An intelligent monitoring method and system for a hydropower station

Abstract

This invention discloses an intelligent monitoring method and system for hydropower stations, belonging to the field of intelligent monitoring technology. The method includes: acquiring historical normal operation data of the hydropower station, extracting historical normal operation time-series features, establishing a normal state baseline model, and constructing an operation knowledge base; acquiring real-time operation data of the hydropower station based on sensors, packaging and substituting it into the normal state baseline model, calculating its deviation from the normal state anomaly score, and associating and matching it with the operation knowledge base to achieve preliminary assessment of the operational health and fault diagnosis; setting dynamic thresholds, continuously monitoring if the health is good, and automatically triggering multi-level alarm commands if the health deteriorates and exceeds the threshold; if multi-level alarm commands are triggered, automatically associating with the target hydropower station's operation knowledge base, and feeding back the final fault cause, handling process, and result data as new samples to the operation knowledge base to achieve adaptive iteration of the hydropower station intelligent monitoring system. This invention improves the operation and maintenance efficiency of hydropower stations.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, specifically to an intelligent monitoring method and system for hydropower stations. Background Technology

[0002] Existing domestic and international technologies for intelligent monitoring of hydropower stations typically rely on single monitoring systems or simple fault diagnosis models. These methods often fail to effectively integrate multi-source heterogeneous data, leading to insufficient data or the inability to identify anomalies in a timely manner under complex and ever-changing hydropower station operating environments. Many existing technologies rely solely on static threshold settings and traditional rule systems, lacking dynamic adaptability. When handling faults, they depend on human experience, which can easily lead to delayed or inaccurate diagnosis. In dealing with multi-dimensional and multi-state environments, the real-time performance and accuracy of existing technologies cannot meet the requirements of efficient operation and maintenance and safety assurance. Summary of the Invention

[0003] To address the aforementioned technical problems, this paper provides an intelligent monitoring method and system for hydropower stations. This technical solution solves the problem that the real-time performance and accuracy of existing technologies cannot meet the requirements of efficient operation and maintenance and safety assurance in multi-dimensional and multi-state environments.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: Firstly, a method for intelligent monitoring of hydropower stations is provided, including: Acquire historical normal operation data of the target hydropower station, extract the historical normal operation time series features of the target hydropower station, establish a normal state baseline model of the target hydropower station, and construct an operation knowledge base of the target hydropower station. Based on multi-source heterogeneous sensors, real-time operation data of the target hydropower station is acquired, packaged and substituted into the baseline model of the target hydropower station's normal state, and its abnormal score deviating from the normal state is calculated. It is then associated and matched with the target hydropower station's operation knowledge base to achieve a preliminary assessment of the target hydropower station's operational health and fault diagnosis. Based on the preliminary assessment of the target hydropower station's operational health and fault diagnosis, a dynamic threshold is set. If the health is good, it is continuously monitored; if the health deteriorates and exceeds the threshold, multi-level alarm commands are automatically triggered. If a multi-level alarm command is triggered, the system will automatically link to the target hydropower station's operation knowledge base. Maintenance personnel will then handle the fault and feed back the final fault cause, handling process, and result data as new samples to the target hydropower station's operation knowledge base, thus enabling the hydropower station's intelligent monitoring system to achieve adaptive iteration.

[0005] Preferably, historical normal operation data of the target hydropower station is obtained, data preprocessing is performed, and timestamps are standardized. The operating data includes: turbine speed, shaft vibration, bearing temperature, cooling water temperature, pressure, flow rate, active / reactive power, circuit breaker status, and valve on / off status; Set a time sliding window with a length of ten minutes, and calculate the mean, variance, and standard deviation of the running data within each window; Calculate the kurtosis and skewness of the running data for each window, obtain their root mean square, get the maximum and minimum values ​​of the running data, construct the time-domain feature matrix of the running data, and extract the time-domain features of the running data. For the running data within each sliding window, the Fast Fourier Transform is used to obtain the frequency domain representation of the running data, the spectral amplitude of the running data is calculated, the maximum frequency and frequency centroid of the spectral amplitude are obtained, the frequency domain feature matrix of the running data is constructed, and the frequency domain features of the running data are extracted. Based on the historical normal operation data of the target hydropower station, the start-up process and temperature rise process are identified. An exponential function model is used for fitting to calculate the rate of change, maximum growth rate and steady state value of the operation process. The starting value, ending value, rise time, slope / rate of change and maximum / minimum value of the operation process are obtained, and the trajectory features of the operation data are extracted. By splicing together the time-domain features, frequency-domain features, and trajectory features of the operational data and performing standardization processing, the historical normal operation time sequence features of the target hydropower station are obtained.

[0006] Preferably, based on the historical normal operation time series characteristics of the target hydropower station, it is divided into a training set and a validation set in chronological order; Based on the historical normal operation time series characteristics of the target hydropower station in the training set, a PCA principal component analysis model is trained, the covariance matrix of the operating data in the training set is calculated, and the corresponding eigenvalues ​​and eigenvectors are obtained. The number of principal components is determined based on the cumulative contribution rate criterion. Based on the historical normal operation data of the target hydropower station in the training set, the data is projected onto the low-dimensional subspace composed of principal components. The T² statistic and squared prediction error of all operation data points are calculated. The upper limit thresholds for the control of the T² statistic and squared prediction error are preset respectively, and a baseline model of the normal state of the target hydropower station is established. The upper limit threshold for the control of the T² statistic and the squared prediction error is taken as the 99th percentile of the historical normal operation data of the target hydropower station. Based on the validation set running data, project it onto the low-dimensional subspace formed by the principal components, calculate the T² statistic and squared prediction error of the running data point, and determine the running data point if it exceeds any corresponding control upper limit threshold. Based on the historical known fault records of the target hydropower station, integrate the historical normal operation data of the target hydropower station to construct the operation knowledge base of the target hydropower station.

[0007] The preferred formula for the combined outlier degree of the T² statistic and the squared prediction error is as follows:

[0008] in, The combined outlier of the T² statistic and the squared prediction error is used to determine the degree of abnormality. For adaptive fusion weights, For statistical purposes, This is the upper limit of the control for the T² statistic. The squared prediction error, This is the upper limit for controlling the squared prediction error. Number of main components For subscript index, For the first data covariance matrix in the running data 1 eigenvalue, This represents the total number of historical normal samples. For the first The squared prediction error value of a normal sample.

[0009] Preferably, real-time operating data of the target hydropower station is acquired based on multi-source heterogeneous sensors, and the data is preprocessed. Based on the preprocessed real-time operation data of the target hydropower station, the data is packaged and substituted into the baseline model of the normal state of the target hydropower station. The projection of the real-time operation data of the target hydropower station in the low-dimensional subspace formed by the principal components is calculated to obtain the T² statistic and squared prediction error of the real-time operation data points. Based on the preset T² statistic and the upper limit threshold of the squared prediction error control, the deviation of real-time running data points is calculated to obtain the abnormal score of real-time running data deviating from the normal state, and to determine whether the real-time running data is abnormal. If an anomaly is identified, the similarity between the real-time feature vector and the fault feature vector in the target hydropower station's operation knowledge base is calculated to achieve a preliminary assessment of the target hydropower station's operational health and fault diagnosis.

[0010] Preferably, the formula for calculating the similarity between the real-time feature vector and the fault feature vector in the target hydropower station's operation knowledge base is as follows:

[0011] in, This represents the comprehensive similarity score between real-time operational data and the j-th fault mode in the target hydropower station's operational knowledge base, where j is the fault mode index, p is the total dimension of the feature vector, and k is the feature variable index. The weights assigned to the k-th feature dimension, This refers to the value of the feature vector obtained after real-time preprocessing and feature extraction in the k-th feature dimension. Let k be the value of the representative feature vector of the j-th fault mode stored in the knowledge base for the operation of the target hydropower station on the k-th feature dimension. Features With fault labels Mutual information between them For the k-th feature variable itself, Let q be the j-th fault label and q be the cycle index. Features With fault labels Mutual information between them This is the smoothing constant.

[0012] Preferably, the operating condition data in the historical operating data of the target hydropower station is obtained, the optimal number of clusters is determined by the silhouette coefficient, several operating condition data are randomly selected as initial cluster centers, the distance from each operating condition data point to several initial cluster centers is calculated, and the operating condition data points are assigned to the nearest cluster centers to form the initial cluster division; The operating condition data includes: steady-state full power generation, steady-state half power generation, startup process, and shutdown process; Calculate the mean of all points in each operating condition data cluster, continuously update the cluster centers until the centers no longer change, and assign a specific operating condition label to each cluster based on the correspondence between the clustering results and the actual operating conditions.

[0013] Preferably, based on assigning specific operating condition labels, dynamic thresholds are set for the health indicators of different operating conditions, and the health of the operating conditions is calculated in real time. If the health is good, it is continuously monitored; if the health deteriorates and exceeds the threshold, multi-level alarm commands are automatically triggered. The multi-level alarm commands include: If the health status is lower than the dynamic threshold and the duration exceeds the set sliding window of five minutes, it will be marked as a level one alarm, and the system will automatically prompt you to pay attention to the trend of deterioration in the operating performance of the hydropower station. If the health status is lower than the dynamic threshold and the rate of deterioration per unit time exceeds the preset limit, it will be marked as a level 2 alarm and automatically notify maintenance personnel to intervene. If the health level falls below the absolute safety threshold, it will be marked as a Level 3 alarm, immediately triggering protective actions and sending a high-priority emergency response notification to maintenance personnel.

[0014] Preferably, if a multi-level alarm command is triggered, the system automatically associates with the target hydropower station's operation knowledge base, and through a similarity matching algorithm, deeply integrates the current operating status and historical fault characteristics to perform high-dimensional spatial pattern recognition and root cause analysis, quickly generating a list of candidate fault causes sorted by confidence level, providing maintenance personnel with a preliminary diagnostic reference; Maintenance personnel handle faults according to system prompts and actual on-site conditions, record the cause of the fault, the handling process and the result data in detail, and organize and code them according to a unified format; The newly processed sample data is fed back into the target hydropower station's operation knowledge base, and the knowledge base is continuously updated using an incremental learning approach to achieve adaptive iterative optimization of the hydropower station's intelligent monitoring system.

[0015] Secondly, a hydropower station intelligent monitoring system is also provided to implement a hydropower station intelligent monitoring method, including: Data acquisition module, fault diagnosis module, multi-level alarm command module and intelligent monitoring module; The data acquisition module is used to acquire historical normal operation data of the target hydropower station, extract the historical normal operation time sequence features of the target hydropower station, establish a normal state baseline model of the target hydropower station, and construct a knowledge base for the operation of the target hydropower station. The fault diagnosis module is used to acquire real-time operating data of the target hydropower station based on multi-source heterogeneous sensors, package and substitute it into the normal state baseline model of the target hydropower station, calculate its abnormal score that deviates from the normal state, and associate and match it with the operating knowledge base of the target hydropower station to achieve preliminary assessment of the health status of the target hydropower station and fault diagnosis. The multi-level alarm instruction module is electrically connected to the fault diagnosis module. It is used to set dynamic thresholds based on the preliminary assessment of the health status of the target hydropower station and fault diagnosis. If the health status is good, it will continue to monitor. If the health status deteriorates and exceeds the threshold, it will automatically trigger the multi-level alarm instruction. The intelligent monitoring module is electrically connected to the multi-level alarm instruction module and the data acquisition module. If the multi-level alarm instruction is triggered, it will automatically associate with the target hydropower station operation knowledge base, and the maintenance personnel will handle the fault. The final fault cause, handling process and result data will be fed back to the target hydropower station operation knowledge base as a new sample, so as to realize the adaptive iteration of the hydropower station intelligent monitoring system.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes an intelligent monitoring method and system for hydropower stations. It establishes a baseline model based on historical normal operation data and combines it with real-time multi-source data for dynamic health assessment, enabling more accurate fault diagnosis and health monitoring. Setting dynamic thresholds and a multi-level alarm system effectively addresses complex operating environments, reducing false alarms and missed alarms, and providing real-time responses to changes in system health. By feeding data from the maintenance process back to a knowledge base, the system can continuously optimize, learn, and iterate adaptively, improving the intelligence and adaptability of the monitoring system, thereby enhancing the operation and maintenance efficiency, safety, and responsiveness of the hydropower station. Attached Figure Description

[0017] Figure 1 A flowchart of a smart monitoring method for hydropower stations; Figure 2This is a framework diagram of an intelligent monitoring system for a hydropower station. Detailed Implementation

[0018] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0019] Reference Figure 1 As shown, an embodiment of the present invention provides a method for intelligent monitoring of a hydropower station, comprising: S1. Obtain historical normal operation data of the target hydropower station, extract the historical normal operation time series features of the target hydropower station, establish a normal state baseline model of the target hydropower station, and construct a knowledge base for the operation of the target hydropower station. Step S1 includes the following: Obtain historical normal operation data of the target hydropower station, perform data preprocessing, and unify timestamps; The operating data includes: turbine speed, shaft vibration, bearing temperature, cooling water temperature, pressure, flow rate, active / reactive power, circuit breaker status, and valve on / off status; Set a time sliding window with a length of ten minutes, and calculate the mean, variance, and standard deviation of the running data within each window; Calculate the kurtosis and skewness of the running data for each window, obtain their root mean square, get the maximum and minimum values ​​of the running data, construct the time-domain feature matrix of the running data, and extract the time-domain features of the running data. For the running data within each sliding window, the Fast Fourier Transform is used to obtain the frequency domain representation of the running data, the spectral amplitude of the running data is calculated, the maximum frequency and frequency centroid of the spectral amplitude are obtained, the frequency domain feature matrix of the running data is constructed, and the frequency domain features of the running data are extracted. Based on the historical normal operation data of the target hydropower station, the start-up process and temperature rise process are identified. An exponential function model is used for fitting to calculate the rate of change, maximum growth rate and steady state value of the operation process. The starting value, ending value, rise time, slope / rate of change and maximum / minimum value of the operation process are obtained, and the trajectory features of the operation data are extracted. By splicing together the time-domain features, frequency-domain features, and trajectory features of the operational data and performing standardization processing, the historical normal operation time sequence features of the target hydropower station are obtained.

[0020] Step S1 also includes the following: Based on the historical normal operation time sequence characteristics of the target hydropower station, it is divided into a training set and a validation set according to time sequence; Based on the historical normal operation time series characteristics of the target hydropower station in the training set, a PCA principal component analysis model is trained, the covariance matrix of the operating data in the training set is calculated, and the corresponding eigenvalues ​​and eigenvectors are obtained. The number of principal components is determined based on the cumulative contribution rate criterion. Based on the historical normal operation data of the target hydropower station in the training set, the data is projected onto the low-dimensional subspace composed of principal components. The T² statistic and squared prediction error of all operation data points are calculated. The upper limit thresholds for the control of the T² statistic and squared prediction error are preset respectively, and a baseline model of the normal state of the target hydropower station is established. The upper limit threshold for the control of the T² statistic and the squared prediction error is taken as the 99th percentile of the historical normal operation data of the target hydropower station. Based on the validation set running data, project it onto the low-dimensional subspace formed by the principal components, calculate the T² statistic and squared prediction error of the running data point. If it exceeds any corresponding upper control threshold, the running data point is determined to be invalid. The formula is as follows:

[0021] in, The combined outlier of the T² statistic and the squared prediction error is used to determine the degree of abnormality. For adaptive fusion weights, For statistical purposes, This is the upper limit of the control for the T² statistic. The squared prediction error, This is the upper limit for controlling the squared prediction error. Number of main components For subscript index, For the first data covariance matrix in the running data 1 eigenvalue, This represents the total number of historical normal samples. For the first The squared prediction error value of a normal sample; Based on the historical known fault records of the target hydropower station, integrate the historical normal operation data of the target hydropower station to construct the operation knowledge base of the target hydropower station.

[0022] When using it, refer to the content of step S1 above: Existing domestic and international technologies for monitoring and detecting anomalies in hydropower station operation data largely rely on simple statistical methods or traditional machine learning algorithms, focusing primarily on single time-domain or frequency-domain features, lacking comprehensive extraction and deep fusion of multi-dimensional features. Furthermore, existing technologies often fail to fully consider the diversity and complexity of operational data, posing a risk of misjudgment or missed detection of anomalies. This is especially problematic in large-scale hydropower station applications, where data noise or system fluctuations can easily interfere, leading to low accuracy and reliability in fault prediction. This approach introduces comprehensive analysis of time-domain, frequency-domain, and trajectory features, combined with principal component analysis for dimensionality reduction, to construct a more comprehensive and efficient baseline model for hydropower station operation. This method accurately reflects the operating status of the hydropower station from multiple dimensions, reducing blind spots caused by single-feature analysis. By calculating the T² statistic and squared prediction error and setting reasonable upper control thresholds, anomaly detection becomes more precise. Integrating historical known fault records into a knowledge base enhances the model's adaptability and intelligence. This significantly improves the predictive performance and fault early warning capabilities of the hydropower station operation monitoring system, thereby enhancing the safety and stability of the hydropower station.

[0023] S2. Based on multi-source heterogeneous sensors, real-time operation data of the target hydropower station is acquired, packaged and substituted into the baseline model of the normal state of the target hydropower station, and its abnormal score deviating from the normal state is calculated. It is then associated and matched with the operation knowledge base of the target hydropower station to achieve a preliminary assessment of the health status of the target hydropower station and fault diagnosis. Step S2 includes the following: Based on multi-source heterogeneous sensors, real-time operation data of the target hydropower station is acquired and data preprocessing is performed. Based on the preprocessed real-time operation data of the target hydropower station, the data is packaged and substituted into the baseline model of the normal state of the target hydropower station. The projection of the real-time operation data of the target hydropower station in the low-dimensional subspace formed by the principal components is calculated to obtain the T² statistic and squared prediction error of the real-time operation data points. Based on the preset T² statistic and the upper limit threshold of the squared prediction error control, the deviation of real-time running data points is calculated to obtain the abnormal score of real-time running data deviating from the normal state, and to determine whether the real-time running data is abnormal. If an anomaly is identified, the similarity between the real-time feature vector and the fault feature vector in the target hydropower station's operation knowledge base is calculated to achieve a preliminary assessment of the target hydropower station's operational health and fault diagnosis. The formula is as follows:

[0024] in, This represents the comprehensive similarity score between real-time operational data and the j-th fault mode in the target hydropower station's operational knowledge base, where j is the fault mode index, p is the total dimension of the feature vector, and k is the feature variable index. The weights assigned to the k-th feature dimension, This refers to the value of the feature vector obtained after real-time preprocessing and feature extraction in the k-th feature dimension. Let k be the value of the representative feature vector of the j-th fault mode stored in the knowledge base for the operation of the target hydropower station on the k-th feature dimension. Features With fault labels Mutual information between them For the k-th feature variable itself, Let q be the j-th fault label and q be the cycle index. Features With fault labels Mutual information between them This is the smoothing constant.

[0025] When using it, refer to the content of step S2 above: Existing domestic and international technologies for real-time operation monitoring and fault diagnosis in hydropower stations often rely on single-sensor data or simple models for anomaly detection, neglecting the integration and comprehensive analysis of multi-source heterogeneous sensor data. This leads to error accumulation and missed diagnoses when facing complex and ever-changing hydropower station operating environments. Most existing methods depend on traditional pattern recognition techniques, lacking in-depth data mining and intelligent processing, resulting in low timeliness and accuracy of fault diagnosis, particularly in multi-dimensional and multi-state health assessment capabilities. This step, through real-time data acquisition and preprocessing from multi-source heterogeneous sensors, substitutes the target hydropower station's operating data into a normal state baseline model constructed through principal component analysis. This allows for the calculation of the T² statistic and squared prediction error of the real-time operating data from a low-dimensional subspace. By comparing this with preset control thresholds, the degree of data deviation is accurately determined, anomalies are detected promptly, and real-time diagnosis is performed. By comparing real-time feature vectors with fault feature vectors in the operation knowledge base, a correlation pairing between fault diagnosis and health assessment can be achieved, significantly improving the accuracy and timeliness of fault detection and diagnosis, and enhancing the safety assurance and operation and maintenance efficiency of hydropower stations.

[0026] S3. Based on the preliminary assessment of the health status of the target hydropower station and fault diagnosis, set dynamic thresholds. If the health status is good, continue to monitor it. If the health status deteriorates and exceeds the threshold, automatically trigger multi-level alarm commands. Step S3 includes the following: Obtain the operating condition data from the historical operating data of the target hydropower station, determine the optimal number of clusters through the silhouette coefficient, randomly select several operating condition data points as initial cluster centers, calculate the distance from each operating condition data point to several initial cluster centers, assign the operating condition data points to the nearest cluster centers, and form the initial cluster partition; The operating condition data includes: steady-state full power generation, steady-state half power generation, startup process, and shutdown process; Calculate the mean of all points in each operating condition data cluster, continuously update the cluster centers until the centers no longer change, and assign a specific operating condition label to each cluster based on the correspondence between the clustering results and the actual operating conditions.

[0027] Step S3 also includes the following: Based on assigning specific operating condition labels, dynamic thresholds are set for the health indicators of different operating conditions, and the health of the operating conditions is calculated in real time. If the health is good, it is continuously monitored; if the health deteriorates and exceeds the threshold, multi-level alarm commands are automatically triggered. The multi-level alarm commands include: If the health status is lower than the dynamic threshold and the duration exceeds the set sliding window of five minutes, it will be marked as a level one alarm, and the system will automatically prompt you to pay attention to the trend of deterioration in the operating performance of the hydropower station. If the health status is lower than the dynamic threshold and the rate of deterioration per unit time exceeds the preset limit, it will be marked as a level 2 alarm and automatically notify maintenance personnel to intervene. If the health level falls below the absolute safety threshold, it will be marked as a Level 3 alarm, immediately triggering protective actions and sending a high-priority emergency response notification to maintenance personnel.

[0028] When using it, refer to the content of step S3 above: Existing domestic and international technologies for assessing the operational health and diagnosing faults in hydropower stations typically rely on fixed threshold settings and static models, failing to dynamically adapt to different operating conditions and their changes. Existing methods also have limitations in alarm triggering mechanisms, struggling to adjust alarm thresholds in real time according to actual operating conditions, easily leading to false alarms or missed alarms, especially in complex, multi-state environments where diagnostic timeliness and accuracy are poor. This step utilizes cluster analysis to identify different operating conditions and sets dynamic thresholds based on the characteristics of each condition. By calculating health status in real time and monitoring its changing trends, the health status of the hydropower station can be more accurately identified, triggering multi-level alarm commands based on specific circumstances. This dynamic, real-time monitoring mechanism effectively reduces the risk of false alarms or missed alarms, improves the system's adaptability under complex operating conditions, and thus enhances the safety and operational efficiency of the hydropower station.

[0029] S4. If a multi-level alarm command is triggered, the system will automatically associate with the target hydropower station's operation knowledge base. Maintenance personnel will then handle the fault and feed back the final fault cause, handling process, and result data as new samples to the target hydropower station's operation knowledge base, thereby enabling the adaptive iteration of the hydropower station's intelligent monitoring system. Step S4 includes the following: If a multi-level alarm command is triggered, the system automatically associates with the target hydropower station's operation knowledge base, and through a similarity matching algorithm, deeply integrates the current operating status and historical fault characteristics to perform high-dimensional spatial pattern recognition and root cause analysis, quickly generating a list of candidate fault causes sorted by confidence level, providing maintenance personnel with a preliminary diagnostic reference. Maintenance personnel handle faults according to system prompts and actual on-site conditions, record the cause of the fault, the handling process and the result data in detail, and organize and code them according to a unified format; The newly processed sample data is fed back into the target hydropower station's operation knowledge base, and the knowledge base is continuously updated using an incremental learning approach to achieve adaptive iterative optimization of the hydropower station's intelligent monitoring system.

[0030] When using it, refer to the content of step S4 above: Existing domestic and international technologies for fault diagnosis and adaptive iteration in hydropower stations typically lack real-time knowledge base feedback mechanisms and rely on human experience during fault handling, making it difficult to quickly and accurately provide maintenance personnel with diagnostic references based on the current operating status. Existing methods, when processing historical fault data, often rely solely on traditional rules or static models, failing to dynamically adjust and optimize according to actual conditions, resulting in low system intelligence and an inability to adapt to new fault modes or equipment changes in a timely manner. This step, by introducing similarity matching algorithms and high-dimensional spatial pattern recognition technology, can deeply integrate the current operating status and historical fault characteristics when a fault occurs, quickly generating a candidate list of fault causes sorted by confidence level, providing maintenance personnel with accurate preliminary diagnostic references. By feeding new sample data during processing back to the operating knowledge base and updating it using incremental learning, the system can achieve adaptive iteration and optimization, continuously improving the intelligence level of fault diagnosis and handling. This dynamic feedback mechanism not only improves the system's real-time responsiveness but also effectively reduces human intervention and errors, enhancing the accuracy and efficiency of fault diagnosis.

[0031] Reference Figure 2 As shown, one embodiment of the present invention also provides an intelligent monitoring system for hydropower stations, comprising: Data acquisition module, fault diagnosis module, multi-level alarm command module and intelligent monitoring module; The data acquisition module is used to acquire historical normal operation data of the target hydropower station, extract the historical normal operation time sequence features of the target hydropower station, establish a normal state baseline model of the target hydropower station, and construct a knowledge base for the operation of the target hydropower station. The fault diagnosis module is used to acquire real-time operating data of the target hydropower station based on multi-source heterogeneous sensors, package and substitute it into the normal state baseline model of the target hydropower station, calculate its abnormal score that deviates from the normal state, and associate and match it with the operating knowledge base of the target hydropower station to achieve preliminary assessment of the health status of the target hydropower station and fault diagnosis. The multi-level alarm instruction module is electrically connected to the fault diagnosis module. It is used to set dynamic thresholds based on the preliminary assessment of the health status of the target hydropower station and fault diagnosis. If the health status is good, it will continue to monitor. If the health status deteriorates and exceeds the threshold, it will automatically trigger the multi-level alarm instruction. The intelligent monitoring module is electrically connected to the multi-level alarm instruction module and the data acquisition module. If the multi-level alarm instruction is triggered, it will automatically associate with the target hydropower station operation knowledge base, and the maintenance personnel will handle the fault. The final fault cause, handling process and result data will be fed back to the target hydropower station operation knowledge base as a new sample, so as to realize the adaptive iteration of the hydropower station intelligent monitoring system.

[0032] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for intelligent monitoring of a hydropower station, characterized in that, include: S1. Obtain historical normal operation data of the target hydropower station, extract the historical normal operation time series features of the target hydropower station, establish a normal state baseline model of the target hydropower station, and construct a knowledge base for the operation of the target hydropower station. S2. Based on multi-source heterogeneous sensors, real-time operation data of the target hydropower station is acquired, packaged and substituted into the baseline model of the normal state of the target hydropower station, and its abnormal score deviating from the normal state is calculated. It is then associated and matched with the operation knowledge base of the target hydropower station to achieve a preliminary assessment of the health status of the target hydropower station and fault diagnosis. S3. Based on the preliminary assessment of the health status of the target hydropower station and fault diagnosis, set dynamic thresholds. If the health status is good, continue to monitor it. If the health status deteriorates and exceeds the threshold, automatically trigger multi-level alarm commands. S4. If a multi-level alarm command is triggered, the system will automatically link to the target hydropower station's operation knowledge base. Maintenance personnel will then handle the fault and feed back the final fault cause, handling process, and result data as new samples to the target hydropower station's operation knowledge base, thereby enabling the adaptive iteration of the hydropower station's intelligent monitoring system.

2. The intelligent monitoring method for a hydropower station according to claim 1, characterized in that, S1 includes: Obtain historical normal operation data of the target hydropower station, perform data preprocessing, and unify timestamps; The operating data includes: turbine speed, shaft vibration, bearing temperature, cooling water temperature, pressure, flow rate, active / reactive power, circuit breaker status, and valve on / off status; Set a time sliding window with a length of ten minutes, and calculate the mean, variance, and standard deviation of the running data within each window; Calculate the kurtosis and skewness of the running data for each window, obtain their root mean square, get the maximum and minimum values ​​of the running data, construct the time-domain feature matrix of the running data, and extract the time-domain features of the running data. For the running data within each sliding window, the Fast Fourier Transform is used to obtain the frequency domain representation of the running data, the spectral amplitude of the running data is calculated, the maximum frequency and frequency centroid of the spectral amplitude are obtained, the frequency domain feature matrix of the running data is constructed, and the frequency domain features of the running data are extracted. Based on the historical normal operation data of the target hydropower station, the start-up process and temperature rise process are identified. An exponential function model is used for fitting to calculate the rate of change, maximum growth rate and steady state value of the operation process. The starting value, ending value, rise time, slope / rate of change and maximum / minimum value of the operation process are obtained, and the trajectory features of the operation data are extracted. By splicing together the time-domain features, frequency-domain features, and trajectory features of the operational data and performing standardization processing, the historical normal operation time sequence features of the target hydropower station are obtained.

3. The intelligent monitoring method for a hydropower station according to claim 2, characterized in that, S1 further includes: Based on the historical normal operation time sequence characteristics of the target hydropower station, it is divided into a training set and a validation set according to time sequence; Based on the historical normal operation time series characteristics of the target hydropower station in the training set, a PCA principal component analysis model is trained, the covariance matrix of the operating data in the training set is calculated, and the corresponding eigenvalues ​​and eigenvectors are obtained. The number of principal components is determined based on the cumulative contribution rate criterion. Based on the historical normal operation data of the target hydropower station in the training set, the data is projected onto the low-dimensional subspace composed of principal components. The T² statistic and squared prediction error of all operation data points are calculated. The upper limit thresholds for the control of the T² statistic and squared prediction error are preset respectively, and a baseline model of the normal state of the target hydropower station is established. The upper limit threshold for the control of the T² statistic and the squared prediction error is taken as the 99th percentile of the historical normal operation data of the target hydropower station. Based on the validation set running data, project it onto the low-dimensional subspace formed by the principal components, calculate the T² statistic and squared prediction error of the running data point, and determine the running data point if it exceeds any corresponding control upper limit threshold. Based on the historical known fault records of the target hydropower station, integrate the historical normal operation data of the target hydropower station to construct the operation knowledge base of the target hydropower station.

4. The intelligent monitoring method for a hydropower station according to claim 3, characterized in that, The formula for the combined outlier degree of the T² statistic and the squared prediction error is as follows: in, The combined outlier of the T² statistic and the squared prediction error is used to determine the degree of abnormality. For adaptive fusion weights, For statistical purposes, This is the upper limit of the control for the T² statistic. The squared prediction error, This is the upper limit for controlling the squared prediction error. Number of main components For subscript index, For the first data covariance matrix in the running data 1 eigenvalue, This represents the total number of historical normal samples. For the first The squared prediction error value of a normal sample.

5. The intelligent monitoring method for a hydropower station according to claim 1, characterized in that, S2 includes: Based on multi-source heterogeneous sensors, real-time operation data of the target hydropower station is acquired and data preprocessing is performed. Based on the preprocessed real-time operation data of the target hydropower station, the data is packaged and substituted into the normal state baseline model of the target hydropower station. The projection of the real-time operation data of the target hydropower station in the low-dimensional subspace formed by the principal components is calculated to obtain the T² statistic and squared prediction error of the real-time operation data points. Based on the preset T² statistic and the upper limit threshold for squared prediction error control, the deviation of real-time running data points is calculated to obtain the abnormal score of real-time running data deviating from the normal state, and to determine whether the real-time running data is abnormal. If an anomaly is identified, the similarity between the real-time feature vector and the fault feature vector in the target hydropower station's operation knowledge base is calculated to achieve a preliminary assessment of the target hydropower station's operational health and fault diagnosis.

6. The intelligent monitoring method for a hydropower station according to claim 5, characterized in that, The formula for calculating the similarity between the real-time feature vector and the fault feature vector in the target hydropower station's operation knowledge base is as follows: in, This represents the comprehensive similarity score between real-time operational data and the j-th fault mode in the target hydropower station's operational knowledge base, where j is the fault mode index, p is the total dimension of the feature vector, and k is the feature variable index. The weights assigned to the k-th feature dimension, This refers to the value of the feature vector obtained after real-time preprocessing and feature extraction in the k-th feature dimension. Let k be the value of the representative feature vector of the j-th fault mode stored in the knowledge base for the operation of the target hydropower station on the k-th feature dimension. Features With fault labels Mutual information between them For the k-th feature variable itself, Let q be the j-th fault label and q be the cycle index. Features With fault labels Mutual information between them This is the smoothing constant.

7. The intelligent monitoring method for a hydropower station according to claim 5, characterized in that, S3 includes: Obtain the operating condition data from the historical operating data of the target hydropower station, determine the optimal number of clusters through the silhouette coefficient, randomly select several operating condition data points as initial cluster centers, calculate the distance from each operating condition data point to several initial cluster centers, assign the operating condition data points to the nearest cluster centers, and form the initial cluster partition; The operating condition data includes: steady-state full power generation, steady-state half power generation, startup process, and shutdown process; Calculate the mean of all points in each operating condition data cluster, continuously update the cluster centers until the centers no longer change, and assign a specific operating condition label to each cluster based on the correspondence between the clustering results and the actual operating conditions.

8. The intelligent monitoring method for a hydropower station according to claim 7, characterized in that, S3 further includes: Based on assigning specific operating condition labels, dynamic thresholds are set for the health indicators of different operating conditions, and the health of the operating conditions is calculated in real time. If the health is good, it is continuously monitored; if the health deteriorates and exceeds the threshold, multi-level alarm commands are automatically triggered. The multi-level alarm commands include: If the health status is lower than the dynamic threshold and the duration exceeds the set sliding window of five minutes, it will be marked as a level one alarm, and the system will automatically prompt you to pay attention to the trend of deterioration in the operating performance of the hydropower station. If the health status is lower than the dynamic threshold and the rate of deterioration per unit time exceeds the preset limit, it will be marked as a level 2 alarm and automatically notify maintenance personnel to intervene. If the health level falls below the absolute safety threshold, it will be marked as a Level 3 alarm, immediately triggering protective actions and sending a high-priority emergency response notification to maintenance personnel.

9. The intelligent monitoring method for a hydropower station according to claim 8, characterized in that, S4 includes: If a multi-level alarm command is triggered, the system automatically associates with the target hydropower station's operation knowledge base, and through a similarity matching algorithm, deeply integrates the current operating status and historical fault characteristics to perform high-dimensional spatial pattern recognition and root cause analysis, quickly generating a list of candidate fault causes sorted by confidence level, providing maintenance personnel with a preliminary diagnostic reference. Maintenance personnel handle faults according to system prompts and actual on-site conditions, record the cause of the fault, the handling process and the result data in detail, and organize and code them according to a unified format; The newly processed sample data is fed back into the target hydropower station's operation knowledge base, and the knowledge base is continuously updated using an incremental learning approach to achieve adaptive iterative optimization of the hydropower station's intelligent monitoring system.

10. A smart monitoring system for a hydropower station, characterized in that, The intelligent monitoring system for hydropower stations, which is used to implement the intelligent monitoring method for hydropower stations according to any one of claims 1-9, includes a data acquisition module, a fault diagnosis module, a multi-level alarm instruction module, and an intelligent monitoring module. The data acquisition module is used to acquire historical normal operation data of the target hydropower station, extract the historical normal operation time sequence features of the target hydropower station, establish a normal state baseline model of the target hydropower station, and construct a knowledge base for the operation of the target hydropower station. The fault diagnosis module is used to acquire real-time operating data of the target hydropower station based on multi-source heterogeneous sensors, package and substitute it into the normal state baseline model of the target hydropower station, calculate its abnormal score that deviates from the normal state, and associate and match it with the operating knowledge base of the target hydropower station to achieve preliminary assessment of the health status of the target hydropower station and fault diagnosis. The multi-level alarm instruction module is electrically connected to the fault diagnosis module. It is used to set dynamic thresholds based on the preliminary assessment of the health status of the target hydropower station and fault diagnosis. If the health status is good, it will continue to monitor. If the health status deteriorates and exceeds the threshold, it will automatically trigger the multi-level alarm instruction. The intelligent monitoring module is electrically connected to the multi-level alarm instruction module and the data acquisition module. If the multi-level alarm instruction is triggered, it will automatically associate with the target hydropower station operation knowledge base, and the maintenance personnel will handle the fault. The final fault cause, handling process and result data will be fed back to the target hydropower station operation knowledge base as a new sample, so as to realize the adaptive iteration of the hydropower station intelligent monitoring system.

Images