning signals according to preset fault modes and thresholds. These thresholds are dynamically set based on the historical fault data and operating characteristics of various parts of the thermal power unit, thereby reducing false alarms and missed alarms caused by fixed thresholds.

[0061] Improve the efficiency and accuracy of fault diagnosis. The fault diagnosis module of the present invention can analyze the warning signal generated by the intelligent warning module, determine the cause of the fault, and provide maintenance suggestions. This method makes fault diagnosis more scientific and systematic, and improves the efficiency and accuracy of fault handling.

[0062] To enhance the adaptive capability of the system, the feedback adjustment unit of the present invention can adjust the parameters of the individual state estimation unit, the individual anomaly detection unit and the comprehensive anomaly detection unit according to the early warning signal and the feedback of the actual fault occurrence, optimize the accuracy and response speed of the early warning system, and enhance the adaptive capability of the system.

[0063] The individual state estimation units can monitor different parts of the thermal power unit specifically, improving the pertinence and accuracy of fault warning. The comprehensive state estimation unit integrates the status of each part and provides an assessment of the overall operating status of the thermal power unit, which helps to fully understand the health status of the unit. The combination of the individual anomaly detection unit and the comprehensive anomaly detection unit improves the system's ability to identify various abnormal patterns, including local and overall anomalies.

[0064] The present invention integrates multiple technologies such as data acquisition, preprocessing, dynamic multivariate state estimation, intelligent early warning and fault diagnosis, and provides a comprehensive thermal power unit fault early warning solution, which can effectively improve the operating safety and efficiency of thermal power units. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 The present invention is a schematic diagram of a module of a fault warning system embodiment of a thermal power unit.

[0066] Figure 2 The present invention is a schematic diagram of a dynamic multivariate state estimation module of a fault warning system for a thermal power unit according to an embodiment of the present invention.

[0067] Figure 3 The present invention is a schematic flow chart of a method for early warning of a thermal power unit fault according to an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0069] The following detailed description is an exemplary description, which is intended to provide further detailed description of the present invention. Unless otherwise specified, all technical terms used in the present invention have the same meaning as those generally understood by those skilled in the art to which the present application belongs. The terms used in the present invention are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention.

[0070] Example 1

[0071] like Figure 1 As shown, a fault early warning method for a thermal power unit includes:

[0072] Data acquisition module, which consists of multiple sensor units, including temperature, pressure, flow and vibration sensors. These sensor units are distributed and deployed in key parts of thermal power units, such as boilers, steam turbines and generators. Each sensor unit is equipped with a wireless communication module, which can send the collected data to the data preprocessing module in real time.

[0073] The data preprocessing module receives the original signal from the data acquisition module and performs filtering, denoising and normalization. Filtering uses a low-pass filter to remove high-frequency noise, denoising uses an adaptive threshold algorithm to reduce random noise in the signal, and normalization converts all sensor data to the [0, 1] interval for subsequent processing.

[0074] The dynamic multivariate state estimation module dynamically constructs a historical memory matrix based on the preprocessed parameter data. The module first collects the multivariate parameter data of the thermal power unit under normal operation to form an initial historical memory matrix. Then, the real-time operating parameter data is normalized, and the normalized real-time observation vector is used to estimate the similarity between the real-time observation vector and the historical observation vector in the historical memory matrix through the Euclidean distance. Finally, several historical observation vectors with the highest similarity to the real-time observation vector are selected to form a dynamic historical memory matrix.

[0075] Intelligent early warning module, when the dynamic multivariate state estimation module identifies an abnormal state, the intelligent early warning module generates a corresponding early warning signal according to the preset fault mode and threshold. These fault modes and thresholds are set based on historical fault data and operating characteristics, and can be dynamically adjusted according to the real-time operating status of the thermal power unit.

[0076] The fault diagnosis module analyzes the warning signals generated by the intelligent warning module, determines the cause of the fault, and provides maintenance suggestions. This module uses machine learning algorithms, such as decision trees or neural networks, to classify and analyze the causes of warning signals, thereby providing targeted maintenance suggestions.

[0077] Example 2

[0078] The system includes a human-computer interaction module connected to a fault diagnosis module for displaying warning information, fault diagnosis results and maintenance suggestions to operators, and receiving control instructions from operators.

[0079] Composition and functions of human-computer interaction module

[0080] User interface design, the human-computer interaction module includes a high-resolution touch screen for displaying real-time operating data, early warning information, fault diagnosis results and maintenance suggestions of the thermal power unit.

[0081] The interface is designed as an intuitive Graphical User Interface (GUI) for quick operator understanding and response.

[0082] Real-time data display: including real-time trend graphs and values ​​of key parameters such as temperature, pressure, flow, etc.

[0083] Warning information display: Use different colors or icons to identify different levels of warning signals, such as red for emergency and yellow for warning.

[0084] Fault diagnosis results: Displays the results of the fault diagnosis module analysis, including the cause of the fault, the scope of impact, and recommended repair measures.

[0085] Control command input, providing a control panel that allows operators to enter control commands, such as adjusting operating parameters or performing specific maintenance operations.

[0086] The control instructions are transmitted to the corresponding control unit via wireless or wired network to achieve real-time control of the thermal power unit.

[0087] Feedback mechanism, the operator's control instructions and system response results are recorded for subsequent fault analysis and system optimization.

[0088] The system provides feedback confirmation to ensure that every instruction from the operator is received and executed by the system.

[0089] Interactive help system, an interactive help system is integrated to provide operators with guidance and suggestions for troubleshooting.

[0090] The system can recommend the best maintenance strategy based on the current fault status and historical data.

[0091] The human-computer interaction module supports multiple languages ​​to accommodate operators with different language backgrounds.

[0092] The human-computer interaction module includes user authentication and permission management, ensuring that only authorized personnel can access sensitive data and control systems.

[0093] Operation logs record all user activities for easy auditing and monitoring.

[0094] The human-computer interaction module not only provides an intuitive user interface, but also realizes efficient interaction with operators, ensuring the effectiveness and practicality of the thermal power unit fault early warning system.

[0095] Example 3

[0096] The data acquisition module in the thermal power unit fault warning system includes multiple distributed sensor units, a central data collector, a data calibration unit, a data quality monitoring unit, and an intelligent data sampling strategy unit. The deployment and types of sensor units include sensor units: various types of sensor units are deployed in key parts of thermal power units, such as boilers, steam turbines, and generators. Each sensor unit includes at least a temperature sensor, a pressure sensor, a flow sensor, and a vibration sensor to monitor the operating status of the thermal power unit in real time. Each sensor unit is equipped with a wireless communication module for sending the collected data to the data preprocessing module in real time. These modules can transmit data through wireless network protocols (such as Wi-Fi, Bluetooth, or ZigBee).

[0097] The central data collector establishes a communication connection with the wireless communication modules of all sensor units to collect the data sent by each sensor unit. The collector has data processing capabilities and can perform preliminary data synchronization and verification on the received data.

[0098] The data calibration unit is responsible for regularly calibrating the sensor unit to ensure the accuracy and consistency of the data. The calibration process can be completed through automatic calibration algorithms or manual intervention to adapt to environmental changes and drift in sensor performance.

[0099] The data quality monitoring unit monitors data loss, errors or anomalies that may occur during data collection. When an anomaly is detected, the unit triggers an alarm for inspection or replacement to ensure the reliability and integrity of the data.

[0100] The intelligent data sampling strategy unit dynamically adjusts the data collection frequency and collection parameters according to the operating status and historical fault data of the thermal power unit. The unit uses advanced algorithms (such as adaptive filters or machine learning models) to optimize the efficiency and coverage of data collection and improve the response speed and accuracy of the early warning system.

[0101] Example 4

[0102] A data preprocessing module in a fault warning system of a thermal power unit includes a signal enhancement unit, a feature extraction unit, a data fusion unit and a data transmission module. The signal enhancement unit is designed to receive raw signals from a data acquisition module, which may become weak due to sensor noise or environmental interference. The unit uses a specific algorithm, such as an adaptive filter or wavelet enhancement technology, to enhance the strength of the signal and improve the signal-to-noise ratio. The enhanced signal will be output to the next processing stage. The specific formula may involve time domain amplification and filtering processing of the signal, for example: Where S(t) is the original signal, N(t) is the noise estimate, and G(t) is the gain function used to adjust the signal strength.

[0103] The feature extraction unit is responsible for processing the enhanced signal and extracting key features from it. This includes statistical features (such as mean, variance), frequency domain features (such as Fourier transform results) and time-frequency domain features (such as wavelet transform coefficients). The extracted features will form the key feature set of each part of the thermal power unit. For example, the formula for feature extraction using wavelet transform is: wavelet =ψ(σt-τ) where ψ is the wavelet function, t and τ are the time scale and displacement parameters respectively, and σ is the scale parameter.

[0104] The data fusion unit integrates the key feature sets of each part of the thermal power unit to form a comprehensive feature set of the equipment status. This process may adopt a decision-level fusion strategy, such as a statistical rule-based method (sum rule, product rule, etc.), or a model-level fusion strategy, such as a neural network fusion model. The choice of fusion strategy depends on the correlation and complementarity between features.

[0105] The data transmission module is responsible for inputting the key feature sets of each part of the thermal power unit and the comprehensive feature sets of the equipment status as pre-processed parameter data into the dynamic multivariate state estimation module. This module ensures the accurate transmission of data, which may involve the implementation of data encapsulation and network transmission protocols.

[0106] Example 5

[0107] like Figure 2As shown, a dynamic multivariate state estimation module in a fault warning system of a thermal power unit includes multiple individual state estimation units, a comprehensive state estimation unit, multiple individual anomaly detection units, a comprehensive anomaly detection unit, and a fault evolution prediction unit. Each individual state estimation unit corresponds to a part of the thermal power unit (such as a boiler, a steam turbine, and a generator), and dynamically constructs a corresponding individual history memory matrix based on the key feature set after preprocessing of the corresponding part. This process involves the following steps:

[0108] Collect multivariate parameter data of each part under normal operating conditions to form an initial historical memory matrix.

[0109] The real-time operation parameter data is normalized to obtain a normalized real-time observation vector.

[0110] Using the normalized real-time observation vector, the similarity between the real-time observation vector and the historical observation vector in the historical memory matrix is ​​estimated through the Euclidean distance.

[0111] Select several historical observation vectors with the highest similarity to the real-time observation vector to form a dynamic historical memory matrix.

[0112] Formula Example

[0113] D dynarmic ={x 1 ″(1), x 1 ″(2),...,x 1 ″(Z), x 2 ″(1),...,x 2 ″(Z),...,x M ″(1),...,x M ″(Z)}.

[0114] The integrated state estimation unit integrates the history memory matrix and the integrated feature set of each individual state estimation unit to dynamically construct the integrated history memory matrix. This process involves the following steps:

[0115] Collect the history memory matrices of all individual state estimation units.

[0116] Combined with the comprehensive feature set, a comprehensive historical memory matrix is ​​formed, which reflects the status of the entire thermal power unit.

[0117] Multiple individual anomaly detection units

[0118] Implementation method: Each individual anomaly detection unit is connected to a corresponding individual state estimation unit, and uses a pre-trained machine learning algorithm to compare real-time data with an individual historical memory matrix to identify individual anomaly patterns. This process involves the following steps:

[0119] Machine learning algorithms (such as support vector machines (SVMs)) are used to compare the real-time data and the historical memory matrix of each part.

[0120] Data points that are significantly different from the historical memory matrix are identified and marked as abnormal patterns.

[0121] The integrated anomaly detection unit connects the integrated state estimation unit and all individual anomaly detection units to compare the real-time data with the integrated historical memory matrix to identify the overall anomaly pattern. This process involves the following steps:

[0122] The results of all individual anomaly detection units are combined with the history memory matrix of the comprehensive state estimation unit.

[0123] Identify overall abnormal patterns that may involve coordinated abnormalities in multiple parts.

[0124] Fault Evolution Prediction Unit

[0125] Implementation method: The fault evolution prediction unit receives the overall abnormal pattern and the individual abnormal pattern to perform comprehensive fault evolution and prediction to obtain the prediction result. This process involves the following steps:

[0126] Use time series analysis and machine learning models (such as random forests or neural networks) to predict the evolution of abnormal patterns.

[0127] Predict the development trend and potential impact of failures and provide decision support for repair and maintenance.

[0128] Steps for pre-training machine learning algorithms:

[0129] 1. Collect a large amount of historical data of thermal power units under normal operation and abnormal conditions, including parameters such as temperature, pressure, flow, vibration, etc.

[0130] The collected historical data is normalized to eliminate the dimensionality effect and improve the modeling accuracy, including methods such as abnormal data elimination, feature parameter screening, and data dimension reduction.

[0131] 2. Extract key features from the normalized data, including statistical features, frequency domain features, and time-frequency domain features.

[0132] Grey correlation analysis, least squares method, average influence value method, mutual information analysis and other methods were used to screen parameters with a greater correlation with the predicted variables as input parameters.

[0133] 3. Select a suitable machine learning model, such as support vector machine, neural network, decision tree, etc., and train the model using the training set.

[0134] Genetic algorithm or particle swarm algorithm is used to optimize model parameters and improve calculation accuracy.

[0135] 4. Choose a suitable pre-trained model, such as BERT, UniLM, etc. These models have been pre-trained on a large amount of text data and have language generation capabilities.

[0136] Based on the pre-trained model, for the specific task of thermal power unit fault warning, some or all of the model's parameters are further trained and adjusted on a small-scale labeled dataset of the new task, so that the model can better adapt to the new task.

[0137] 5. Use the test set to evaluate the model. Common evaluation indicators include accuracy, recall, F1 value, etc.

[0138] According to the evaluation results, the model is optimized through methods such as cross-validation to improve the accuracy and efficiency of the model.

[0139] Example 6

[0140] An intelligent early warning module in a fault early warning system of a thermal power unit comprises a threshold setting unit, a decision logic unit and a feedback adjustment unit.

[0141] The threshold setting unit sets the dynamic threshold for abnormal mode detection based on the historical fault data and operating characteristics of each part of the thermal power unit. This process involves statistical analysis of historical data to determine the frequency and severity of faults and their relationship with operating parameters. The threshold is set based on the following formula: Threshold = f (historical fault data, operating characteristics) where f is a function that determines the threshold based on historical data and operating characteristics. For example, statistical methods such as standard deviation or percentile can be used to determine the threshold.

[0142] The decision logic unit connects the threshold setting unit, the individual anomaly detection unit, the comprehensive anomaly detection unit, and the fault evolution prediction unit. It generates warning signals based on the prediction results, dynamic thresholds, identified individual anomaly patterns, and overall anomaly patterns, and triggers warnings of corresponding levels. The implementation of the decision logic unit involves the following steps:

[0143] Collect and integrate the output data of each unit.

[0144] Apply logical rules or machine learning models, such as decision trees, random forests, or neural networks, to analyze data.

[0145] Based on the analysis results, a warning signal is generated and the warning level is determined.

[0146] Trigger early warning, including sound and light alarm, SMS notification, etc.

[0147] The feedback adjustment unit adjusts the parameters of the individual state estimation unit, the individual anomaly detection unit, and the integrated anomaly detection unit based on the early warning signal and the feedback of the actual fault occurrence. This process involves the following steps:

[0148] Collect feedback on actual conditions after early warning.

[0149] Analyze early warning accuracy and response effectiveness.

[0150] Based on the feedback results, adjust the threshold settings and parameters of the anomaly detection algorithm to optimize the accuracy and response speed of the early warning system.

[0151] Apply an optimization algorithm, such as gradient descent or genetic algorithm, to tune the parameters.

[0152] Example 7

[0153] A state evaluation unit in a fault warning system of a thermal power unit uses a dynamic multivariate state estimation technique (DMSET) to dynamically construct a historical memory matrix. The following are the specific steps:

[0154] 1. Select M interrelated measurement point parameters of the thermal power unit as monitoring variables. These parameters include but are not limited to key operating parameters such as temperature, pressure, and flow.

[0155] The vector composed of these M variables in a normal operating state of the equipment is recorded as the observation vector X(j). Specifically, X(j)=[x 1 (j), x 2 (j), ..., x i (j), ..., X M (j)]T, where x i (j) is the measured value of measuring point i in normal state j.

[0156] Select N historical normal states of the device, which cover the entire range of normal operation of the device. The observation vectors under these N states form a static historical memory matrix D, that is, D = [x 1 (1), x 1 (2), ..., x 1 (N); x 2 (1), x 2 (2), ..., x 2 (N); ...; x M (1), x M (2), ..., x M (N)].

[0157] 2. Normalize the data of each measuring point of the equipment according to their respective extreme values, convert the measured values ​​to the interval [0, 1], and the normalized static history matrix is ​​D′. The normalization algorithm is x′i(j)=(xi(j)-xi,min) / (xi,max-xi,min), where xi,max and xi,min are the maximum and minimum values ​​of xi(j) respectively.

[0158] 3. The real-time observation vector of the device at a certain moment is Xobs, which is normalized to X′obs. Calculate the Euclidean distance between X'obs and each column vector in the static historical memory matrix D′, and select Z historical observation vectors with the smallest Euclidean distance to X'obs to form the dynamic matrix Dd.

[0159] Let the weight vector W = [w 1 , w 2 , ..., w z ]T, the estimated vector X'est of the current state of the device is a linear combination of the Z historical observation vectors in the dynamic matrix Dd, that is, X'est = W·Dd.

[0160] Calculate the Euclidean distance between the estimated vector X'est and the real-time observation vector X'obs. When the distance exceeds the preset threshold, the system sends a warning signal to indicate a possible abnormal state.

[0161] Example 8

[0162] The fault evolution prediction unit specifically includes:

[0163] Fault correlation analysis engine: used to analyze the potential connection between the individual abnormal modes and the comprehensive abnormal modes of each part of the thermal power unit, evaluate the impact that the abnormal state of one part may have on other parts, and obtain the fault correlation analysis results. The engine collects historical fault data and real-time monitoring data of each part, and uses statistical analysis and machine learning algorithms (such as association rule mining) to identify the correlation between different abnormal modes. In specific implementation, the following formula can be used to calculate the correlation between abnormal modes:

[0164] The evolutionary path simulation engine is used to simulate the propagation path of abnormal conditions. By simulating different fault scenarios, it predicts how a single abnormal mode may cause or aggravate abnormalities in other parts, and how the combined abnormal mode affects the stability of the entire system, and obtains the evolutionary path simulation results. The engine uses the system dynamics model or Monte Carlo simulation method to simulate the fault propagation process based on the data in the fault propagation model library. During the simulation process, the following formula can be used to calculate the probability of fault propagation P (fault propagation) = 1-e -λt Where λ is the fault propagation rate and t is the time.

[0165] The fault propagation model library is used to store various fault propagation models based on historical data and expert knowledge, and is used to represent the propagation mechanism and evolution law between different abnormal modes. The data in the model library can be updated and maintained in the following way: Model update = new data × weight + old model × (1-weight) where the weight is determined based on the reliability of the new data and the historical accuracy of the old model.

[0166] The prediction result analysis unit is used to conduct a comprehensive comparative analysis of fault correlation analysis, evolution path simulation and fault propagation model library to form a comprehensive prediction result. This unit uses machine learning algorithms such as decision trees, random forests or neural networks to comprehensively analyze the outputs of the above three units to form the final fault evolution prediction result.

[0167] Example 9

[0168] A decision logic unit in a fault warning system of a thermal power unit comprises a data analysis subunit, a threshold application subunit, a risk assessment subunit and a warning signal decision subunit.

[0169] The data analysis subunit receives and analyzes individual abnormal pattern data and overall abnormal pattern data, as well as the fault evolution prediction data provided by the fault evolution prediction unit. This unit uses statistical analysis and machine learning algorithms (such as random forests and support vector machines SVM) to analyze the data and obtain analysis results. The analysis results include feature extraction and classification of abnormal patterns, as well as trend prediction of fault evolution.

[0170] The threshold application subunit applies the dynamic threshold provided by the threshold setting unit to the analysis result of the data analysis subunit. The unit determines whether the abnormal mode exceeds the normal operating parameter range based on the dynamic threshold. The setting of the dynamic threshold is based on historical fault data and operating characteristics, as well as changes in real-time operating status. The specific formula may involve statistical methods, such as standard deviation or percentile, to determine the threshold: Threshold = f (historical fault data, operating characteristics) where f is a function that determines the threshold based on historical data and operating characteristics.

[0171] The risk assessment subunit evaluates the risk level of abnormal modes to the operation of thermal power units, including short-term risks and long-term evolution risks, based on the judgment results of the threshold application subunit. This unit uses risk assessment models (such as fault tree analysis FTA) combined with expert knowledge and historical data to quantitatively assess risks. The risk assessment results are used to determine the urgency and type of early warning signals.

[0172] The early warning signal decision subunit decides whether to generate an early warning signal, as well as the urgency and type of the signal, based on the assessment results of the risk assessment subunit. The unit uses algorithms such as decision trees or neural networks to generate early warning signals based on risk assessment results and preset early warning strategies. The generation of early warning signals includes various forms such as sound and light alarms and SMS notifications to ensure that operators can respond in a timely manner.

[0173] Example 10

[0174] like Figure 3 As shown, a fault early warning method for a thermal power unit comprises the following steps:

[0175] The data acquisition module is used to collect the operating parameters of the thermal power unit in real time, including but not limited to temperature, pressure, flow rate and vibration. These parameters are collected through multiple distributed sensor units, each of which includes a temperature sensor, a pressure sensor, a flow sensor and a vibration sensor, and is equipped with a wireless communication module to send the collected data to the data preprocessing module in real time.

[0176] The received operating parameters are filtered, denoised and normalized by the data preprocessing module to obtain preprocessed parameter data. The signal enhancement unit is used to enhance the signal strength, improve the signal-to-noise ratio, and output the enhanced signal. The feature extraction unit processes the enhanced signal and extracts key features from it, including statistical features, frequency domain features and time-frequency domain features, to obtain the key feature set of each part of the thermal power unit.

[0177] The dynamic multivariate state estimation module is used to dynamically construct a historical memory matrix based on the preprocessed parameter data, and the operating status of the thermal power unit is monitored by comparing the real-time data with the historical memory matrix. The module includes multiple separate state estimation units, each of which corresponds to a part of the thermal power unit and dynamically constructs a corresponding separate historical memory matrix.

[0178] When the dynamic multivariate state estimation module identifies an abnormal state, the intelligent early warning module generates a corresponding early warning signal according to the preset fault mode and threshold. The threshold setting unit sets the abnormal mode detection threshold according to the historical fault data and operating characteristics of each part of the thermal power unit, and dynamically adjusts the abnormal mode detection threshold based on the operating state.

[0179] The fault diagnosis module analyzes the warning signal generated by the intelligent warning module, determines the cause of the fault, and provides maintenance suggestions. This module uses the fault evolution prediction unit to receive the overall abnormal pattern and the individual abnormal pattern to perform comprehensive fault evolution and prediction to obtain the prediction result.

[0180] The human-machine interaction module displays warning information, fault diagnosis results and maintenance suggestions to the operator, and receives control instructions from the operator. The module provides an intuitive user interface, displays real-time data, warning information and maintenance suggestions, and provides a manual control interface for operator intervention.

[0181] The specific steps include:

[0182] Multiple individual state estimation units correspond to different parts of the thermal power unit (such as boilers, steam turbines, generators, etc.), and dynamically construct corresponding individual historical memory matrices based on the preprocessed key feature sets of the corresponding parts. Each unit collects multivariate parameter data of the part under normal operating conditions to form an initial historical memory matrix, which is updated in real time to reflect the current state of the equipment.

[0183] The integrated state estimation unit integrates the historical memory matrix and the integrated feature set of each individual state estimation unit to dynamically construct a comprehensive historical memory matrix. This process involves a weighted average or voting mechanism to form a comprehensive estimate of the operating state of the entire thermal power unit.

[0184] Multiple individual anomaly detection units use pre-trained machine learning algorithms (such as support vector machines (SVMs) and random forests) to compare real-time data with individual historical memory matrices to identify individual anomaly patterns. These algorithms are trained on historical data to identify data points that are significantly different from normal operation.

[0185] The integrated anomaly detection unit compares real-time data with an integrated historical memory matrix to identify overall abnormal patterns. This unit uses similar machine learning algorithms but focuses on anomalies in the overall system rather than individual parts.

[0186] The fault evolution prediction unit receives the overall abnormal pattern and the individual abnormal pattern, performs comprehensive fault evolution and prediction, and obtains the prediction result. The unit uses time series analysis and machine learning models (such as recurrent neural network RNN) to predict the development trend of faults.

[0187] The threshold setting unit sets the abnormal mode detection threshold of each part of the thermal power unit according to the historical fault data and operation characteristics of each part of the thermal power unit, and dynamically adjusts the abnormal mode detection threshold based on the operation status of each part of the thermal power unit. The threshold setting can use statistical methods such as standard deviation or percentile.

[0188] The decision logic unit generates a warning signal and triggers a warning of the corresponding level based on the prediction results of the fault evolution prediction unit, the dynamic threshold set by the threshold setting unit, the individual abnormality pattern identified by the individual abnormality detection unit, and the overall abnormality pattern identified by the comprehensive abnormality detection unit. The unit uses logical rules or machine learning models to synthesize this information and decide the type and urgency of the warning signal.

[0189] The feedback adjustment unit adjusts the parameters of the individual state estimation unit, individual anomaly detection unit, and integrated anomaly detection unit based on the early warning signal and the feedback of the actual fault occurrence to optimize the accuracy and response speed of the early warning system. The unit uses an online learning algorithm to adjust the model parameters based on the actual results to improve the performance of the system.

[0190] It is known from common technical knowledge that the present invention can be implemented by other embodiments that do not deviate from its spirit or essential features. Therefore, the above disclosed embodiments are only illustrative in all respects and are not exclusive. All changes within the scope of the present invention or within the scope equivalent to the present invention are included in the present invention.

Claims

1. A fault warning system for a thermal power unit, characterized in that: include, Data acquisition module, used to collect real-time operating parameters of thermal power units, including but not limited to temperature, pressure, flow and vibration; The data preprocessing module is used to filter, denoise and normalize the received operating parameters to obtain preprocessed parameter data; Dynamic multivariate state estimation module, which is used to dynamically build a historical memory matrix based on preprocessed parameter data and monitor the operating status of the thermal power unit by comparing real-time data with the historical memory matrix; An intelligent early warning module, used to generate a corresponding early warning signal according to a preset fault mode and threshold when the dynamic multivariate state estimation module identifies an abnormal state; The fault diagnosis module is used to analyze the warning signal generated by the intelligent warning module, determine the cause of the fault, and provide maintenance suggestions.

2. A fault warning system for a thermal power unit according to claim 1, characterized in that: It also includes a human-computer interaction module, which is connected to the fault diagnosis module and is used to display warning information, fault diagnosis results and maintenance suggestions to the operator, and receive control instructions from the operator.

3. A fault warning system for a thermal power unit according to claim 1, characterized in that: The data acquisition module specifically includes: A plurality of distributedly deployed sensor units, each of which includes but is not limited to a temperature sensor, a pressure sensor, a flow sensor and a vibration sensor, and the sensor unit is configured with a wireless communication module for sending the collected data to the data preprocessing module in real time; A central data collector establishes a communication connection with the wireless communication modules of all the sensor units to collect the data sent by each sensor unit and perform preliminary data synchronization and verification; A data calibration unit, used for periodically calibrating the sensor unit; Data quality monitoring unit, used to monitor data loss, errors or anomalies that may occur during data collection and trigger alarms for inspection or replacement; The intelligent data sampling strategy unit is used to dynamically adjust the data collection frequency and collection parameters according to the operating status and historical fault data of the thermal power unit, and optimize the efficiency and coverage of data collection.

4. A fault warning system for a thermal power unit according to claim 3, characterized in that: The data preprocessing module includes: A signal enhancement unit, used to receive the original signal from the data acquisition module, enhance the strength of the signal, improve the signal-to-noise ratio, and output the enhanced signal; A feature extraction unit, connected to the signal enhancement unit, for processing the enhanced signal and extracting key features therefrom, including but not limited to statistical features, frequency domain features and time-frequency domain features, to obtain a key feature set of each part of the thermal power unit; A data fusion unit, connected to the feature extraction unit, for synthesizing the key feature sets of various parts of the thermal power unit to form a comprehensive feature set of the equipment status; The data transmission module is used to input the key feature sets of each part of the thermal power unit and the comprehensive feature sets of the equipment status as pre-processed parameter data into the dynamic multivariate state estimation module.

5. A fault warning system for a thermal power unit according to claim 4, characterized in that: The dynamic multivariate state estimation module specifically includes: A plurality of separate state estimation units, each of which corresponds to a part of a thermal power unit, including a boiler, a steam turbine and a generator, and each of which is used to dynamically construct a corresponding separate history memory matrix based on a key feature set preprocessed for the corresponding part; A comprehensive state estimation unit, used to integrate the historical memory matrix and the comprehensive feature set of each of the individual state estimation units to dynamically construct a comprehensive historical memory matrix; A plurality of individual anomaly detection units, each of which is connected to a corresponding individual state estimation unit, and uses a pre-trained machine learning algorithm to compare real-time data with an individual historical memory matrix to identify individual anomaly patterns; a comprehensive anomaly detection unit, connected to the comprehensive state estimation unit and all the anomaly detection units, for comparing the real-time data with the comprehensive historical memory matrix to identify the overall anomaly pattern; The fault evolution prediction unit is used to receive the overall abnormal pattern and the individual abnormal pattern to perform comprehensive fault evolution and prediction to obtain a prediction result.

6. A fault warning system for a thermal power unit according to claim 5, characterized in that: The intelligent early warning module includes: A threshold setting unit, used to set an abnormal mode detection threshold of each part of the thermal power unit according to historical fault data and operating characteristics of each part of the thermal power unit, and dynamically adjust the abnormal mode detection threshold based on the operating status of each part of the thermal power unit; a decision logic unit, connected to the threshold setting unit, the plurality of individual anomaly detection units, the comprehensive anomaly detection unit and the fault evolution prediction unit, for generating an early warning signal based on the prediction result, the set dynamic threshold, the identified individual anomaly mode and the overall anomaly mode, and triggering an early warning of a corresponding level; A feedback adjustment unit is connected to the decision logic unit and is used to adjust the parameters of the individual state estimation unit, the individual anomaly detection unit and the comprehensive anomaly detection unit according to the early warning signal and the feedback of the actual fault occurrence, so as to optimize the accuracy and response speed of the early warning system.

7. A fault warning system for a thermal power unit according to claim 5, characterized in that: The state evaluation unit uses dynamic multivariate state estimation technology to dynamically construct a historical memory matrix, including the following steps: Acquire multivariate parameter data of each part of the thermal power unit under normal operating conditions to form an initial historical memory matrix of each part of the thermal power unit; Normalize the real-time operating parameter data of each part of the thermal power unit to obtain the normalized real-time observation vector of each part of the thermal power unit; Using the normalized real-time observation vector, the similarity between the real-time observation vector and the historical observation vector in the historical memory matrix is ​​estimated through the Euclidean distance; Several historical observation vectors with the highest similarity to the real-time observation vector are selected to form the dynamic historical memory matrix of each part of the thermal power unit.

8. A fault warning system for a thermal power unit according to claim 5, characterized in that: The fault evolution prediction unit specifically includes: Fault correlation analysis engine, used to analyze the potential connection between the individual abnormal modes and the comprehensive abnormal modes of each part of the thermal power unit, evaluate the impact that the abnormal state of one part may have on other parts, and obtain the fault correlation analysis results; The evolution path simulation engine is used to simulate the propagation path of abnormal states. By simulating different failure scenarios, it predicts how a single abnormal mode may cause or aggravate abnormalities in other parts, and how the combined abnormal mode affects the stability of the entire system, and obtains the evolution path simulation results. Fault propagation model library, which is used to store various fault propagation models based on historical data and expert knowledge, and is used to represent the propagation mechanism and evolution law between different abnormal modes; The prediction result analysis unit is used to conduct comprehensive comparative analysis of fault correlation analysis, evolution path simulation and fault propagation model library to form a comprehensive prediction result.

9. A fault warning system for a thermal power unit according to claim 6, characterized in that: The decision logic unit specifically includes: A data analysis subunit is used to receive and analyze the individual abnormal mode data and the overall abnormal mode data, as well as the fault evolution prediction data provided by the fault evolution prediction unit, to obtain an analysis result; A threshold application subunit, used to apply the dynamic threshold provided by the threshold setting unit to the analysis result of the data analysis subunit to determine whether the abnormal mode exceeds the normal operating parameter range; A risk assessment subunit, used to assess the risk level of the abnormal mode to the operation of the thermal power unit, including short-term risk and long-term evolution risk, according to the judgment result of the threshold application subunit; The early warning signal decision subunit determines whether it is necessary to generate an early warning signal, as well as the urgency and type of the signal based on the assessment result of the risk assessment subunit.

10. A fault warning method for a thermal power unit, based on a fault warning system for a thermal power unit according to any one of claims 1 to 9, characterized in that: The following steps are included: Use the data acquisition module to collect the operating parameters of the thermal power unit in real time, including but not limited to temperature, pressure, flow and vibration; The received operating parameters are filtered, denoised and normalized by the data preprocessing module to obtain preprocessed parameter data; Use the dynamic multivariate state estimation module to dynamically build a historical memory matrix based on preprocessed parameter data, and monitor the operating status of the thermal power unit by comparing real-time data with the historical memory matrix; When the dynamic multivariate state estimation module identifies an abnormal state, the intelligent early warning module generates a corresponding early warning signal according to the preset fault mode and threshold; The fault diagnosis module analyzes the warning signal generated by the intelligent warning module, determines the cause of the fault, and provides maintenance suggestions; The human-computer interaction module displays warning information, fault diagnosis results and maintenance suggestions to the operator, and receives control instructions from the operator.

Images