High-efficiency thermal power plant DCS fault prediction and diagnosis system
By using singular value decomposition and support vector machine to analyze electromagnetic signals in thermal power plants, combining dynamic time bending and Bayesian network to analyze pressure pulse data, early detection and accurate identification of thermal power plants faults is achieved, and the reaction in traditional methods is solved The problems of time delay and insufficient accuracy improve the accuracy of fault prediction and system reliability.
Patent Information
- Application Number
- CN202411071662.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-06
- Publication Date
- 2024-11-29
AI Technical Summary
Traditional thermal power plant fault diagnosis methods have problems such as delayed reaction time, insufficient accuracy and insufficient data utilization, and cannot effectively predict faults or provide maintenance suggestions, resulting in increased risk of equipment damage and maintenance costs.
A high-efficiency thermal power plant DCS fault prediction and diagnosis system is designed, including data acquisition module, signal analysis module and fault output module. The electromagnetic signal is analyzed in depth using Singular Value Decomposition (SVD) and Support Vector Machine (SVM) and combined with Dynamic Time Bending (DTW) and Bayesian networks analyze pressure pulse data to monitor and predict potential failures in real time.
It realizes early detection and accurate identification of electrical and hydraulic system faults in thermal power plants, improves diagnostic accuracy and prediction capabilities, and reduces the risk of equipment damage and maintenance costs.
Smart Images

Figure CN119042194A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of fault prediction of thermal power plants, and in particular to a DCS fault prediction and diagnosis system for high-efficiency thermal power plants. Background Art
[0002] The operating efficiency and reliability of thermal power plants greatly depend on the stability of their key equipment, such as electrical and hydraulic systems. Traditional fault diagnosis methods mainly rely on periodic physical inspections or monitoring systems based on simple thresholds. These methods have the following shortcomings:
[0003] Delayed response time: Traditional systems often detect problems only after they occur, which leads to processing delays, increasing the risk of equipment damage and repair costs.
[0004] Lack of accuracy: Methods that rely on simple thresholds cannot accurately identify complex failure modes, especially in hydraulic and electrical systems, where failures often manifest as subtle signal changes and are easily overlooked.
[0005] Insufficient data utilization: Existing technologies fail to fully utilize the large amount of collected data for in-depth analysis, making it impossible to predict failures or provide maintenance recommendations, resulting in the potential value of the data not being fully tapped. Summary of the invention
[0006] Based on the above objectives, the present invention provides a high-efficiency thermal power plant DCS fault prediction and diagnosis system.
[0007] High-efficiency thermal power plant DCS fault prediction and diagnosis system, including the following modules:
[0008] Data acquisition module: collects operating data of electrical equipment and hydraulic systems in thermal power plants, including electromagnetic signals and pressure pulse data;
[0009] Signal analysis module:
[0010] Electromagnetic signal analysis submodule: performs real-time analysis on electromagnetic signals collected from electrical equipment to detect abnormal changes in signals and predict potential electrical faults;
[0011] Pressure pulse analysis submodule: monitors and analyzes the pressure pulse data of the hydraulic system to identify leaks or blockages in the hydraulic system and predict potential hydraulic failures;
[0012] Fault output module: Provides diagnostic results after detecting electrical faults and hydraulic fault signals and analyzing the fault causes and types.
[0013] Furthermore, the data acquisition module specifically includes:
[0014] Electromagnetic signal acquisition device: It is installed on electrical equipment, including transformers, motors and switchgear. The electromagnetic signal acquisition device uses sensors to continuously monitor and record changes in electromagnetic waveforms, and automatically adjusts the sampling frequency to adapt to changes in the operating status of the equipment to capture abnormal electromagnetic signals;
[0015] Pressure pulse acquisition device: installed in key parts of the hydraulic system, including hydraulic pipelines, valves and pump systems, using pressure sensors to monitor pressure fluctuations in real time. The pressure pulse acquisition device records short-term pressure pulses and long-term pressure trends to analyze leaks and blockages in the hydraulic system.
[0016] Furthermore, the electromagnetic signal analysis submodule specifically includes:
[0017] Acquire electromagnetic signals of electrical equipment collected by a data acquisition module;
[0018] Perform preliminary processing on the collected electromagnetic signals, including denoising, standardization and time window cutting;
[0019] Apply singular value decomposition (SVD): Submit the preprocessed electromagnetic signal matrix to singular value decomposition (SVD). SVD decomposes the signal matrix into the product of several matrices, namely U, Σ, and V. * , where Σ contains all singular values, which represent the energy distribution and characteristics of the data;
[0020] Singular value analysis: Analyze the singular values in Σ to determine the main components of the signal. By setting thresholds or using historical data to establish the normal range of singular values, monitor in real time whether the singular values of the current signal have abnormal changes.
[0021] Anomaly detection and fault prediction: When a singular value is detected that deviates from the normal range, it is considered a signal anomaly, indicating a potential fault. A pre-trained model will be used to predict the fault type based on the type and severity of the anomaly.
[0022] Furthermore, the application of singular value decomposition (SVD) specifically includes:
[0023] Signal matrix construction: convert the electromagnetic signal arrangement collected by the electrical equipment into time series data, convert the time series data into a matrix form X, divide the continuously collected signal into segments, each segment contains n data points, and place the signal segments side by side to form an m×n matrix, where m is the number of segments;
[0024] Perform singular value decomposition: The matrix X is decomposed into the product of three matrices through SVD:
[0025] X=U∑V T, where U and V are orthogonal matrices, ∑ is a diagonal matrix, and the elements on the diagonal are singular values, arranged in descending order, representing the energy distribution in the data;
[0026] The singular value analysis also includes:
[0027] Energy analysis: The size of the singular value indicates the "energy" or "importance" of the corresponding component in the data. By analyzing the size of the singular value, we can understand the main variables and structure of the data;
[0028] Cumulative energy contribution: Calculate the cumulative contribution of the top-ranked singular values to determine what percentage of the information (energy) in the data can be explained by the principal components;
[0029] Determine the number of components to be retained: According to the needs, set a threshold to decide how many components to retain. Set the components with cumulative contributions of more than 90% to be retained;
[0030] Minimizing information loss: When choosing the number of components to retain, balance information retention with computational complexity;
[0031] Use the retained principal components for subsequent analysis: Use the retained principal components to reconstruct the approximate signal. If you choose to retain the first k components, the reconstructed signal X k It is expressed as:
[0032] Among them U k ,∑ k , and V k are the first k columns or rows of the original matrices U, ∑ and V respectively;
[0033] In real-time monitoring or fault prediction, by comparing the singular values of real-time data with the singular values under normal conditions, deviations from the normal range can be identified, indicating potential problems or faults.
[0034] Furthermore, the pre-trained model used to predict the fault type adopts a support vector machine SVM model, which is as follows:
[0035] Construct eigenvectors: extract multiple top-ranked singular values from the matrix ∑ as eigenvectors;
[0036] SVM model training: Map the data to high-dimensional space through the kernel function, and find the best segmentation plane in the high-dimensional space. For electromagnetic signal data, the kernel function uses the radial basis RBF kernel;
[0037] The SVM model classification form is expressed as:
[0038] Among them, w is the normal vector of the splitting hyperplane, b is the bias term, ξ iis a slack variable used to handle nonlinear separable situations, C is a regularization parameter that controls the complexity of the model and the tolerance to errors. For each training sample (x i ,y i ), the goal of SVM is to maximize the margin while minimizing the classification error, and cross-validation is used to select the best C value and kernel function parameters;
[0039] The SVM model is trained using the extracted singular value features and labeled data.
[0040] Furthermore, the pressure pulse analysis submodule includes using dynamic time warping (DTW) to analyze pressure pulse data at different time points, find the minimum bending cost path, identify micro-leaks or blockages occurring at different operating speeds, and combine Bayesian networks for probabilistic inference to estimate the probability of failure based on current and historical data.
[0041] Furthermore, the use of dynamic time warping (DTW) to analyze the pressure pulse data at different time points includes comparing the real-time collected pressure pulse data with historical data (data of normal and known fault conditions), and using the DTW algorithm to calculate the similarity between the real-time data and each historical data. DTW dynamically adjusts the points in the time series to minimize the total distance between the two sequences and finds the best matching path. The calculation method is as follows:
[0042] Among them, A and B are two time series, a i and b j is a point in the sequence;
[0043] Leakage and blockage detection: Analyze the minimum tortuosity cost path obtained by DTW. If it is highly similar to the path of a known failure mode (leakage or blockage) or different from the path of the normal operating mode, it is considered to be a problem.
[0044] Further, the probabilistic inference in combination with the Bayesian network includes constructing a Bayesian network model, establishing a Bayesian network whose nodes include various faults (no fault, leakage, blockage) and factors affecting the faults (pressure fluctuations, operating conditions, etc.), and edges represent probabilistic dependencies between nodes;
[0045] Define the conditional probability table CPT, based on historical data, to describe the relationship between each factor and the fault;
[0046] The real-time data features obtained from DTW analysis (measurement of the minimum tortuous cost path) are input into the Bayesian network, which is used to calculate the posterior probabilities of various faults under the current observed data. Based on the calculated fault probabilities, an early warning is generated to instruct the maintenance team to check or take measures.
[0047] Furthermore, the calculation of the minimum bending cost path includes:
[0048] Construct the cost matrix: Suppose there are two time series A and B, create an m×n matrix for A and B, where m and n are the lengths of the two sequences respectively, and each element d(i,j) of the matrix represents the distance between the i-th point in sequence A and the j-th point in sequence B, calculated using the Euclidean distance;
[0049] Fill the cumulative cost matrix: Use dynamic programming to fill the cumulative cost matrix D. The value of each element D(i,j) is based on the minimum value of its adjacent elements (left, above, and above-left) plus the current cost d(i,j):
[0050] D(i,j)=d(i,j)+min(D(i-1,j),D(i,j-1),D(i-1,j-1));
[0051] Trace the minimum cost path: Starting from the lower right corner of the matrix D(m,n), select the path with the minimum cost and trace it back to the upper left corner D(1,1). This path is the minimum curvature cost path between the two time series.
[0052] Further, the posterior probability calculation is expressed as: Among them, P(F|E) is the posterior probability of fault F after observing evidence E, P(F) is the prior probability of fault F, which is obtained based on historical data statistics, P(E|F) is the probability of observing evidence E assuming that fault F occurs, which is obtained by analyzing the data of similar historical events, and P(E) is the total probability of observing evidence E, which is P(E|F) for all possible fault types.
[0053] The sum of P(F): P(E) = ∑ F P(E|F)·P(F).
[0054] Beneficial effects of the present invention:
[0055] The present invention, by combining dynamic time warping (DTW) and Bayesian networks, can more accurately identify and analyze subtle anomalies in the hydraulic system of a thermal power plant. DTW can effectively compare the time series of pressure pulse data, accurately match waveforms even when operating conditions change, and identify tiny leaks or blockages. The Bayesian network is used to analyze the probability of fault occurrence based on historical and current data, thereby improving the accuracy of diagnosis.
[0056] The present invention uses singular value decomposition (SVD) and support vector machine (SVM) to conduct in-depth analysis of the electromagnetic signals of electrical equipment, so that the system can detect potential problems in the early stages of fault occurrence. Singular value analysis helps determine the main components of the signal, and SVM uses these components to accurately predict the type of fault, so that measures can be taken in advance before the problem becomes serious. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0058] Figure 1 A schematic diagram of system function modules according to an embodiment of the present invention;
[0059] Figure 2 Schematic diagram of an electromagnetic signal analysis submodule according to an embodiment of the present invention. DETAILED DESCRIPTION
[0060] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.
[0061] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0062] like Figure 1-Figure 2 As shown in the figure, the high-efficiency thermal power plant DCS fault prediction and diagnosis system includes the following modules:
[0063] Data acquisition module: collects operating data of electrical equipment and hydraulic systems in thermal power plants, including electromagnetic signals and pressure pulse data;
[0064] Signal analysis module:
[0065] Electromagnetic signal analysis submodule: performs real-time analysis on electromagnetic signals collected from electrical equipment to detect abnormal changes in signals and predict potential electrical faults;
[0066] Pressure pulse analysis submodule: monitors and analyzes the pressure pulse data of the hydraulic system to identify leaks or blockages in the hydraulic system and predict potential hydraulic failures;
[0067] Fault output module: Provides diagnostic results after detecting electrical faults and hydraulic fault signals and analyzing the fault causes and types.
[0068] The data acquisition module specifically includes:
[0069] Electromagnetic signal acquisition device: It is installed on electrical equipment, including transformers, motors and switchgear. It uses sensors to continuously monitor and record changes in electromagnetic waveforms, and automatically adjusts the sampling frequency to adapt to changes in the operating status of the equipment, capturing abnormal electromagnetic signals that may indicate potential electrical faults;
[0070] Pressure pulse acquisition device: installed in key parts of the hydraulic system, including hydraulic pipelines, valves and pump systems, using pressure sensors to monitor pressure fluctuations in real time. The pressure pulse acquisition device records short-term pressure pulses and long-term pressure trends to analyze leaks and blockages in the hydraulic system.
[0071] The electromagnetic signal analysis submodule specifically includes:
[0072] Acquire electromagnetic signals of electrical equipment collected by a data acquisition module;
[0073] Perform preliminary processing on the collected electromagnetic signals, including denoising, standardization and time window cutting;
[0074] Apply singular value decomposition (SVD): Submit the preprocessed electromagnetic signal matrix to singular value decomposition (SVD). SVD decomposes the signal matrix into the product of several matrices, namely U, Σ, and V. * , where Σ contains all singular values, which represent the energy distribution and characteristics of the data;
[0075] Singular value analysis: Analyze the singular values in Σ to determine the main components of the signal. Under normal operating conditions, the signal's singular value distribution should be stable. By setting a threshold or using historical data to establish a normal range of singular values, monitor in real time whether the singular value of the current signal has abnormal changes;
[0076] Anomaly detection and fault prediction: When a singular value is detected that deviates from the normal range, it is considered a signal anomaly, indicating a potential fault. A pre-trained model will be used to predict the fault type based on the type and severity of the anomaly.
[0077] The application of singular value decomposition (SVD) specifically includes:
[0078] Signal matrix construction: convert the electromagnetic signal arrangement collected by the electrical equipment into time series data, convert the time series data into a matrix form X, divide the continuously collected signal into segments, each segment contains n data points, and place the signal segments side by side to form an m×n matrix, where m is the number of segments;
[0079] Perform singular value decomposition: The matrix X is decomposed into the product of three matrices through SVD:
[0080] X=U∑V T , where U and V are orthogonal matrices, ∑ is a diagonal matrix, and the elements on the diagonal are singular values, arranged in descending order, representing the energy distribution in the data;
[0081] The singular value analysis also includes:
[0082] Energy analysis: The size of the singular value indicates the "energy" or "importance" of the corresponding component in the data. A larger singular value corresponds to a more important feature or component in the data. By analyzing the size of the singular value, we can understand the main variables and structure of the data.
[0083] Cumulative energy contribution: Calculate the cumulative contribution of the top-ranked singular values to determine what percentage of the information (energy) in the data can be explained by the main components. For example, if the sum of the first three singular values accounts for 80% of the total sum of singular values, it can be considered that these first three components capture most of the dynamics of the data.
[0084] Determine the number of components to be retained: According to the needs, set a threshold to decide how many components to retain. Set the components with cumulative contributions of more than 90% to be retained;
[0085] Minimize information loss: When choosing the number of components to retain, balance information retention with computational complexity, select enough components to minimize information loss, and exclude those with smaller contributions to simplify the model and reduce noise;
[0086] Use the retained principal components for subsequent analysis: Use the retained principal components to reconstruct the approximate signal. If you choose to retain the first k components, the reconstructed signal X k It is expressed as:
[0087] Among them U k ,∑ k , and V k are the first k columns or rows of the original matrices U, ∑ and V respectively;
[0088] In real-time monitoring or fault prediction, by comparing the singular values of real-time data with the singular values under normal conditions, deviations from the normal range can be identified, indicating potential problems or faults.
[0089] The pre-trained model used to predict the fault type adopts a support vector machine SVM model, which is as follows:
[0090] Constructing eigenvectors: Extracting multiple top-ranked singular values from the matrix ∑ as eigenvectors. Singular values encapsulate the main information and dynamic changes of the signal and are suitable for further learning and classification.
[0091] SVM model training: Map the data to high-dimensional space through the kernel function, and find the best segmentation plane in the high-dimensional space. For electromagnetic signal data, the kernel function uses the radial basis RBF kernel. The RBF kernel performs well in dealing with practical problems due to its flexibility.
[0092] The SVM model classification form is expressed as:
[0093] Among them, w is the normal vector of the splitting hyperplane, b is the bias term, ξ i is a slack variable used to handle nonlinear separable situations, C is a regularization parameter that controls the complexity of the model and the tolerance to errors. For each training sample (x i ,y i ), the goal of SVM is to maximize the margin while minimizing the classification error, and cross-validation is used to select the best C value and kernel function parameters;
[0094] The SVM model is trained using the extracted singular value features and labeled data. This process involves adjusting w and b to find the best splitting hyperplane so that data points of different categories are separated as much as possible.
[0095] Applied to fault prediction and classification, the steps are as follows:
[0096] Real-time feature extraction: Perform the same SVD processing on the electromagnetic signals collected in real time to extract features.
[0097] Use the SVM model for prediction: Input the real-time features into the trained SVM model to predict the fault type.
[0098] Result analysis and response: Based on the prediction results of SVM, the potential fault type is determined. If the model predicts a certain fault, the corresponding maintenance measures can be taken immediately.
[0099] The pressure pulse analysis submodule includes using dynamic time warping (DTW) to analyze pressure pulse data at different time points, find the minimum bending cost path, identify micro-leaks or blockages occurring at different operating speeds, and perform probabilistic inference in combination with Bayesian networks to estimate the probability of failure based on current and historical data.
[0100] The use of dynamic time warping (DTW) to analyze pressure pulse data at different time points includes comparing the real-time collected pressure pulse data with historical data (data of normal and known fault conditions), and using the DTW algorithm to calculate the similarity between the real-time data and each historical data. DTW dynamically adjusts the points in the time series to minimize the total distance between the two series and finds the best matching path. The calculation method is as follows:
[0101] Among them, A and B are two time series, a i and b j is a point in the sequence;
[0102] Leakage and blockage detection: Analyze the minimum tortuosity cost path obtained by DTW. If it is highly similar to the path of a known failure mode (leakage or blockage) or different from the path of the normal operating mode, it is considered to be a problem.
[0103] The probabilistic inference in combination with the Bayesian network includes constructing a Bayesian network model, establishing a Bayesian network, whose nodes include various faults (no fault, leakage, blockage) and factors affecting the faults (pressure fluctuations, operating conditions, etc.), and edges represent probabilistic dependencies between nodes, such as operating conditions may directly affect whether the pressure reading is normal or not;
[0104] Define the conditional probability table CPT, based on historical data, to describe the relationship between each factor and the fault;
[0105] The real-time data features obtained from DTW analysis (measurement of the minimum tortuous cost path) are input into the Bayesian network, which is used to calculate the posterior probabilities of various faults under the current observed data. Based on the calculated fault probabilities, an early warning is generated to instruct the maintenance team to check or take measures.
[0106] The calculation of the minimum bending cost path includes:
[0107] Construct the cost matrix: Suppose there are two time series A and B, create an m×n matrix for A and B, where m and n are the lengths of the two sequences respectively, and each element d(i,j) of the matrix represents the distance between the i-th point in sequence A and the j-th point in sequence B, calculated using the Euclidean distance;
[0108] Fill the cumulative cost matrix: Use dynamic programming to fill the cumulative cost matrix D. The value of each element D(i,j) is based on the minimum value of its adjacent elements (left, above, and above-left) plus the current cost d(i,j):
[0109] D(i,j)=d(i,j)+min(D(i-1,j),D(i,j-1),D(i-1,j-1));
[0110] Trace the minimum cost path: Starting from the lower right corner of the matrix D(m,n), select the path with the minimum cost and trace it back to the upper left corner D(1,1). This path is the minimum curvature cost path between the two time series.
[0111] Analyze the minimum bending cost path:
[0112] Path interpretation: Analyze the shape and pattern of the path. The directness or curvature of the path can reflect the similarity and change pattern between time series. A straighter path indicates that the two series are very similar, while a curved or wavy path indicates anomalies or changes.
[0113] Abnormal pattern recognition: Compare the minimum bending cost path of the real-time data sequence with the known normal and abnormal pattern sequences. If the path of the real-time data is highly similar to the path of a certain abnormal pattern, it indicates that a similar fault has occurred.
[0114] Quantitative analysis: Calculate the average distance or total distance on the path, which is used as an indicator to quantify the difference between two sequences. The abnormal threshold can be set based on historical data for automatic detection and alarm.
[0115] The posterior probability calculation is expressed as: Where P(F|E) is the posterior probability of fault F after observing evidence E, P(F) is the prior probability of fault F, which is obtained based on historical data statistics, P(E|F) is the probability of observing evidence E assuming that fault F occurs, which is obtained by analyzing data from similar historical events, and P(E) is the total probability of observing evidence E, which is the sum of P(E|F)·P(F) for all possible fault types: P(E)=∑ F P(E|F)·P(F).
[0116] Failure probability update and decision:
[0117] Data input: The latest pressure pulse analysis results obtained from DTW are input into the Bayesian network as evidence E.
[0118] Calculate the posterior probability: Use the posterior probability formula to update the posterior probability of each fault type based on the current observed data.
[0119] Decision support: Based on the calculated posterior probability, the most likely fault type is identified and corresponding maintenance and repair suggestions are provided.
[0120] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present invention is limited to these examples. Under the concept of the present invention, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the present invention as described above, which are not provided in detail for the sake of simplicity.
[0121] The present invention is intended to cover all such substitutions, modifications and variations that fall within the broad scope of the claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. High-efficiency thermal power plant DCS fault prediction and diagnosis system, characterized by: Includes the following modules: Data acquisition module: collects operating data of electrical equipment and hydraulic systems in thermal power plants, including electromagnetic signals and pressure pulse data; Signal analysis module: Electromagnetic signal analysis submodule: performs real-time analysis on electromagnetic signals collected from electrical equipment to detect abnormal changes in signals and predict potential electrical faults; Pressure pulse analysis submodule: monitors and analyzes the pressure pulse data of the hydraulic system to identify leaks or blockages in the hydraulic system and predict potential hydraulic failures; Fault output module: Provides diagnostic results after detecting electrical faults and hydraulic fault signals and analyzing the fault causes and types.
2. The high-efficiency thermal power plant DCS fault prediction and diagnosis system according to claim 1 is characterized in that: The data acquisition module specifically includes: Electromagnetic signal acquisition device: It is installed on electrical equipment, including transformers, motors and switchgear. The electromagnetic signal acquisition device uses sensors to continuously monitor and record changes in electromagnetic waveforms, and automatically adjusts the sampling frequency to adapt to changes in the operating status of the equipment to capture abnormal electromagnetic signals; Pressure pulse acquisition device: installed in key parts of the hydraulic system, including hydraulic pipelines, valves and pump systems, using pressure sensors to monitor pressure fluctuations in real time. The pressure pulse acquisition device records short-term pressure pulses and long-term pressure trends to analyze leaks and blockages in the hydraulic system.
3. The high-efficiency thermal power plant DCS fault prediction and diagnosis system according to claim 1 is characterized in that: The electromagnetic signal analysis submodule specifically includes: Acquire electromagnetic signals of electrical equipment collected by a data acquisition module; Perform preliminary processing on the collected electromagnetic signals, including denoising, standardization and time window cutting; Apply singular value decomposition (SVD): Submit the preprocessed electromagnetic signal matrix to singular value decomposition (SVD). SVD decomposes the signal matrix into the product of several matrices, namely U, ∑, and V*, where ∑ contains all singular values, which represent the energy distribution and characteristics of the data. Singular value analysis: Analyze the singular values in ∑ to determine the main components of the signal. By setting thresholds or using historical data to establish a normal range for singular values, monitor in real time whether the singular values of the current signal have abnormal changes. Anomaly detection and fault prediction: When a singular value is detected that deviates from the normal range, it is considered a signal anomaly, indicating a potential fault. A pre-trained model will be used to predict the fault type based on the type and severity of the anomaly.
4. The high-efficiency thermal power plant DCS fault prediction and diagnosis system according to claim 3 is characterized in that: The application of singular value decomposition (SVD) specifically includes: Signal matrix construction: convert the electromagnetic signal arrangement collected by the electrical equipment into time series data, convert the time series data into a matrix form X, divide the continuously collected signal into segments, each segment contains n data points, and place the signal segments side by side to form an m×n matrix, where m is the number of segments; Perform singular value decomposition: The matrix X is decomposed into the product of three matrices through SVD: X=U∑V T , where U and v are orthogonal matrices, ∑ is a diagonal matrix, and the elements on the diagonal are singular values, arranged in descending order, representing the energy distribution in the data; The singular value analysis also includes: Energy analysis: The size of the singular value indicates the "energy" or "importance" of the corresponding component in the data. By analyzing the size of the singular value, we can understand the main variables and structure of the data; Cumulative energy contribution: Calculate the cumulative contribution of the top-ranked singular values to determine what percentage of the information in the data can be explained by the principal components; Determine the number of components to be retained: According to the needs, set a threshold to decide how many components to retain. Components with cumulative contributions of more than 90% are retained; Minimizing information loss: When choosing the number of components to retain, balance information retention with computational complexity; Use the retained principal components for subsequent analysis: Use the retained principal components to reconstruct the approximate signal. If you choose to retain the first k components, the reconstructed signal X k It is expressed as: Among them U k ,∑ k , and V k are the first k columns or rows of the original matrices U, ∑ and v respectively; In real-time monitoring or fault prediction, by comparing the singular values of real-time data with the singular values under normal conditions, deviations from the normal range can be identified, indicating potential problems or faults.
5. The high-efficiency thermal power plant DCS fault prediction and diagnosis system according to claim 4 is characterized in that: The pre-trained model used to predict the fault type adopts a support vector machine SVM model, which is as follows: Construct eigenvectors: extract multiple top-ranked singular values from the matrix ∑ as eigenvectors; SVM model training: Map the data to high-dimensional space through the kernel function, and find the best segmentation plane in the high-dimensional space. For electromagnetic signal data, the kernel function uses the radial basis RBF kernel; The SVM model classification form is expressed as: Among them, w is the normal vector of the splitting hyperplane, b is the bias term, ξ i is a slack variable used to handle nonlinear separable situations, C is a regularization parameter that controls the complexity of the model and the tolerance to errors. For each training sample (x i ,y i ), the goal of SVM is to maximize the margin while minimizing the classification error, and cross-validation is used to select the best C value and kernel function parameters; The SVM model is trained using the extracted singular value features and labeled data.
6. The high-efficiency thermal power plant DCS fault prediction and diagnosis system according to claim 1 is characterized in that: The pressure pulse analysis submodule includes using dynamic time warping (DTW) to analyze pressure pulse data at different time points, find the minimum bending cost path, identify micro-leaks or blockages occurring at different operating speeds, and perform probabilistic inference in combination with Bayesian networks to estimate the probability of failure based on current and historical data.
7. The high-efficiency thermal power plant DCS fault prediction and diagnosis system according to claim 6 is characterized in that: The use of dynamic time warping (DTW) to analyze pressure pulse data at different time points includes comparing the real-time collected pressure pulse data with historical data, and using the DTW algorithm to calculate the similarity between the real-time data and each historical data. DTW dynamically adjusts the points in the time series to minimize the total distance between the two series and finds the best matching path. The calculation method is as follows: Among them, A and B are two time series, a i and b j is a point in the sequence; Leakage and blockage detection: Analyze the minimum tortuosity cost path obtained by DTW. If it is highly similar to the path of the known fault mode or different from the path of the normal operation mode, it is considered that there is a problem.
8. The high-efficiency thermal power plant DCS fault prediction and diagnosis system according to claim 7 is characterized in that: The probabilistic inference in combination with the Bayesian network includes constructing a Bayesian network model, establishing a Bayesian network, whose nodes include various faults and factors affecting the faults, and whose edges represent the probabilistic dependency relationship between the nodes; Define the conditional probability table CPT, based on historical data, to describe the relationship between each factor and the fault; The real-time data features obtained from DTW analysis are input into the Bayesian network, which is used to calculate the posterior probability of various faults under the current observation data. Based on the calculated fault probability, an early warning is generated to instruct the maintenance team to check or take measures.
9. The high-efficiency thermal power plant DCS fault prediction and diagnosis system according to claim 7 is characterized in that: The calculation of the minimum bending cost path includes: Construct a cost matrix: Suppose there are two time series A and B. Create an m×n matrix for A and B, where m and n are the lengths of the two sequences respectively. Each element d(i, j) of the matrix represents the distance between the i-th point in sequence A and the j-th point in sequence B, calculated using the Euclidean distance. Fill the cumulative cost matrix: Use dynamic programming to fill the cumulative cost matrix D. The value of each element D(i, j) is based on the minimum value of its neighboring elements plus the current cost d(i, j): D(i,j)=d(i,j)+min(D(i-1,j), D(i,j-1), D(i-1,j-1)); Trace the minimum cost path: Starting from the lower right corner of the matrix D(m, n), select the path with the minimum cost and trace it back to the upper left corner D(1, 1). This path is the minimum curvature cost path between the two time series.
10. The high-efficiency thermal power plant DCS fault prediction and diagnosis system according to claim 8, characterized in that: The posterior probability calculation is expressed as: Where P(F|E) is the posterior probability of fault F after observing evidence E, P(F) is the prior probability of fault F, which is obtained based on historical data statistics, P(E|F) is the probability of observing evidence E assuming that fault F occurs, which is obtained by analyzing data from similar historical events, and P(E) is the total probability of observing evidence E, which is the sum of P(E|F)·P(F) for all possible fault types: P(E)=∑ F P(E|F)·P(F).
Citation Information
Cited By
Christmas tree fault propagation path development trajectory prediction method based on deep learning
CN119226746A
Coal-fired power plant operation efficiency optimization management system based on electromagnetic sensing
CN119623751A
An operation efficiency optimization management system for coal-fired power plants based on electromagnetic sensing
CN119623751B
Intelligent processing and manufacturing management system
CN119850025A
Distribution network resource utilization diagnostic analysis system and method based on power supply grid
CN120069627A
