Method and system for realizing intelligent disk monitoring of hydraulic power plant

Abstract

The invention discloses an intelligent monitoring implementation method and system for a hydraulic power plant, belongs to the technical field of hydraulic power plant monitoring, and aims at solving the problems that a traditional monitoring method depends on manpower and is low in efficiency and accuracy. The method comprises the following steps: firstly, collecting mechanical, thermal and electrical multi-modal data of a hydraulic power plant in real time, and carrying out standardized preprocessing; secondly, constructing a special anomaly recognition model, and extracting mechanical, thermal and electrical anomaly features by using wavelet packet decomposition, an LSTM algorithm, a CNN and other technologies; then, constructing a coupling anomaly recognition model, and analyzing a coupling relationship among the anomaly features; then calculating a global abnormal coefficient based on the abnormal coefficient output by each model, evaluating an alarm confidence coefficient, and implementing a three-level alarm suppression strategy; and finally, pushing an AR picture to visually display an equipment state, and issuing a repair task to realize closed-loop management. The system integrates the functional modules in the method. According to the invention, the intelligent level of monitoring is obviously improved, and the operation safety and stability of the hydraulic power plant are enhanced.

Description

Technical Field

[0001] This invention relates to the field of hydropower plant operation monitoring technology, and in particular to a method and system for implementing intelligent monitoring panels in hydropower plants. Background Technology

[0002] In the field of hydropower plant operation monitoring, monitoring panels are a crucial means of ensuring the safe, reliable, and stable operation of power plants and have always been highly valued by the industry. Traditional monitoring methods mainly rely on on-duty personnel to conduct real-time monitoring, data analysis, operational adjustments, and accident handling of power plant equipment, facilities, and systems through control room feedback screens, monitoring systems, and online monitoring devices. However, this method has revealed many problems in practical applications, affecting the efficiency and accuracy of monitoring panels.

[0003] Currently, several technologies are attempting to improve the intelligence level of hydropower plant monitoring. For example, CN118260542A proposes an intelligent monitoring system and method, which includes a data acquisition module, a model building module, and an operation guidance module. The data acquisition module acquires historical and real-time operating data of industrial production equipment and performs preprocessing; the model building module constructs a neural network model and obtains a fault diagnosis model through training on historical data; the operation guidance module provides real-time guidance based on real-time operating data and the fault diagnosis model. Furthermore, CN119472357A proposes an unmanned intelligent monitoring system and method for power plants, which collects and processes real-time relevant data from power plant pipelines to construct pipeline early warning rules, thereby achieving unmanned intelligent monitoring.

[0004] Although the above technologies have improved the intelligence level of hydropower plant monitoring systems to some extent, the following main problems still exist: 1. Insufficient Multimodal Data Fusion: Existing technologies often focus on the analysis of single-type data (such as mechanical data only or electrical data only), failing to fully consider the coupling and correlation between multimodal data such as mechanical, thermal, and electrical data. This single-modal analysis method is prone to false alarms or false negatives, and cannot comprehensively and accurately reflect the equipment status. For example, CN118260542A and CN119472357A do not mention the fusion analysis of multimodal data, which may lead to misjudgment of complex equipment faults.

[0005] 2. Lack of Coupling Analysis: Equipment failures are often not caused by a single factor, but rather the result of the interaction of multiple factors. However, most existing technologies lack in-depth analysis of the coupling between multimodal data, making it impossible to accurately identify complex failure modes. For example, abnormal mechanical vibration may be closely related to changes in electrical parameters, but existing technologies often fail to reveal this correlation.

[0006] 3. Inadequate alarm suppression and closed-loop handling mechanisms: Traditional monitoring methods generate numerous alarm messages when equipment malfunctions, but lack effective suppression measures and closed-loop handling methods, resulting in low monitoring efficiency and high reliance on manual experience. Although existing technologies have proposed some intelligent monitoring methods, they still fall short in terms of alarm suppression strategies and closed-loop handling mechanisms.

[0007] 4. The Balance Between Real-Time Performance and Accuracy: Hydropower plant equipment status changes rapidly, requiring monitoring systems to possess a high degree of real-time performance. However, existing technologies often sacrifice real-time performance in pursuit of high accuracy, resulting in delayed responses to equipment anomalies. How to improve real-time performance while ensuring accuracy is a pressing issue that current technologies need to address.

[0008] To address the aforementioned issues, this invention proposes a method and system for implementing intelligent monitoring in hydropower plants. The aim is to comprehensively analyze multimodal data, including mechanical, thermal, and electrical data, to reveal the coupling and correlation between these data, thereby improving the comprehensiveness and accuracy of intelligent monitoring. Simultaneously, it aims to improve alarm suppression and closed-loop handling mechanisms, enabling real-time and accurate monitoring and early warning of hydropower plant equipment status. Summary of the Invention

[0009] The technical problem this invention aims to solve is to provide a method and system for implementing intelligent monitoring in hydropower plants, addressing the issues of low efficiency and insufficient accuracy in the monitoring process. Traditional monitoring methods rely on manual review of monitoring pages and trend analysis, making it difficult to promptly detect abnormal equipment states before fixed threshold alarms. Furthermore, in the event of equipment failure, the sheer volume of information lacks effective suppression measures and closed-loop handling procedures, resulting in low monitoring efficiency and a high dependence on human experience. To overcome these limitations, this invention proposes a method and system for implementing intelligent monitoring in hydropower plants, which improves the level of intelligence in monitoring by integrating multimodal data analysis and intelligent recognition technologies.

[0010] To achieve the above objectives, the present invention adopts a comprehensive technical solution, namely, a method and system for implementing intelligent monitoring panels in hydropower plants.

[0011] The specific steps for implementing intelligent monitoring panels in hydropower plants are as follows: 1. Multimodal data acquisition and preprocessing: The data acquisition module acquires real-time mechanical data (such as vibration signal data of turbine runner and main shaft), thermal data (such as thermal images of generator stator and rotor windings), and electrical data (such as insulation status data and operating parameter data of generator stator windings and main transformer). The collected multimodal data undergoes standardized preprocessing, including noise reduction, smoothing, pixel value conversion, and Z-Score standardization, to ensure data quality and consistency.

[0012] 2. Construction and operation of a specialized anomaly identification model: A first mechanical data anomaly identification model is constructed, and mechanical anomaly features are extracted using wavelet packet decomposition, LSTM algorithm and other techniques, and the mechanical anomaly feature matrix and anomaly coefficients are output; A second thermal anomaly identification model is constructed, and CNN is used to extract and analyze features from thermal images, outputting thermal anomaly feature matrix and anomaly coefficients; A third electrical anomaly identification model is constructed. Electrical data is processed through machine learning algorithms to identify electrical anomaly characteristics and output electrical anomaly feature matrix and anomaly coefficients.

[0013] 3. Coupling anomaly identification and global anomaly assessment: A fourth coupling anomaly identification model is constructed to comprehensively analyze the coupling relationship between mechanical, thermal and electrical anomaly features, calculate the coupling coefficients of binary and ternary anomaly features, and comprehensively evaluate the equipment status. Based on the anomaly coefficients output by each specific model and the coupled model, the global anomaly coefficient is calculated, and an alarm confidence assessment system is constructed.

[0014] 4. Alarm Suppression and Task Issuance: Based on the global anomaly coefficient and alarm confidence level, a three-level alarm suppression strategy is implemented to ensure the accuracy and timeliness of alarm information; Send AR augmented reality images to monitoring personnel to intuitively display equipment status and abnormal information, facilitating rapid decision-making; It supports task publishing, automatically or manually sending repair notifications to staff to achieve closed-loop management of fault handling.

[0015] To implement the intelligent monitoring method for hydropower plants, this invention also provides an intelligent monitoring system for hydropower plants. This system integrates the various modules and functions described in the above-mentioned implementation method, specifically including: 1. Data Acquisition Module: Responsible for real-time acquisition of mechanical, thermal, and electrical data from the hydropower plant, and performing preliminary processing; 2. Specialized Anomaly Identification Module: Includes a first mechanical data anomaly identification model, a second thermal anomaly identification model, and a third electrical anomaly identification model, which are used to extract and analyze mechanical, thermal, and electrical anomaly features, respectively; 3. Coupling Anomaly Identification Module: Also known as the fourth coupling anomaly identification model, it is used to analyze the coupling relationship between various specific anomaly features and output the coupling anomaly coefficient; 4. Global Anomaly Assessment and Alarm Module: Based on the anomaly coefficients output by each specialized model and the coupled model, calculate the global anomaly coefficient, construct an alarm confidence assessment system, and implement a three-level alarm suppression strategy. 5. AR Image Push and Task Release Module: Responsible for sending AR augmented reality images to monitoring personnel, displaying equipment status and abnormal information, and supporting task release function to achieve closed-loop management of fault handling.

[0016] The intelligent monitoring method and system for hydropower plants of the present invention significantly improves the efficiency and accuracy of monitoring in hydropower plants by integrating multimodal data analysis and intelligent recognition technologies, reduces false alarms and missed alarms, and reduces reliance on human experience, thus providing strong support for the intelligent development of the hydropower industry.

[0017] The intelligent monitoring system for hydropower plants provided by this invention has the following beneficial effects: 1. This invention breaks through the limitation of a single data source in traditional systems, and for the first time realizes the synchronous acquisition and joint analysis of mechanical, thermal and electrical three-modal data, filling the gap in cross-modal correlation analysis technology; it reveals the coupling correlation between multimodal data, solves the problem of traditional methods ignoring the intrinsic relationship between data, overcomes the technical limitation of existing technologies that do not consider the interaction between mechanical vibration and thermal expansion, electrical discharge and insulation aging, making it difficult to accurately identify multi-factor coupled faults, and successfully solves the technical problem of traditional systems relying only on a single type of data (such as only vibration or temperature), resulting in a high rate of missed detection of early equipment anomalies and a low rate of identification of composite faults.

[0018] 2. This invention is the first to propose a three-level alarm strategy based on a global anomaly coefficient. Combined with the alarm confidence level, the handling method is dynamically adjusted to effectively suppress false alarms and ensure that high-risk anomalies are triggered in a timely manner. This solves the problem in the existing technology where fixed threshold alarms cause an "alarm storm" (more than 100 false alarms per day) and the high rate of missed alarms for high-risk anomalies, which seriously threatens equipment safety.

[0019] 3. This invention is the first in the industry to apply AR technology to the monitoring scenario of hydropower plants. By overlaying equipment status parameters in real time through a three-dimensional visualization interface, it improves the efficiency of anomaly location and overcomes the limitation of traditional two-dimensional monitoring interfaces that cannot intuitively present key parameters such as equipment thermal distribution and vibration spectrum, resulting in excessively long fault diagnosis time.

[0020] 4. This invention effectively obtains various anomaly feature matrices and anomaly coefficients by using multimodal data fusion and coupling analysis, and by constructing a first mechanical data anomaly identification model, a second thermal anomaly identification model, a third electrical anomaly identification model, and a fourth coupled anomaly identification model. Based on these coefficients, a global anomaly coefficient is obtained, which avoids the problem of limited accuracy in intelligent monitoring anomaly analysis of hydropower plants due to single-modal data, and comprehensively improves the comprehensiveness and accuracy of intelligent monitoring anomaly analysis.

[0021] 5. After the implementation of this invention, the early fault identification rate is significantly improved compared with the industry average, and key indicators such as mechanical vibration prediction accuracy and insulation aging early warning lead time reach the industry leading level.

[0022] 6. The implementation of this invention significantly reduces the false alarm rate, ensures that 100% of high-risk anomalies trigger audible and visual alarms, and significantly reduces the annual misoperation rate, effectively solving the problem of the coexistence of "alarm storms" and missed alarms.

[0023] 7. The system designed in this invention improves monitoring efficiency, significantly shortens fault location time, reduces annual maintenance costs per unit, and yields significant economic benefits.

[0024] 8. This invention provides strong support for the maintenance and management of hydropower plant equipment through real-time monitoring and data analysis, which helps to extend the service life of equipment and reduce operation and maintenance costs.

[0025] 9. The implementation of this invention has formed a complete technical chain of "data fusion - feature extraction - coupling analysis - risk quantification - AR presentation", setting a new standard for intelligent monitoring in the hydropower industry.

[0026] 10. Through theoretical derivation and thorough verification by practical application cases, it is shown that the present invention has significant technical advantages and broad application prospects in the field of hydropower plant operation monitoring. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the structure of the first mechanical data anomaly identification model of the present invention: Figure 2 This is a schematic diagram of the structure of the second thermal anomaly identification model of the present invention; Figure 3 This is a schematic diagram of the third electrical anomaly identification model of the present invention; Figure 4 This is a schematic diagram of the fourth coupling anomaly identification model structure of the present invention; Figure 5 This is a schematic diagram of the global anomaly coefficient acquisition process of the present invention. Detailed Implementation

[0028] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments: Example 1 like Figures 1 to 5 As shown in the figure, this embodiment provides a method for implementing intelligent monitoring panels in hydropower plants. The specific methods and steps are as follows: Step 1: Data Acquisition 1) First mechanical data acquisition: In hydropower plants, corresponding sensors are installed for the primary monitoring targets, including the turbine runner, main shaft, guide vanes, and draft tube. For example, vibration sensors are installed on the turbine runner and main shaft to collect runner and main shaft vibration signals; oscillation meters are installed to collect main shaft oscillation signals; pressure sensors are installed at the runner and draft tube to acquire runner and draft tube pressure pulsation signals; and displacement sensors are installed at the guide vanes to collect guide vane opening signal data. These collected data are then transmitted to a data storage and processing unit.

[0029] 2) Acquisition of secondary thermal data: For the second monitoring targets, such as the generator stator, rotor windings, thrust bearing pads, and water pumps, thermal imagers are used to collect thermal images of the stator windings, rotor windings, thrust bearing pads, and water pump bearing surfaces. Simultaneously, multiple zones are divided on the heat exchanger's metal wall, and temperature sensors are used to collect temperature data for each zone. The thermal images and temperature data are then transmitted to the data processing center.

[0030] 3) Third Electrical Data Acquisition: For the third monitoring targets, including the generator stator windings, main transformer, GIS (Gas Insulated Switchgear) switchgear, and excitation system, corresponding electrical monitoring equipment is installed. For example, a partial discharge detector is used to collect partial discharge data from the stator windings, main transformer, and GIS; an insulation resistance tester is used to measure insulation resistance; a harmonic analyzer is used to collect stator harmonic current data; and a current sensor is installed in the excitation system to collect excitation current data. This electrical data is then aggregated and transmitted to the data processing module.

[0031] Step 2: Model Building 1) Construction of the first mechanical data anomaly identification model (1) Input layer: The first mechanical data obtained is used as input.

[0032] (2) Preprocessing layer: The first mechanical data is standardized by using the Z-Score (standard variable) algorithm to obtain the second mechanical data.

[0033] (3) Anomaly feature extraction layer: Wavelet packet decomposition was used to extract the high-frequency energy ratio of the vibration signal data of the impeller and the main shaft, which was then used as the high-frequency energy ratio feature of the impeller and the main shaft. The vibration amplitude of the spindle yaw signal data is analyzed by the LSTM (Long Short-Term Memory) algorithm to predict the vibration amplitude for the next hour, and the spindle yaw residual is calculated by combining the actual vibration amplitude sequence. Bandpass filtering was applied to the pressure pulsation signal data of the runner and tailrace pipe, and then sampled in a 1-second window to obtain the Person correlation coefficient as the cross-correlation coefficient feature of the pressure pulsation. The standard deviation of guide vane opening fluctuation is calculated as the guide vane opening feature, and the extracted features are used to form the first mechanical anomaly feature matrix.

[0034] (4) Abnormal feature analysis layer: The first mechanical abnormal feature matrix is ​​analyzed according to the preset rules and algorithms to obtain the first mechanical abnormal coefficient.

[0035] (5) Output layer: Output the first mechanical anomaly feature matrix and the first mechanical anomaly coefficient.

[0036] 2) Construction of the second thermal anomaly identification model: (1) Input layer: Input the second thermal data, including thermal images and temperature data.

[0037] (2) Preprocessing layer: Denoise and smooth the thermal image, convert the image pixel values ​​into actual temperature values, determine the monitoring points, and standardize the temperature data to obtain the third thermal data.

[0038] (3) Anomaly Feature Extraction Layer: A Convolutional Neural Network (CNN) is used to extract features from the third thermal data to obtain the second thermal anomaly features, including the anomaly degree of stator winding temperature gradient, the temperature fluctuation rate of rotor winding, the temperature non-uniformity of thrust bearing pads, the hot spot identification coefficient of water pump bearing, and the temperature anomaly coefficient of heat exchanger metal wall, etc., and these features are formed into a matrix. These features are calculated based on parameters such as monitoring point temperature, spatial distance, standard deviation, mean, maximum and minimum temperature, pixel temperature, hot spot threshold, regional average temperature, and reference average temperature using corresponding formulas.

[0039] (4) Anomaly Feature Analysis Layer: Analyze the second thermal anomaly feature matrix to obtain the second thermal anomaly coefficient.

[0040] (5) Output layer: Output the second thermal anomaly feature matrix and the second thermal anomaly coefficient.

[0041] 3) Construction of the third electrical anomaly identification model: (1) Input layer: Input the third electrical data.

[0042] (2) Preprocessing layer: The third electrical data is standardized using the Z-Score algorithm to obtain the fourth electrical data.

[0043] (3) Anomaly Feature Extraction Layer: Machine learning algorithms (such as random forest, support vector machine, etc.) are used to extract the third electrical anomaly features, including the stator winding partial discharge intensity index, transformer partial discharge energy accumulation rate, GIS partial discharge frequency anomaly degree, insulation resistance comprehensive attenuation rate, current anomaly degree, etc., and form a matrix. These features are calculated based on parameters such as the partial discharge pulse amplitude time function, transformer partial discharge energy, partial discharge frequency, insulation resistance value, current harmonic distortion rate, and excitation current fluctuation rate through corresponding formulas.

[0044] (4) Abnormal feature analysis layer: Analyze the third electrical abnormal feature matrix to obtain the third electrical abnormal coefficient.

[0045] (5) Output layer: Output the third electrical anomaly feature matrix and the third electrical anomaly coefficient.

[0046] 4) Construction of the fourth coupling anomaly identification model: (1) Input layer: Input the first mechanical, second thermal and third electrical anomaly feature matrices.

[0047] (2) Coupling anomaly feature extraction layer: Based on the Person correlation function of the first mechanical, second thermal and third electrical anomaly feature matrices and the mean value of each matrix feature, the bivariate (mechanical-thermal, mechanical-electrical, thermal-electrical anomaly coupling coefficient) and ternary (mechanical-thermal-electrical anomaly coupling coefficient) anomaly feature coupling coefficients are calculated by corresponding formula.

[0048] (3) Coupled anomaly feature analysis layer: The fourth coupled anomaly coefficient is calculated based on the weight coefficients of the binary and ternary anomaly feature coupling coefficients.

[0049] (4) Output layer: Output the fourth coupling anomaly coefficient.

[0050] Step 3: Anomaly Identification 1) The first mechanical data anomaly identification model, the second thermal anomaly identification model, the third electrical anomaly identification model, and the fourth coupling anomaly identification model are used to identify anomalies in the input data.

[0051] 2) The first mechanical data anomaly identification model outputs the first mechanical anomaly feature matrix and the first mechanical anomaly coefficient; the second thermal anomaly identification model outputs the second thermal anomaly feature matrix and the second thermal anomaly coefficient; the third electrical anomaly identification model outputs the third electrical anomaly feature matrix and the third electrical anomaly coefficient; and the fourth coupling anomaly identification model outputs the fourth coupling anomaly coefficient.

[0052] Step 4: Coefficient Calculation and Alarm 1) Calculation of global anomaly coefficient: Based on the first mechanical anomaly coefficient, the second thermal anomaly coefficient, the third electrical anomaly coefficient, and the fourth coupling anomaly coefficient, the global anomaly coefficient is calculated using a weighted average or other suitable algorithm.

[0053] 2) Construction of the first alarm confidence level: The first alarm confidence level is constructed based on the global anomaly coefficient and the preset rules. The confidence level can be set to [0, 1].

[0054] 3) Determination of the three-level alarm suppression strategy: (1) Level 1 suppression: When the confidence level of the first alarm is lower than the preset low threshold, the low confidence alarm is automatically suppressed and no prompt is given; (2) Secondary suppression: When the confidence level of the first alarm is between the preset low threshold and the high threshold, the medium confidence alarm is pushed to the monitoring interface and marked as "to be confirmed" to remind the monitoring personnel to pay attention; (3) Level 3 suppression: When the confidence level of the first alarm is higher than the preset high threshold, an audible and visual alarm is triggered, which attracts the attention of the monitoring personnel.

[0055] 4) Alarm Information Sending: Send AR (Augmented Reality) images to monitoring personnel so that they can understand the abnormal equipment situation more intuitively. The monitoring personnel can issue tasks based on the alarm information. At the same time, send repair notices to staff to notify them to repair the abnormal equipment in a timely manner.

[0056] Example 2 In another preferred embodiment, based on the above embodiment 1, this embodiment provides a detailed description of the specific implementation method and steps of the intelligent monitoring panel implementation method for hydropower plants: Step 1: Data Acquisition Acquire the first mechanical data of the first monitoring object in the hydropower plant; acquire the second thermal data of the second monitoring object in the hydropower plant; acquire the third electrical data of the third monitoring object in the hydropower plant.

[0057] The first monitoring objects include the turbine runner, main shaft, guide vanes, and draft tube; the first mechanical data includes runner vibration signal data, main shaft vibration signal data, main shaft swing signal data, runner pressure pulsation signal data, draft tube pressure pulsation signal data, and guide vane opening signal data.

[0058] The second monitoring objects include the generator stator, rotor winding, thrust bearing pads, and water pump; the second thermal data includes thermal images of the stator winding, rotor winding, thrust bearing pads, water pump bearing surface, and heat exchanger metal wall after division into regions.

[0059] The third monitoring object includes the generator stator winding, main transformer, GIS switchgear and excitation system; the third electrical data includes insulation status data and operating parameter data; the insulation status data includes stator winding partial discharge, main transformer partial discharge, GIS partial discharge, stator winding insulation resistance, main transformer insulation resistance and GIS insulation resistance; the operating parameter data includes stator harmonic current and excitation current.

[0060] A first mechanical data anomaly identification model is constructed, which includes a first input layer, a first preprocessing layer, a first mechanical anomaly feature extraction layer, a first mechanical anomaly feature analysis layer, and a first output layer.

[0061] The first input layer is used to input the first mechanical data into the first mechanical data anomaly recognition model.

[0062] The first preprocessing layer is used to standardize the first mechanical data using the Z-Score algorithm to obtain the second mechanical data.

[0063] The first mechanical anomaly feature extraction layer is used to extract features from the second mechanical data to obtain the first mechanical anomaly feature; the first mechanical anomaly feature includes the proportion of high-frequency energy of the first rotor. High-frequency energy ratio of the second spindle First principal axis runout residual First pressure pulsation interrelationship number and the standard deviation of the first guide vane opening fluctuation The high-frequency energy ratio of the first impeller and the high-frequency energy ratio of the second main shaft are obtained by extracting the high-frequency energy ratios of the impeller vibration signal data and the main shaft vibration signal data through wavelet packet decomposition, respectively. The process of obtaining the first main shaft swing residual is as follows: firstly, the vibration amplitude of the main shaft swing signal data is analyzed using the LSTM algorithm, and the vibration amplitude in the next hour is predicted to obtain the predicted vibration amplitude sequence; finally, the first main shaft swing residual is obtained based on the actual vibration amplitude sequence and the predicted vibration amplitude sequence. The process of obtaining the first pressure pulsation cross-correlation coefficient includes: firstly, bandpass filtering is performed on the impeller pressure pulsation signal data and the tailrace pipe pressure pulsation signal data, and a 1-second window sampling is performed based on a 1KHz sampling frequency to obtain the Person correlation coefficient of the impeller pressure pulsation signal data and the tailrace pipe pressure pulsation signal data within the window; and the first mechanical anomaly feature matrix is ​​obtained based on the first mechanical anomaly feature. .

[0064] The first mechanical anomaly feature analysis layer is used to obtain the first mechanical anomaly coefficient based on the first mechanical anomaly feature.

[0065] The first output layer is used to output the first mechanical anomaly feature matrix and the first mechanical anomaly coefficient.

[0066] The first mechanical anomaly feature matrix is: (1) In the formula, Represents the transpose of a matrix; The first mechanical anomaly coefficient is: (2) In the formula, Indicates the first mechanical anomaly coefficient; , , , and These represent the weights of five different mechanical anomaly features.

[0067] Step 2: Model Building A second thermal anomaly identification model is constructed, which includes a second input layer, a second preprocessing layer, a second thermal anomaly feature extraction layer, a second thermal anomaly feature analysis layer, and a second output layer.

[0068] The second input layer is used to input the second thermal data into the second thermal anomaly identification model.

[0069] The specific process of the second preprocessing layer includes: firstly, denoising and smoothing the second thermal data, and converting the pixel values ​​in each image of the second thermal data into actual temperatures, determining the monitoring points, and obtaining the temperature data of each monitoring point; and then standardizing the temperature data using the Z-Score algorithm to finally obtain the third thermal data.

[0070] The second thermal anomaly feature extraction layer is used to extract features from the third thermal data using a CNN, resulting in second thermal anomaly features. These second thermal anomaly features include the degree of anomaly in the stator winding temperature gradient. Rotor winding temperature fluctuation rate Temperature unevenness of thrust bearing pads Water pump bearing hot spot identification coefficient and the temperature anomaly coefficient of the heat exchanger metal wall ; The specific calculation formula is as follows: (3) In the formula, , , , These represent the stator windings, respectively. , , , Temperature values ​​at each monitoring point; Indicates the spatial distance between adjacent monitoring points; Indicates the number of monitoring points; The specific formula is as follows: (4) In the formula, This represents the standard deviation of the temperature at each monitoring point. This represents the average temperature at each monitoring point; The specific calculation formula is as follows: (5) In the formula, This indicates the highest temperature at each monitoring point; This indicates the lowest temperature at each monitoring point; This represents the average temperature at each monitoring point; The specific calculation formula is as follows: (6) In the formula, This indicates the number of pixels on the surface of the water pump bearing. Indicates the first surface of the water pump bearing Temperature data of each pixel; Indicates the hotspot threshold temperature; This indicates an indicator function that takes the value 1 when the condition is 1, and 0 otherwise. The specific calculation formula is as follows: (7) In the formula, Indicates the first The average temperature at each monitoring point in the region; Indicates the first Reference average temperature at each monitoring point in each region; This indicates the total number of divisions within the metal wall.

[0071] The second thermal anomaly feature analysis layer is used to obtain the second thermal anomaly coefficient and the second thermal anomaly feature matrix based on the second thermal anomaly features. .

[0072] The second output layer is used to output the second thermal anomaly feature matrix and the second thermal anomaly coefficient.

[0073] The second thermal anomaly feature matrix is: (8) The second thermal anomaly coefficient is: (9) In the formula, Indicates the second thermal anomaly coefficient; , , , and The influence weights of five different thermal anomaly characteristics are respectively listed; A third electrical anomaly identification model is constructed, which includes a third input layer, a third preprocessing layer, a third electrical anomaly feature extraction layer, a third electrical anomaly feature analysis layer, and a third output layer. The third input layer is used to input the third electrical data into the third electrical anomaly identification model; The third preprocessing layer is used to standardize the third electrical data using the Z-Score algorithm to obtain the fourth electrical data; The third electrical anomaly feature extraction layer is used to extract features from the fourth electrical data using machine learning algorithms; thus obtaining the third electrical anomaly feature; the third electrical anomaly feature includes the stator winding partial discharge intensity index. Transformer partial discharge energy accumulation rate GIS partial discharge frequency anomaly Overall insulation resistance attenuation rate and current anomaly Current anomaly By stator current harmonic distortion rate and excitation current fluctuation Comprehensive acquisition; The specific formula is as follows: (10) In the formula, This represents the time function of the partial discharge pulse amplitude. Indicates a time period; Indicates time; The specific formula is as follows: (11) in, Indicates the transformer number The energy of a partial discharge; Indicates the reference energy threshold; Indicates the number of partial discharges in the transformer; The specific formula is as follows: (12) In the formula, Indicates the first Partial discharge frequencies in each frequency band; Indicates the first Reference values ​​for partial discharge frequencies in each frequency band; Indicates the first Weighting coefficients for each frequency band; Indicates the number of frequency bands allocated; The specific formula is as follows: (13) in, , and This indicates the current insulation resistance value of the stator winding, main transformer, and GIS; , and This indicates the initial insulation resistance values ​​of the stator windings, main transformer, and GIS. , and This represents three different weighting coefficients; The specific formula is as follows: (14) In the formula, and Indicates the current anomaly weighting coefficient; The specific formula is as follows: (15) In the formula, Indicates the first RMS value of subharmonic current; Indicates the effective value of the fundamental current; Indicates the highest harmonic order; The specific calculation formula is as follows: (16) In the formula, This represents the excitation current data; This represents the maximum value in the excitation current data; This represents the minimum value in the excitation current data; This represents the average value in the excitation current data.

[0074] The third electrical anomaly feature analysis layer is used to obtain the third electrical anomaly feature matrix based on the third electrical anomaly features. and the third electrical anomaly coefficient; The third electrical anomaly feature matrix is ​​as follows: (17) The third electrical anomaly coefficient is: (18) In the formula, Indicates the third electrical anomaly coefficient; , , , and This indicates the weights of five different electrical anomaly characteristics. The third output layer is used to output the third electrical anomaly feature matrix and the third electrical anomaly coefficient.

[0075] A fourth coupling anomaly identification model is constructed, which includes a fourth input layer, a fourth coupling anomaly feature extraction layer, a fourth coupling anomaly feature analysis layer, and a fourth output layer. The fourth input layer is used to input the first mechanical anomaly feature matrix, the second thermal anomaly feature matrix, and the third electrical anomaly feature matrix into the fourth coupled anomaly identification model; The fourth coupling anomaly feature extraction layer is used to perform feature analysis on the first mechanical anomaly feature matrix, the second thermal anomaly feature matrix, and the third electrical anomaly feature matrix; to obtain binary anomaly feature coupling coefficients and ternary anomaly feature coupling coefficients; the binary anomaly feature coupling coefficients include mechanical-thermal anomaly coupling coefficients. Mechanical-electrical abnormal coupling coefficient and thermo-electrical anomaly coupling coefficient The coupling coefficient of the ternary anomaly characteristics is the mechanical-thermal-electrical anomaly coupling coefficient. The formula for calculating the mechanical-thermal anomaly coupling coefficient is as follows: (19) In the formula, The Person correlation function represents the first mechanical anomaly feature matrix and the second thermal anomaly feature matrix; The calculation formula is: (20) in, and These represent the mean values ​​of each feature in the first mechanical anomaly feature matrix and the mean values ​​of each feature in the second thermal anomaly feature matrix, respectively. The specific formula for calculating the mechanical-electrical abnormal coupling coefficient is as follows: (twenty one) in, The Person correlation function represents the first mechanical anomaly feature matrix and the third electrical anomaly feature matrix; The calculation formula is: (twenty two) In the formula, and These represent the mean values ​​of each feature in the first mechanical anomaly feature matrix and the mean values ​​of each feature in the third electrical anomaly feature matrix, respectively. The formula for calculating the thermo-electrical anomaly coupling coefficient is as follows: (twenty three) In the formula, The Person correlation function represents the second thermal anomaly characteristic matrix and the third electrical anomaly characteristic matrix; The calculation formula is: (twenty four) In the formula, and These represent the mean values ​​of each feature in the second thermal anomaly feature matrix and the mean values ​​of each feature in the third electrical anomaly feature matrix, respectively. The mechanical-thermal-electrical abnormal coupling coefficient The calculation formula is: (25) The fourth coupling anomaly feature analysis layer is used to obtain the fourth coupling anomaly coefficient based on the binary anomaly feature coupling coefficient and the ternary anomaly feature coupling coefficient. The fourth coupling anomaly coefficient for: (26) In the formula, and These represent the weighting coefficients of the binary anomaly feature coupling coefficient and the ternary anomaly feature coupling coefficient, respectively. The fourth output layer is used to output the fourth coupling anomaly coefficient.

[0076] The global anomaly coefficient is obtained based on the first mechanical anomaly coefficient, the second thermal anomaly coefficient, the third electrical anomaly coefficient, and the fourth coupling anomaly coefficient. The formula for calculating the global anomaly coefficient is: (27) In the formula, , , , These represent the influence weights of each anomaly coefficient; A first alarm confidence level is constructed based on the correlation between the global anomaly coefficient and the actual alarm situation. The first alarm confidence level is: (28) In the formula, Indicates the first level of confidence in the alarm; Indicates the global anomaly coefficient influence factor; This indicates the historical frequency of correct alarms based on the global anomaly coefficient. This represents the influence factor on the historical alarm correctness frequency based on the global anomaly coefficient. A three-tiered alarm suppression strategy is determined based on the first alarm confidence level and a preset threshold set; specifically, it includes first-level suppression, second-level suppression, and third-level suppression, as follows: Level 1 suppression: Low confidence alarm ( The alarm will be automatically suppressed and no notification will be triggered. Secondary suppression: Medium confidence alarm ( This will be pushed to the monitoring interface and marked as "Pending Confirmation"; Level 3 suppression: High confidence alarm ( Trigger an audible and visual alarm; , These represent the first preset confidence threshold and the second preset confidence threshold, respectively.

[0077] AR images are sent to monitoring personnel based on the global anomaly coefficient and the first alarm confidence level; monitoring personnel issue tasks based on the global anomaly coefficient, the first alarm confidence level, and the three-level alarm suppression strategy; and repair notices are sent to staff.

[0078] Example 3 In another preferred embodiment, based on embodiments 1 and 2 above, this embodiment provides an intelligent monitoring system for hydropower plants, used to implement an intelligent monitoring method for hydropower plants, as detailed below: The intelligent monitoring system for hydropower plants mainly consists of a data acquisition module, a model building module, an anomaly identification module, a coefficient calculation module, and an alarm processing module. The modules transmit and interact with each other through a data bus.

[0079] 1. Data Acquisition Module 1) First Mechanical Data Acquisition Submodule: It is responsible for communicating with sensors installed at the turbine runner, main shaft, guide vanes, and draft tube to collect data such as runner vibration signals, main shaft vibration signals, main shaft swing signals, runner pressure pulsation signals, draft tube pressure pulsation signals, and guide vane opening signals, and transmits these data to the data processing unit.

[0080] 2) Second thermal data acquisition submodule: The thermal imager is controlled to acquire thermal images of the generator stator, rotor windings, thrust bearing pads, and water pump bearing surfaces. At the same time, it reads temperature sensor data from various areas of the heat exchanger metal wall. The thermal images and temperature data are then integrated and transmitted to the data processing center.

[0081] 3) Third Electrical Data Acquisition Submodule: It connects to devices such as partial discharge detectors, insulation resistance testers, harmonic analyzers, and current sensors in the excitation system to collect electrical data such as partial discharge, insulation resistance, stator harmonic current, and excitation current of the stator winding, main transformer, and GIS, and transmits them to the data processing module.

[0082] 2. Model building module 1) First Mechanical Data Anomaly Identification Model Construction Submodule: (1) Following the construction steps of the first mechanical data anomaly identification model in the method embodiment, an input layer, a preprocessing layer, an anomaly feature extraction layer, an anomaly feature analysis layer and an output layer are constructed to realize the construction of the first mechanical data anomaly identification model.

[0083] (2) Sub-module for constructing the second thermal anomaly identification model: Based on the construction method of the second thermal anomaly identification model, an input layer, a preprocessing layer, an anomaly feature extraction layer, an anomaly feature analysis layer, and an output layer are constructed to complete the construction of the second thermal anomaly identification model.

[0084] (3) Third Electrical Anomaly Identification Model Construction Submodule: Following the construction process of the third electrical anomaly identification model, an input layer, a preprocessing layer, an anomaly feature extraction layer, an anomaly feature analysis layer, and an output layer are constructed to realize the construction of the third electrical anomaly identification model.

[0085] (4) Fourth Coupling Anomaly Identification Model Construction Submodule: Based on the construction requirements of the fourth coupled anomaly identification model, an input layer, a coupled anomaly feature extraction layer, a coupled anomaly feature analysis layer, and an output layer are constructed to complete the construction of the fourth coupled anomaly identification model.

[0086] 3. Anomaly Detection Module The first mechanical data anomaly identification model, the second thermal anomaly identification model, the third electrical anomaly identification model, and the fourth coupling anomaly identification model, which are constructed using the model building module, are used to identify anomalies in the data collected by the data acquisition module. The first mechanical anomaly feature matrix and the first mechanical anomaly coefficient, the second thermal anomaly feature matrix and the second thermal anomaly coefficient, the third electrical anomaly feature matrix and the third electrical anomaly coefficient, and the fourth coupling anomaly coefficient are output to the coefficient calculation module.

[0087] 4. Coefficient Calculation Module Receive the anomaly coefficients output by the anomaly identification module, and calculate the global anomaly coefficient according to the algorithm in the method embodiment; The first alarm confidence level is constructed based on the global anomaly coefficient, and then transmitted to the alarm processing module.

[0088] 5. Alarm processing module The three-level alarm suppression strategy is determined by comparing the first alarm confidence level with the preset threshold set. When an alarm is triggered, an AR image is sent to the monitoring personnel, allowing them to intuitively understand the abnormal situation of the equipment. At the same time, the monitoring personnel can issue tasks in the system. Send repair notices to staff to promptly inspect and repair any malfunctioning equipment, ensuring the normal operation of the hydropower plant's equipment.

[0089] Through the above method and system implementation examples, the intelligent monitoring function of hydropower plants has been realized, which can identify abnormal equipment conditions in a timely and accurate manner, and take corresponding alarm suppression strategies according to different alarm confidence levels, thereby improving the reliability and safety of hydropower plant equipment operation.

[0090] In the preferred embodiment, the first monitoring object in step 1 includes the turbine runner, main shaft, guide vanes, and draft tube, and the first mechanical data includes runner vibration signal, main shaft vibration signal, main shaft swing signal, runner pressure pulsation signal, draft tube pressure pulsation signal, and guide vane opening signal data. The above settings can comprehensively cover the key parameters of turbine operation, ensuring accurate monitoring of the unit's status. Through real-time collection and analysis of these data, potential faults can be detected in a timely manner, providing strong support for the stable operation of the turbine.

[0091] In a preferred embodiment, the second monitoring objects in step 1 include the generator stator, rotor windings, thrust bearing pads, and water pump. The second thermal data includes thermal images of the stator windings, rotor windings, thrust bearing pads, and water pump bearing surfaces, as well as thermal images of the heat exchanger metal walls after region division. This configuration enables comprehensive monitoring of the temperature of key components of the generator and water pump, and, combined with the thermal distribution of the heat exchanger metal walls, timely detection of potential overheating or abnormal wear problems, ensuring the stable operation of the power generation system and improving the accuracy of fault early warning.

[0092] In the preferred embodiment, the third monitoring object in step 1 includes the generator stator winding, main transformer, GIS switchgear, and excitation system. The third electrical data includes insulation status data (partial discharge quantity and insulation resistance of stator winding, main transformer, and GIS) and operating parameter data (stator harmonic current and excitation current). The above settings can comprehensively cover the key components of the generator set, ensuring accurate monitoring of the insulation performance and operating status of the generator stator winding, main transformer, GIS switchgear, and excitation system, timely detection of potential faults, and improvement of the stability and safety of the power system.

[0093] In the preferred embodiment, the first mechanical data anomaly identification model in step 2 includes an input layer, a preprocessing layer, an anomaly feature extraction layer, an anomaly feature analysis layer, and an output layer. The input layer inputs the first mechanical data, the preprocessing layer standardizes the data using the Z-Score algorithm to obtain the second mechanical data, the anomaly feature extraction layer extracts the first mechanical anomaly features and forms a matrix, the anomaly feature analysis layer obtains the first mechanical anomaly coefficients, and the output layer outputs the matrix and coefficients. This configuration effectively improves the accuracy and efficiency of anomaly identification. The data standardization in the preprocessing layer ensures that the model has consistent sensitivity to data of different magnitudes, while the anomaly feature extraction and analysis layer uses deep learning algorithms to deeply mine potential anomaly patterns in the data, providing strong support for mechanical fault early warning.

[0094] In the preferred embodiment, the first mechanical anomaly features include the high-frequency energy proportion of the runner and main shaft, the main shaft swing residual, the pressure pulsation cross-correlation coefficient, and the standard deviation of the guide vane opening fluctuation. The high-frequency energy proportion of the runner and main shaft is obtained by extracting the high-frequency energy proportion of the runner and main shaft vibration signal data through wavelet packet decomposition. The main shaft swing residual is obtained by analyzing the vibration amplitude of the main shaft swing signal data using the LSTM algorithm, predicting the vibration amplitude for the next hour, and combining it with the actual vibration amplitude sequence. The pressure pulsation cross-correlation coefficient is obtained by bandpass filtering the pressure pulsation signal data of the runner and tailrace pipe, and obtaining the Person correlation coefficient through 1-second window sampling. The standard deviation of the guide vane opening fluctuation is measured by statistically analyzing the standard deviation of the guide vane opening signal data to measure its fluctuation degree. The extraction and analysis of these mechanical anomaly features can comprehensively monitor the operating status of the hydro-generator unit, promptly detect potential faults, and provide strong support for the stable operation of the equipment.

[0095] In a preferred embodiment, the second thermal anomaly identification model in step 2 includes an input layer, a preprocessing layer, an anomaly feature extraction layer, an anomaly feature analysis layer, and an output layer. The input layer inputs the second thermal data; the preprocessing layer denoises, smooths, converts pixel values ​​to actual temperatures, determines monitoring points, and standardizes to obtain the third thermal data; the anomaly feature extraction layer uses a CNN to extract the second thermal anomaly features and form a matrix; the anomaly feature analysis layer obtains the second thermal anomaly coefficients; and the output layer outputs the matrix and coefficients. This configuration enables the second thermal anomaly identification model to efficiently identify anomaly features from complex thermal data, improving the accuracy and efficiency of anomaly detection and providing a reliable data foundation for subsequent fault warning and decision support.

[0096] In the preferred embodiment, the second thermal anomaly features include stator winding temperature gradient anomaly, rotor winding temperature fluctuation rate, thrust bearing pad temperature non-uniformity, water pump bearing hot spot identification coefficient, and heat exchanger metal wall temperature anomaly coefficient. These are calculated using corresponding formulas based on parameters such as monitoring point temperature, spatial distance, standard deviation, mean, maximum and minimum temperatures, pixel temperature, hot spot threshold, regional average temperature, and reference average temperature. These settings comprehensively and accurately reflect the thermal state of the generator set, promptly identifying potential overheating, wear, or uneven cooling issues, providing a strong basis for fault early warning and diagnosis, ensuring the safe and stable operation of the unit, and optimizing maintenance strategies to reduce the risk of unplanned shutdowns.

[0097] In a preferred embodiment, the third electrical anomaly identification model in step 2 includes an input layer, a preprocessing layer, an anomaly feature extraction layer, an anomaly feature analysis layer, and an output layer. The input layer inputs the third electrical data, the preprocessing layer uses the Z-Score algorithm to standardize the data to obtain the fourth electrical data, the anomaly feature extraction layer uses a machine learning algorithm to extract the third electrical anomaly features and form a matrix, the anomaly feature analysis layer obtains the third electrical anomaly coefficients, and the output layer outputs the matrix and coefficients. This configuration ensures that the model can efficiently and accurately identify the third type of anomaly in the electrical system. The Z-Score algorithm improves the consistency and comparability of the data, and the introduction of the machine learning algorithm enhances the model's ability to capture complex anomaly features, providing a strong guarantee for the stable operation of the electrical system.

[0098] In the preferred embodiment, the third electrical anomaly characteristic includes the stator winding partial discharge intensity index, transformer partial discharge energy accumulation rate, GIS partial discharge frequency anomaly degree, insulation resistance comprehensive attenuation rate, and current anomaly degree. These are calculated using corresponding formulas based on parameters such as the partial discharge pulse amplitude-time function, transformer partial discharge energy, partial discharge frequency, insulation resistance value, current harmonic distortion rate, and excitation current fluctuation rate. This configuration effectively monitors key anomaly indicators of electrical equipment, improving the accuracy and timeliness of fault warnings. Real-time data analysis allows for the timely detection of potential problems, providing a scientific basis for operation and maintenance decisions and ensuring the safe and stable operation of the power system.

[0099] In the preferred embodiment, the fourth coupling anomaly identification model in step 2 includes an input layer, a coupling anomaly feature extraction layer, a coupling anomaly feature analysis layer, and an output layer. The input layer inputs the first mechanical, second thermal, and third electrical anomaly feature matrices. The coupling anomaly feature extraction layer obtains binary (mechanical-thermal, mechanical-electrical, and thermal-electrical anomaly coupling coefficients) and ternary (mechanical-thermal-electrical anomaly coupling coefficients) anomaly feature coupling coefficients. The coupling anomaly feature analysis layer obtains the fourth coupling anomaly coefficient. The output layer outputs the fourth coupling anomaly coefficient. This configuration effectively improves the accuracy and efficiency of anomaly identification. Through multi-level feature coupling analysis, it can accurately capture complex anomaly correlations between mechanical, thermal, and electrical systems, providing strong data support for fault early warning and diagnosis.

[0100] In the preferred embodiment, the binary and ternary anomaly feature coupling coefficients are calculated using corresponding formulas based on the Person correlation function of the first mechanical, second thermal, and third electrical anomaly feature matrices and the mean value of each matrix feature; the fourth coupling anomaly coefficient is calculated based on the weighting coefficients of the binary and ternary anomaly feature coupling coefficients. The above settings aim to improve the accuracy and efficiency of fault diagnosis. By comprehensively considering the multi-dimensional anomaly features of mechanical, thermal, and electrical systems, combining the analysis of the correlation between features using the Person correlation function, and optimizing the coupling anomaly coefficients using weighting coefficients, accurate identification of fault modes can be achieved.

[0101] In the preferred embodiment, the three-level alarm suppression strategy in step 4 includes first-level suppression (automatic suppression of low-confidence alarms), second-level suppression (medium-confidence alarms are pushed to the monitoring interface and marked "pending confirmation"), and third-level suppression (high-confidence alarms trigger audible and visual alarms). These settings ensure effective hierarchical management of alarm information, reduce false alarm interference, and improve the response efficiency of maintenance personnel. Specifically, low-confidence alarms are automatically filtered, medium-confidence alarms require manual verification, and high-confidence alarms immediately attract attention, ensuring that emergency situations are handled rapidly.

[0102] In summary, the intelligent monitoring method and system for hydropower plants provided by this invention achieve a systemic breakthrough in addressing long-standing technical bottlenecks in the field of hydropower plant operation monitoring. This solution first resolves three core issues: insufficient multimodal data fusion, lack of coupling analysis, and imperfect alarm handling mechanisms. It successfully overcomes the limitations of existing technologies that rely solely on single-modal data, lack data interrelationship mining, have high false alarm rates, and cannot achieve closed-loop processing.

[0103] At the data acquisition and analysis level, this solution simultaneously acquires mechanical, thermal, and electrical data to construct a multimodal comprehensive analysis framework. Compared to traditional systems that rely on only a single type of data for detection, this multi-dimensional data acquisition model significantly improves the comprehensiveness and accuracy of anomaly detection. To further optimize feature capture accuracy, the solution designs three dedicated anomaly identification models: the first mechanical data model extracts high-frequency features using specific techniques and combines them with prediction algorithms to capture vibration anomalies; the second thermal model uses deep learning networks to identify temperature distribution anomalies; and the third electrical model uses machine learning algorithms to analyze discharge characteristics. These three models run in parallel, optimizing features for different monitoring objects, and compared to traditional single-model identification methods, they can more accurately capture anomaly features across different dimensions.

[0104] To reveal the interaction patterns of multiphysics fields, this solution introduces a fourth coupling anomaly identification model. By calculating the coupling coefficients of binary and ternary anomaly features, it quantitatively analyzes the correlation between mechanical, thermal, and electrical anomalies. This innovation overcomes the limitation of traditional methods that ignore the inherent connections between data, providing a deeper analytical perspective for anomaly detection. Based on this, the solution constructs a global anomaly coefficient, weighted and fused with the single-dimensional anomaly coefficients of mechanical, thermal, and electrical anomalies and the coupling coefficients to form a comprehensive risk index. This index provides monitoring personnel with an intuitive and comprehensive basis for anomaly assessment, solving the problems of scattered and difficult-to-quantify indicators in traditional systems.

[0105] In terms of alarm management, the solution proposes a three-tiered suppression strategy based on the global anomaly coefficient and alarm confidence level: low-confidence alarms are automatically suppressed, medium-confidence alarms are marked as "pending confirmation," and high-confidence alarms trigger audible and visual alarms. This hierarchical handling mechanism effectively reduces the false alarm rate while ensuring timely triggering of high-risk anomalies, effectively solving the industry dilemma of "alarm storms" and missed alarms. To further improve monitoring efficiency, the solution applies AR augmented reality technology to push a 3D visualization interface to monitoring personnel based on the global anomaly coefficient and alarm confidence level, overlaying key equipment status parameters in real time. Practical verification has shown that this technology significantly improves anomaly location efficiency and greatly shortens fault diagnosis time.

[0106] At the feature engineering level, the solution features a meticulously designed anomaly identification model for each anomaly detection model: the mechanical model achieves high-precision prediction through a spindle runout prediction residual algorithm; the thermal model develops temperature non-uniformity detection technology to enhance the ability to identify minute anomalies; and the electrical model constructs a comprehensive insulation status assessment model to achieve early warning of aging trends. These core technologies significantly improve early fault identification capabilities, representing a substantial improvement over the industry average, and significantly enhancing the sensitivity and specificity of anomaly detection. The entire system forms a complete technical closed loop of "data fusion - feature extraction - coupled analysis - risk quantification - AR presentation," setting a new technical benchmark for intelligent monitoring in the hydropower industry.

Claims

1. A method for implementing intelligent monitoring panels in hydropower plants, characterized in that, Includes the following steps: Step 1: Data acquisition, acquiring the first mechanical data of the first monitoring object of the hydropower plant, the second thermal data of the second monitoring object, and the third electrical data of the third monitoring object; Step 2: Model building, constructing the first mechanical data anomaly identification model, the second thermal anomaly identification model, the third electrical anomaly identification model, and the fourth coupling anomaly identification model; Step 3: Anomaly identification. Anomalies are identified using each model to obtain the first mechanical anomaly feature matrix and coefficients, the second thermal anomaly feature matrix and coefficients, the third electrical anomaly feature matrix and coefficients, and the fourth coupling anomaly coefficients. Step 4: Coefficient Calculation and Alarm. Calculate the global anomaly coefficient based on each anomaly coefficient, construct the first alarm confidence level, determine the three-level alarm suppression strategy, send AR images to monitoring personnel and issue tasks, and send repair notices to staff.

2. The method for implementing intelligent monitoring panels in hydropower plants according to claim 1, characterized in that: The first monitoring object in step 1 includes the turbine runner, main shaft, guide vanes and tailrace pipe, and the first mechanical data includes runner vibration signal, main shaft vibration signal, main shaft swing signal, runner pressure pulsation signal, tailrace pipe pressure pulsation signal and guide vane opening signal data.

3. The method for implementing intelligent monitoring panels in hydropower plants according to claim 2, characterized in that: The second monitoring objects in step 1 include the generator stator, rotor winding, thrust bearing pads, and water pump. The second thermal data includes thermal images of the stator winding, rotor winding, thrust bearing pads, water pump bearing surfaces, and thermal images of the heat exchanger metal walls after dividing the area.

4. The method for implementing intelligent monitoring panels in hydropower plants according to claim 3, characterized in that: The third monitoring object in step 1 includes the generator stator winding, main transformer, GIS switchgear and excitation system, and the third electrical data includes insulation status data and operating parameter data.

5. The method for implementing intelligent monitoring panels in a hydropower plant according to claim 4, characterized in that: The first mechanical data anomaly identification model in step 2 includes an input layer, a preprocessing layer, an anomaly feature extraction layer, an anomaly feature analysis layer, and an output layer. The input layer inputs the first mechanical data, the preprocessing layer uses the Z-Score algorithm to standardize the second mechanical data, the anomaly feature extraction layer extracts the first mechanical anomaly features and forms a matrix, the anomaly feature analysis layer obtains the first mechanical anomaly coefficients, and the output layer outputs the matrix and coefficients.

6. The method for implementing intelligent monitoring panels in a hydropower plant according to claim 5, characterized in that: The first mechanical anomaly features include the high-frequency energy ratio of the impeller and main shaft, the main shaft swing residual, the pressure pulsation cross-correlation coefficient, and the standard deviation of the guide vane opening fluctuation. The high-frequency energy ratio of the impeller and main shaft is obtained by extracting the high-frequency energy ratio of the impeller and main shaft vibration signal data through wavelet packet decomposition. The main shaft swing residual is obtained by analyzing the vibration amplitude of the main shaft swing signal data using the LSTM algorithm, predicting the vibration amplitude for the next hour, and combining it with the actual vibration amplitude sequence. The pressure pulsation cross-correlation coefficient is obtained by bandpass filtering and 1-second window sampling of the pressure pulsation signal data of the impeller and tailrace pipe to obtain the Person correlation coefficient.

7. The method for implementing intelligent monitoring panels in a hydropower plant according to claim 6, characterized in that: The second thermal anomaly identification model in step 2 includes an input layer, a preprocessing layer, an anomaly feature extraction layer, an anomaly feature analysis layer, and an output layer. The input layer inputs the second thermal data, the preprocessing layer denoises, smooths, converts pixel values ​​to actual temperatures, determines monitoring points, and standardizes to obtain the third thermal data, the anomaly feature extraction layer uses CNN to extract the second thermal anomaly features and form a matrix, the anomaly feature analysis layer obtains the second thermal anomaly coefficients, and the output layer outputs the matrix and coefficients.

8. The method for implementing intelligent monitoring panels in hydropower plants according to claim 7, characterized in that: The second thermal anomaly features include stator winding temperature gradient anomaly, rotor winding temperature fluctuation rate, thrust bearing pad temperature non-uniformity, water pump bearing hot spot identification coefficient, and heat exchanger metal wall temperature anomaly coefficient; these are calculated using corresponding formulas based on parameters such as monitoring point temperature, spatial distance, standard deviation, mean, highest and lowest temperature, pixel temperature, hot spot threshold, regional average temperature, and reference average temperature.

9. The method for implementing intelligent monitoring panels in a hydropower plant according to claim 8, characterized in that: The third electrical anomaly identification model in step 2 includes an input layer, a preprocessing layer, an anomaly feature extraction layer, an anomaly feature analysis layer, and an output layer. The input layer inputs the third electrical data, the preprocessing layer uses the Z-Score algorithm to standardize the data to obtain the fourth electrical data, the anomaly feature extraction layer uses a machine learning algorithm to extract the third electrical anomaly features and form a matrix, the anomaly feature analysis layer obtains the third electrical anomaly coefficients, and the output layer outputs the matrix and coefficients.

10. The method for implementing intelligent monitoring panels in a hydropower plant according to claim 9, characterized in that: The third electrical anomaly features include the stator winding partial discharge intensity index, transformer partial discharge energy accumulation rate, GIS partial discharge frequency anomaly degree, insulation resistance comprehensive attenuation rate, and current anomaly degree; these are calculated using corresponding formulas based on parameters such as the partial discharge pulse amplitude time function, transformer partial discharge energy, partial discharge frequency, insulation resistance value, current harmonic distortion rate, and excitation current fluctuation rate.

11. The method for implementing intelligent monitoring panels in a hydropower plant according to claim 1, characterized in that: The fourth coupling anomaly identification model in step 2 includes an input layer, a coupling anomaly feature extraction layer, a coupling anomaly feature analysis layer, and an output layer. The input layer inputs the first mechanical, second thermal, and third electrical anomaly feature matrices. The coupling anomaly feature extraction layer obtains the binary and ternary anomaly feature coupling coefficients. The coupling anomaly feature analysis layer obtains the fourth coupling anomaly coefficient. The output layer outputs the fourth coupling anomaly coefficient.

12. The method for implementing intelligent monitoring panels in a hydropower plant according to claim 11, characterized in that: The coupling coefficients of the binary and ternary anomaly features are calculated using corresponding formulas based on the Person correlation function of the first mechanical, second thermal, and third electrical anomaly feature matrices and the mean value of each matrix feature. The fourth coupling anomaly coefficient is calculated based on the weighting coefficients of the binary and ternary anomaly feature coupling coefficients.

13. The method for implementing intelligent monitoring panels in a hydropower plant according to claim 12, characterized in that: The three-level alarm suppression strategy in step 4 includes first-level suppression, second-level suppression, and third-level suppression.

14. A hydropower plant intelligent monitoring system, which is a system for implementing the hydropower plant intelligent monitoring method as described in claim 13, characterized in that, include: Data acquisition module: used to acquire the first mechanical data of the first monitoring object of the hydropower plant, the second thermal data of the second monitoring object, and the third electrical data of the third monitoring object; Model building module: used to build the first mechanical data anomaly identification model, the second thermal anomaly identification model, the third electrical anomaly identification model, and the fourth coupling anomaly identification model; Anomaly identification module: Each model is used to identify anomalies, and the first mechanical anomaly feature matrix and coefficients, the second thermal anomaly feature matrix and coefficients, the third electrical anomaly feature matrix and coefficients, and the fourth coupling anomaly coefficients are obtained. Coefficient calculation module: Calculates the global anomaly coefficient based on each anomaly coefficient, and constructs the first alarm confidence level; Alarm processing module: Determines a three-level alarm suppression strategy based on the first alarm confidence level and the preset threshold set, sends AR images to monitoring personnel, the monitoring personnel issue tasks, and sends repair notices to staff.

Images