Multi-parameter fusion-based wind turbine fault self-diagnosis and early warning method and system

Abstract

The application discloses a wind turbine fault self-diagnosis and early warning method and system based on multi-parameter fusion, relates to the technical field of wind power generation equipment fault monitoring and early warning, and comprises the following steps: collecting wind turbine monitoring data, constructing a physical information three-stage cleaning mechanism, constructing a multi-view adaptive space-time feature map, extracting wind turbine feature data, adopting a deep embedded clustering algorithm and a source domain-target domain migration adaptation technology, outputting the working condition label of the wind turbine, reversely adjusting the multi-view adaptive space-time feature map, extracting the latest wind turbine feature data, outputting deep fusion features by using an adaptive polynomial KAN network and a meta-learning module, constructing a wind turbine fault self-diagnosis model, outputting a fault self-diagnosis coefficient, evaluating the fault grade of the wind turbine by combining a double-stage calibration evaluation method, and outputting a wind turbine fault early warning report, so that the wind turbine fault self-diagnosis and early warning are realized.

Description

Technical Field

[0001] This invention relates to the field of wind power equipment fault monitoring and early warning technology, specifically to a method and system for self-diagnosis and early warning of wind turbine generator faults based on multi-parameter fusion. Background Technology

[0002] Currently, most wind turbine condition monitoring systems still rely primarily on single-parameter alarm mechanisms based on fixed thresholds and offline diagnostic tools such as signal spectrum analysis. These traditional technologies are no longer sufficient to meet the needs for early and accurate early warning under complex operating conditions. On the one hand, single-parameter models have weak anti-interference capabilities and are prone to frequent false alarms due to sudden environmental changes or fluctuations in operating conditions, and they cannot characterize multi-component coupled faults. On the other hand, feature extraction relies on human experience, making it difficult to automatically mine effective fault precursor information from massive, high-dimensional operating data. In addition, existing methods generally lack transfer and adaptation capabilities, resulting in limited generalization performance. With the continuous growth of wind turbine installed capacity and its trend towards larger and more complex structures, developing a fault self-diagnosis system that can deeply integrate multi-source heterogeneous data, achieve early intelligent early warning, and possess adaptive learning capabilities is of decisive significance for improving the operational reliability of wind turbines and promoting the transformation of operation and maintenance modes from periodic maintenance to predictive maintenance.

[0003] Traditional wind turbine condition monitoring and fault diagnosis technologies mainly rely on single-parameter early warnings based on fixed thresholds or single-model outputs. While these technologies can identify some typical faults, they suffer from insufficient accuracy and reliability in practical applications. Single parameters are easily affected by operating conditions and the environment, leading to frequent false alarms and missed alarms. Furthermore, they have poor generalization and adaptive capabilities in fault diagnosis and early warning. Therefore, how to achieve intelligent fault self-diagnosis and early warning capabilities that adaptively integrate multi-source heterogeneous data, accurately capture fault characteristics, and have cross-condition migration capabilities has become a key technical challenge for improving the operation and maintenance efficiency of wind turbines and ensuring their safe and stable operation. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for self-diagnosis and early warning of wind turbine generator faults based on multi-parameter fusion, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: Firstly, a method for self-diagnosis and early warning of wind turbine generator faults based on multi-parameter fusion, comprising the following steps: Step 1: Collect monitoring data of wind turbine generator sets and construct a three-level physical information cleaning mechanism to efficiently preprocess the collected data, effectively remove noise and outliers, and improve the reliability of the collected data. The monitoring data of wind turbine generator sets includes operation data, component data, environmental data, and topology data. Step 2: Based on operational data, component data, and topology data, construct a multi-view adaptive spatiotemporal feature map to fully explore the spatiotemporal correlation characteristics of the data and extract the characteristic data of the wind turbine generator set; Step 3: Using deep embedded clustering algorithms and source-target domain migration adaptation technology, accurate clustering and migration adaptation across operating conditions are achieved through wind turbine generator feature data, operation data and environmental data, and the operating condition labels of wind turbine generators are output. Step 4: Based on the operating condition labels, adjust the multi-view adaptive spatiotemporal feature map and extract the latest wind turbine generator feature data. Combine the adaptive multinomial KAN network and meta-learning module to deeply fuse the latest wind turbine generator feature data, operating data and component data, significantly enhancing the adaptability and generalization ability of feature expression, and outputting deeply fused features. Step 5: Using a neural network algorithm and based on deep fusion features, construct a fault self-diagnosis model for wind turbine generator sets and output fault self-diagnosis coefficients; Step 6: Using a two-stage calibration evaluation method, based on the fault self-diagnosis coefficient, the fault level of the wind turbine generator is evaluated, which greatly improves the robustness and accuracy of fault level determination. Combined with digital twin visualization technology, a fault early warning report of the wind turbine generator is output.

[0006] A further improvement to the technical solution of this invention lies in the fact that, in step one, the process of collecting monitoring data from wind turbine generators and constructing a three-level physical information cleaning mechanism includes: Different types of data acquisition equipment are deployed to collect monitoring data from wind turbine generator sets. These acquisition devices include power transmitters, speed sensors, current transformers, voltage transformers, grid frequency transmitters, vibration acceleration sensors, magnetoelectric velocity sensors, fiber optic grating sensors, patch temperature sensors, platinum resistance temperature sensors, online oil viscosity sensors, ultrasonic anemometers, ultrasonic wind vanes, atmospheric temperature sensors, humidity sensors, atmospheric pressure sensors, laser alignment instruments, 3D laser scanners, wire-type displacement sensors, and articulated arm measuring machines. The operating data includes the wind turbine generator's power, speed, pitch angle, three-phase current, three-phase voltage, power, speed, and the grid frequency at the connection point; The component data includes the vibration acceleration, vibration velocity, vibration displacement and temperature of the wind turbine blades, hub, main shaft, gearbox, generator, converter and transformer, as well as the lubricating oil temperature and viscosity of the wind turbine main shaft, gearbox and generator. Environmental data includes wind speed, wind direction, temperature, humidity, air pressure, wind shear index, turbulence intensity, atmospheric stability, and air density of the environment in which the wind turbine is located; The topology data represents the physical topology of the wind turbine drivetrain. Specifically, the power, speed, pitch angle, three-phase current, three-phase voltage, power, speed and grid frequency of the wind turbine generator set are collected through the power transmitter, speed sensor, absolute encoder of the pitch motor, current transformer, voltage transformer and grid frequency transmitter. Among them, the pitch motor is one of the key components of the wind turbine generator set. Vibration acceleration sensors, magnetoelectric velocity sensors, fiber optic grating sensors, and patch temperature sensors are used to collect vibration acceleration, vibration velocity, vibration displacement, and temperature of wind turbine blades, hubs, main shafts, gearboxes, generators, converters, and transformers. The lubricating oil temperature and viscosity of the main shaft, gearbox, and generator are collected through platinum resistance temperature sensors and online oil viscosity sensors. Combining ultrasonic anemometers, ultrasonic wind vanes, atmospheric temperature sensors, humidity sensors, and atmospheric pressure sensors, the wind speed, wind direction, temperature, humidity, and air pressure of the wind turbine's environment are collected. Using the temperature, humidity, and air pressure of the wind turbine's environment, referencing the formula... Calculate the air density of the environment in which the wind turbine is located, where, The air density of the environment in which the wind turbine is located. , and The data includes the temperature, humidity, and air pressure of the environment where the wind turbine is located; two ultrasonic anemometers are installed to collect wind speeds at two different heights of the wind turbine, referencing the formula... Calculate the wind shear index of the environment where the wind turbine is located, where... and They are respectively at the height of and The wind speed in the environment where the wind turbine is located. The wind shear index is used. The standard deviation and average wind speed formulas are used to calculate the standard deviation and average wind speed of the environment where the wind turbine is located. The turbulence intensity of the environment is calculated by the ratio of the standard deviation to the average wind speed. Two layers of atmospheric temperature sensors are installed to collect atmospheric temperatures at two different altitudes where wind turbines are located. The calculation is based on the formula... Calculate the atmospheric stability of the environment in which the wind turbine is located, where, The atmospheric stability of the environment in which the wind turbine is located. and They are respectively at the height of and The formula is based on the principle of temperature gradient calculation, which measures the atmospheric temperature of the environment in which the wind turbine is located. By combining a laser alignment instrument, a 3D laser scanner, a wire-type displacement sensor, an articulated arm measuring machine, and a vibration displacement sensor, the physical topology of the wind turbine drive chain is collected. A three-level physical information cleaning mechanism is constructed. Based on the wind turbine theoretical control curve, physical constraint boundaries are constructed. By comparing the collected operation data, environmental data, and topology data, abnormal data is automatically corrected and removed, realizing the first level of physical information cleaning, ensuring that the data conforms to physical laws, and eliminating invalid interference data from the source. An adaptive wavelet threshold denoising algorithm is adopted to dynamically adjust the denoising parameters and remove noise from the data after the first-level cleaning of physical information and the component data, thereby achieving the second-level cleaning of physical information. Based on an event-triggered asynchronous data fusion and alignment algorithm, high-precision spatiotemporal synchronization processing is performed on wind turbine monitoring data to unify the analysis benchmark and achieve three-level cleaning of physical information.

[0007] A further improvement to the technical solution of this invention lies in the fact that, in step two, the process of constructing a multi-view adaptive spatiotemporal feature map based on operational data, component data, and topology data, and extracting the feature data of the wind turbine generator set, includes: The characteristic data of wind turbine generator sets include time-domain characteristic data, frequency-domain characteristic data, and time-frequency-domain characteristic data; Each component of the wind turbine generator is set as a graph node. Based on the physical topology of the wind turbine generator drive chain, the connection weights between each graph node are initialized, and a static physical adjacency matrix is ​​constructed. The power, speed and pitch angle of the wind turbine generator set are extracted from the operating data. The vibration acceleration, vibration velocity, vibration displacement and temperature of the blades, hub, main shaft, gearbox, generator, converter and transformer of the wind turbine generator are extracted from the component data. The lubricating oil temperature of the main shaft, gearbox and generator of the wind turbine generator is also extracted. A multi-head attention mechanism is introduced, which inputs data extracted from running data and component data in real time, identifies abnormal graph nodes and their related nodes, automatically learns the relationship between graph nodes through graph network, dynamically adjusts the connection weight between abnormal graph nodes and their related nodes by referring to the connection weight between each graph node after initialization, and converts the real-time input data into the node attributes of each graph node, so as to realize the construction of multi-view adaptive spatiotemporal feature map. Using multi-view adaptive spatiotemporal feature maps as the carrier, graph convolution and causal convolution are performed on the node attributes of each graph node to obtain spatial topological features and time series features, respectively. Finally, a three-dimensional feature tensor of time-space-frequency is output. Combined with Fourier transform, time-domain feature data, frequency-domain feature data and time-frequency-domain feature data are obtained.

[0008] A further improvement to the technical solution of this invention lies in the fact that, in step three, the process of outputting the operating condition label of the wind turbine generator set using deep embedded clustering algorithm and source-target domain migration adaptation technology, based on the wind turbine generator set feature data, operating data, and environmental data, includes: The operating condition labels for wind turbine generators include low wind speed shear, strong turbulence, high temperature and humidity, grid frequency fluctuation, low air density, and atmospheric instability. The three-phase current, three-phase voltage, power, speed, and grid frequency at the grid connection point of the wind turbine generator set are selected from the operating data. The selected data, environmental data, and wind turbine generator set characteristic data are standardized. Using the feature dimension unification technology, the standardized data are spliced ​​together according to the time dimension to construct a standardized feature matrix. An autoencoder is used to reduce the dimensionality of the standardized feature matrix, retaining the core features for distinguishing working conditions. A deep embedded clustering algorithm is used to perform K-means clustering on the dimensionality-reduced features, automatically dividing the initial working condition clusters and outputting the initial working condition labels. Historical data of wind turbine generators are obtained from the wind turbine generator data acquisition and monitoring control system. The historical data of wind turbine generators is standardized and a standardized historical feature matrix is ​​constructed. The historical data of wind turbine generators includes historical three-phase current, historical three-phase voltage, historical power, historical speed, grid frequency of historical grid connection point, historical environmental data, and historical wind turbine generator characteristic data. Using the standardized feature matrix and the standardized historical feature matrix as the target domain data and source domain data respectively, calculate the maximum mean difference between the feature distributions of the source domain and the target domain. Set a threshold for the difference in feature distribution between the source and target domains, compare the maximum mean difference between the feature distributions of the source and target domains, adjust the migration adaptation parameters, and output the operating condition label of the wind turbine generator set.

[0009] A further improvement to the technical solution of this invention lies in the fact that, in step four, the process of adjusting the multi-view adaptive spatiotemporal feature map based on the operating condition label and extracting the latest wind turbine generator feature data includes: The operating condition labels of wind turbine generators are encoded. Based on the priority of the impact of different operating conditions on wind turbine generators, basic weight coefficients are assigned to the coding of each type of operating condition label to form an operating condition label weight vector. The correlation matrix between the operating condition label coding of wind turbine generators and the nodes of the multi-view adaptive spatiotemporal feature map is established. Read the current multi-view adaptive spatiotemporal feature map and dynamically adjust the connection weights between nodes of the multi-view adaptive spatiotemporal feature map, driven by the correlation matrix and the working condition label weight vector. Based on the connection weights between nodes in the dynamically adjusted multi-view adaptive spatiotemporal feature map, the multi-view adaptive spatiotemporal feature map is reconstructed. The latest wind turbine generator feature data is extracted by combining the graph convolution and causal convolution of the node attributes of each graph node with Fourier transform.

[0010] A further improvement to the technical solution of this invention lies in the fact that, in step four, the process of deeply fusing the latest wind turbine generator feature data, operational data, and component data by combining the adaptive multinomial KAN network and the meta-learning module to output deeply fused features includes: Select and integrate the three-phase current, three-phase voltage, and power of the wind turbine generator set from the operational data, as well as the lubricating oil viscosity of the wind turbine generator's main shaft, gearbox, and generator from the component data, to obtain basic data for deep fusion features; The deep fusion feature base data and the latest wind turbine generator feature data are normalized; a multi-channel one-dimensional convolutional neural network is used to process the normalized data respectively, extract the deep features of each mode, and generate a joint fault diagnosis feature vector through feature fusion. The joint fault diagnosis feature vector is input into the adaptive multinomial KAN network. The deep nonlinear coupling feature is fitted using a multinomial basis function. The order of the adaptive multinomial is dynamically adjusted according to the operating condition label of the wind turbine generator, the weight of the adaptive multinomial KAN network is optimized, and the deep nonlinear coupling feature is output. Using deep nonlinear coupling features as the training set, a meta-learning module is constructed to output meta-optimized features. The joint fault diagnosis feature vector, deep nonlinear coupling features, and meta-optimized features are aggregated. A cross-modal attention mechanism is introduced to automatically assign weights to each modality feature. Through weighted integration, the joint fault diagnosis feature vector, deep nonlinear coupling features, and meta-optimized features are fused. First, the fused joint fault diagnosis feature vector, deep nonlinear coupling feature, and meta-optimization feature are linearly compressed, and then normalized to obtain the deep fused feature.

[0011] A further improvement to the technical solution of this invention lies in the fact that, in step five, the process of constructing a wind turbine generator fault self-diagnosis model based on deep fusion features using a neural network algorithm and outputting fault self-diagnosis coefficients includes: The deep fusion features are divided into training and test sets in a 7:3 ratio. A neural network algorithm is used, with the training set data as input data and the fault self-diagnosis coefficient as output data. The nonlinear relationship between the deep fusion features and the fault self-diagnosis coefficient is learned to obtain the trained model. The test set data is input into the trained model, the model parameters are adjusted, the model performance is optimized, and the final wind turbine generator fault self-diagnosis model is obtained. Combined with the current deep fusion features, the fault self-diagnosis coefficient is output.

[0012] A further improvement to the technical solution of this invention lies in the fact that, in step six, the process of evaluating the fault level of the wind turbine generator set using a two-stage calibration evaluation method based on the fault self-diagnosis coefficient includes: Wind turbine generator set fault levels are classified into Level I, Level II, Level III, and Level IV faults. A Level I fault indicates that the fault does not affect the normal operation of the wind turbine generator set; a Level II fault indicates that the fault will cause minor abnormalities during operation and generally does not require intervention; a Level III fault indicates that the fault will cause general malfunctions during operation and requires intervention; and a Level IV fault indicates that the fault will cause serious malfunctions during operation, potentially even leading to immediate shutdown, and requires intervention. A two-stage calibration evaluation method is adopted. Based on the wind turbine generator fault self-diagnosis model and combined with the current deep fusion characteristics, the fault self-diagnosis coefficient of the past 30 days is output. The mean and standard deviation of the fault self-diagnosis coefficient of the past 30 days are calculated according to the mean formula and standard deviation formula. Based on the mean and standard deviation of the fault self-diagnosis coefficient over the past 30 days, the boundary thresholds between Level I and Level II faults, Level II and Level III faults, and Level III and Level IV faults were calculated. By using the boundary thresholds between Level I and Level II faults, Level II and Level III faults, and Level III and Level IV faults, a first-order assessment of the fault level of wind turbine generator sets is achieved, that is, a coarse classification of the fault level of wind turbine generator sets is realized. The fault self-diagnosis coefficients for the past 30 days are retrieved for 5 consecutive days. The reliability weights are assigned to the retrieved fault self-diagnosis coefficients for 5 consecutive days, the fault self-diagnosis weighted value is calculated, and then the deviation rate between the current fault self-diagnosis coefficient and the weighted confidence level is calculated. Based on the deviation rate between the current fault self-diagnosis coefficient and the weighted confidence level, a second-order assessment of the fault level of the wind turbine generator is achieved, that is, the accurate classification of the fault level of the wind turbine generator is realized, and the final fault level of the wind turbine generator is obtained.

[0013] A further improvement to the technical solution of this invention lies in the fact that, in step six, the process of outputting a wind turbine generator fault early warning report using digital twin visualization technology includes: Build a digital twin 3D model to achieve synchronous mapping of wind turbine monitoring data, fault self-diagnosis coefficients, and the final fault level of the wind turbine. In the digital twin 3D model, based on the final fault level of the wind turbine generator, the corresponding early warning indicator signal is presented intuitively, and the wind turbine generator fault early warning report is output, realizing the hierarchical and visual early warning display of the fault level of the wind turbine generator. The specific process of displaying tiered, visualized early warnings includes: U1 and Level I faults: Green indicator displayed, no warning indicator; U2 and II level faults: A yellow flashing indicator is displayed, triggering a fault self-diagnosis coefficient fluctuation pop-up window; U3 and Level III faults: An orange highlighted indicator will be displayed, and a pop-up window will be shown to display the current monitoring data changes of the wind turbine generator set. U4 and IV level faults: display a red warning sign and audible and visual prompts, trigger a fault self-diagnosis coefficient fluctuation pop-up window, and simultaneously display a pop-up window showing the current wind turbine generator monitoring data changes.

[0014] Secondly, a wind turbine generator fault self-diagnosis and early warning system based on multi-parameter fusion is used to realize the above-mentioned wind turbine generator fault self-diagnosis and early warning method based on multi-parameter fusion. It includes a monitoring data acquisition module, a feature extraction module, an operating condition division module, a reverse optimization and deep fusion module, a fault self-diagnosis module and a visualization module, wherein each module is interconnected. The monitoring data acquisition module collects monitoring data from wind turbine generators and constructs a three-level physical information cleaning mechanism to preprocess the collected data. The monitoring data from wind turbine generators includes operational data, component data, environmental data, and topology data. The feature extraction module, based on operational data, component data, and topology data, constructs a multi-view adaptive spatiotemporal feature map, breaking through the limitations of single feature extraction and extracting feature data of wind turbine generator sets. The operating condition classification module adopts a deep embedded clustering algorithm and source domain-target domain migration adaptation technology. It outputs the operating condition labels of wind turbine generators through wind turbine generator characteristic data, operation data and environmental data, so as to realize high-precision automatic classification under complex operating conditions. The reverse optimization and deep fusion module adjusts the multi-view adaptive spatiotemporal feature map based on the working condition label and extracts the latest wind turbine feature data. Combined with the adaptive multinomial KAN network and meta-learning module, the latest wind turbine feature data, operating data and component data are deeply fused to output deep fused features. The fault self-diagnosis module uses a neural network algorithm to construct a fault self-diagnosis model for wind turbine generators based on deep fusion features and outputs fault self-diagnosis coefficients. The visualization module adopts a two-stage calibration evaluation method, which assesses the fault level of wind turbine generators based on the fault self-diagnosis coefficient. It overcomes the problem of high false alarm rate of traditional threshold method, and outputs wind turbine generator fault early warning report by combining digital twin visualization technology.

[0015] The beneficial effects of this invention are as follows: Compared with traditional wind turbine generator fault self-diagnosis and early warning technologies, the method and system for wind turbine generator fault self-diagnosis and early warning based on multi-parameter fusion in this invention closely integrates the physical information three-level cleaning technology, multi-view adaptive spatiotemporal feature map construction technology, source domain-target domain migration adaptation technology, adaptive multinomial KAN network deep fusion technology, two-order calibration evaluation method, and digital twin visualization technology. This achieves fault self-diagnosis and early warning for wind turbine generators, solving the problems of frequent false alarms and missed alarms in traditional single-parameter early warning, reliance on manual experience for feature extraction, poor generalization and adaptive ability of evaluation results, and rigid and inaccurate fault level evaluation. This ensures that the method in this invention can refine the dynamic monitoring standards of the wind turbine generator fault self-diagnosis and early warning method and system based on multi-parameter fusion within a more precise range, making the monitored data more accurate indicators under the same conditions. The development and application of this method significantly enhances the intelligence level in the process of wind turbine generator fault self-diagnosis and early warning based on multi-parameter fusion. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a flowchart of the wind turbine generator fault self-diagnosis and early warning method based on multi-parameter fusion according to the present invention. Figure 2 This is a block diagram of the wind turbine generator fault self-diagnosis and early warning system based on multi-parameter fusion of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1, as Figure 1As shown, this invention provides a method for self-diagnosis and early warning of wind turbine generator faults based on multi-parameter fusion, which consists of the following steps: S1. Collect monitoring data of wind turbine generator sets, construct a three-level physical information cleaning mechanism, preprocess the collected data, and lay a high-quality data foundation for subsequent analysis. The monitoring data of wind turbine generator sets includes operation data, component data, environmental data and topology data. S2. Based on operational data, component data, and topology data, construct a multi-view adaptive spatiotemporal feature map and extract feature data of the wind turbine generator set; S3. Employing a deep embedded clustering algorithm and source-target domain migration adaptation technology, the system outputs the operating condition label of the wind turbine generator by jointly analyzing the characteristic data, operation data, and environmental data of the wind turbine generator. S4. Based on the operating condition label, adjust the multi-view adaptive spatiotemporal feature map and extract the latest wind turbine generator feature data. Combine the adaptive multinomial KAN network and meta-learning module to deeply fuse the latest wind turbine generator feature data, operating data and component data, and output the deep fusion feature. S5. Using a neural network algorithm and based on deep fusion features, a self-diagnosis model for wind turbine generator faults is constructed, and fault self-diagnosis coefficients are output, achieving accurate mapping from features to faults. S6. The two-stage calibration evaluation method is adopted. Based on the fault self-diagnosis coefficient, the fault level of the wind turbine generator is evaluated, which greatly enhances the reliability of fault classification. Combined with digital twin visualization technology, the fault early warning report of the wind turbine generator is output.

[0020] In S1, the process of collecting monitoring data from wind turbine generators and constructing a three-level physical information cleaning mechanism includes: Different types of data acquisition equipment are deployed to collect monitoring data from wind turbine generator sets. These acquisition devices include power transmitters, speed sensors, current transformers, voltage transformers, grid frequency transmitters, vibration acceleration sensors, magnetoelectric velocity sensors, fiber optic grating sensors, patch temperature sensors, platinum resistance temperature sensors, online oil viscosity sensors, ultrasonic anemometers, ultrasonic wind vanes, atmospheric temperature sensors, humidity sensors, atmospheric pressure sensors, laser alignment instruments, 3D laser scanners, wire-type displacement sensors, and articulated arm measuring machines. The operating data includes the wind turbine generator's power, speed, pitch angle, three-phase current, three-phase voltage, power, speed, and the grid frequency at the connection point; The component data includes the vibration acceleration, vibration velocity, vibration displacement and temperature of the wind turbine blades, hub, main shaft, gearbox, generator, converter and transformer, as well as the lubricating oil temperature and viscosity of the wind turbine main shaft, gearbox and generator. Environmental data includes wind speed, wind direction, temperature, humidity, air pressure, wind shear index, turbulence intensity, atmospheric stability, and air density of the environment in which the wind turbine is located; The topology data represents the physical topology of the wind turbine drivetrain. Specifically, the power, speed, pitch angle, three-phase current, three-phase voltage, power, speed and grid frequency of the wind turbine generator set are collected through the power transmitter, speed sensor, absolute encoder of the pitch motor, current transformer, voltage transformer and grid frequency transmitter. Among them, the pitch motor is one of the key components of the wind turbine generator set. Vibration acceleration sensors, magnetoelectric velocity sensors, fiber optic grating sensors, and patch temperature sensors are used to collect vibration acceleration, vibration velocity, vibration displacement, and temperature of wind turbine blades, hubs, main shafts, gearboxes, generators, converters, and transformers. The lubricating oil temperature and viscosity of the main shaft, gearbox, and generator are collected through platinum resistance temperature sensors and online oil viscosity sensors. Combining ultrasonic anemometers, ultrasonic wind vanes, atmospheric temperature sensors, humidity sensors, and atmospheric pressure sensors, the wind speed, wind direction, temperature, humidity, and air pressure of the wind turbine's environment are collected. Using the temperature, humidity, and air pressure of the wind turbine's environment, referencing the formula... Calculate the air density of the environment in which the wind turbine is located, where, The air density of the environment in which the wind turbine is located. , and The data includes the temperature, humidity, and air pressure of the environment where the wind turbine is located; two ultrasonic anemometers are installed to collect wind speeds at two different heights of the wind turbine, referencing the formula... Calculate the wind shear index of the environment where the wind turbine is located, where... and They are respectively at the height of and The wind speed in the environment where the wind turbine is located. The wind shear index is used. The standard deviation and average wind speed formulas are used to calculate the standard deviation and average wind speed of the environment where the wind turbine is located. The turbulence intensity of the environment is calculated by the ratio of the standard deviation to the average wind speed. Two layers of atmospheric temperature sensors are installed to collect atmospheric temperatures at two different altitudes where wind turbines are located. The calculation is based on the formula... Calculate the atmospheric stability of the environment in which the wind turbine is located, where, The atmospheric stability of the environment in which the wind turbine is located. and They are respectively at the height of and The formula is based on the principle of temperature gradient calculation, which measures the atmospheric temperature of the environment in which the wind turbine is located. By combining a laser alignment instrument, a 3D laser scanner, a wire-type displacement sensor, an articulated arm measuring machine, and a vibration displacement sensor, the physical topology of the wind turbine drive chain is collected. A three-level physical information cleaning mechanism is constructed. Based on the wind turbine theoretical control curve, physical constraint boundaries are constructed. By comparing the collected operation data, environmental data, and topology data, abnormal data is automatically corrected and removed, realizing the first level of physical information cleaning, ensuring that the data conforms to physical laws, and eliminating invalid interference data from the source. An adaptive wavelet threshold denoising algorithm is adopted to dynamically adjust the denoising parameters and remove noise from the data after the first-level cleaning of physical information and the component data, thereby achieving the second-level cleaning of physical information. Based on an event-triggered asynchronous data fusion and alignment algorithm, high-precision spatiotemporal synchronization processing is performed on wind turbine monitoring data to unify the analysis benchmark and achieve three-level cleaning of physical information.

[0021] In S2, the process of constructing a multi-view adaptive spatiotemporal feature map based on operational data, component data, and topology data, and extracting feature data of the wind turbine generator set, includes: The characteristic data of wind turbine generator sets include time-domain characteristic data, frequency-domain characteristic data, and time-frequency-domain characteristic data; Each component of the wind turbine generator is set as a graph node. Based on the physical topology of the wind turbine generator drive chain, the connection weights between each graph node are initialized, and a static physical adjacency matrix is ​​constructed. The power, speed and pitch angle of the wind turbine generator set are extracted from the operating data. The vibration acceleration, vibration velocity, vibration displacement and temperature of the blades, hub, main shaft, gearbox, generator, converter and transformer of the wind turbine generator are extracted from the component data. The lubricating oil temperature of the main shaft, gearbox and generator of the wind turbine generator is also extracted. A multi-head attention mechanism is introduced, which inputs data extracted from running data and component data in real time, identifies abnormal graph nodes and their related nodes, automatically learns the relationship between graph nodes through graph network, dynamically adjusts the connection weight between abnormal graph nodes and their related nodes by referring to the connection weight between each graph node after initialization, and converts the real-time input data into the node attributes of each graph node, so as to realize the construction of multi-view adaptive spatiotemporal feature map. Using multi-view adaptive spatiotemporal feature maps as the carrier, graph convolution and causal convolution are performed on the node attributes of each graph node to obtain spatial topological features and time series features, respectively. Finally, a three-dimensional feature tensor of time-space-frequency is output. Combined with Fourier transform, time-domain feature data, frequency-domain feature data and time-frequency-domain feature data are obtained.

[0022] In S3, the process of outputting the operating condition label of a wind turbine generator using a deep embedded clustering algorithm and source-target domain migration adaptation technology, based on wind turbine generator characteristic data, operational data, and environmental data, includes: The operating condition labels for wind turbine generators include low wind speed shear, strong turbulence, high temperature and humidity, grid frequency fluctuation, low air density, and atmospheric instability. The three-phase current, three-phase voltage, power, speed, and grid frequency at the grid connection point of the wind turbine generator set are selected from the operating data. The selected data, environmental data, and wind turbine generator set characteristic data are standardized. Using the feature dimension unification technology, the standardized data are spliced ​​together according to the time dimension to construct a standardized feature matrix. An autoencoder is used to reduce the dimensionality of the standardized feature matrix, retaining the core features for distinguishing working conditions. A deep embedded clustering algorithm is used to perform K-means clustering on the dimensionality-reduced features, automatically dividing the initial working condition clusters and outputting the initial working condition labels. Historical data of wind turbine generators are obtained from the wind turbine generator data acquisition and monitoring control system. The historical data of wind turbine generators is standardized and a standardized historical feature matrix is ​​constructed. The historical data of wind turbine generators includes historical three-phase current, historical three-phase voltage, historical power, historical speed, grid frequency of historical grid connection point, historical environmental data, and historical wind turbine generator characteristic data. The standardized feature matrix and the standardized historical feature matrix are used as the target domain data and the source domain data, respectively. Since the target domain data and the source domain data are corresponding but acquired at different times, the maximum difference between each corresponding data in the source domain and the target domain is calculated by the difference between the source domain data and the target domain data. The maximum mean difference (i.e., MMD value) of the feature distribution between the source domain and the target domain is obtained to evaluate the difference in feature distribution. Set a threshold for the difference in feature distribution between the source domain and the target domain, compare the maximum mean difference between the feature distributions of the source domain and the target domain, adjust the migration adaptation parameters, and output the operating condition label of the wind turbine generator set. The specific comparison process described above includes the following steps: when the maximum mean difference between the feature distributions of the source domain and the target domain is lower than the threshold for the difference between the feature distributions of the source domain and the target domain, the migration adaptation parameters are not adjusted; when the maximum mean difference between the feature distributions of the source domain and the target domain is higher than or equal to the threshold for the difference between the feature distributions of the source domain and the target domain, the migration adaptation parameters are adjusted until the maximum mean difference between the feature distributions of the source domain and the target domain is lower than the threshold for the difference between the feature distributions of the source domain and the target domain.

[0023] In S4, the process of adjusting the multi-view adaptive spatiotemporal feature map based on the operating condition label and extracting the latest wind turbine generator feature data includes: The operating condition labels of wind turbine generators are encoded. Based on the priority of the impact of different operating conditions on the wind turbine generators, a basic weight coefficient is assigned to each type of operating condition label encoding, forming an operating condition label weight vector. A correlation matrix is ​​established between the operating condition label encoding of wind turbine generators and the nodes of the multi-view adaptive spatiotemporal feature map. Specifically, the operating condition labels for low wind speed shear, strong turbulence, high temperature and humidity, grid frequency fluctuation, low air density, and atmospheric instability are encoded as follows: , , , , and The working condition label weight vector is ; The current multi-view adaptive spatiotemporal feature map is read, and the connection weights between nodes in the multi-view adaptive spatiotemporal feature map are dynamically adjusted, driven by the correlation matrix and the working condition label weight vector. Specifically, this is done through the formula... ,in, To dynamically adjust the connection weights between nodes in the multi-view adaptive spatiotemporal feature map, Dynamically adjust the connection weights between nodes in the multi-view adaptive spatiotemporal feature map. It is the identity matrix. This is the weight vector for the working condition labels. It is an incidence matrix; Based on the connection weights between nodes in the dynamically adjusted multi-view adaptive spatiotemporal feature map, the multi-view adaptive spatiotemporal feature map is reconstructed. The latest wind turbine generator feature data is extracted by combining the graph convolution and causal convolution of the node attributes of each graph node with Fourier transform.

[0024] In S4, the process of combining the adaptive multinomial KAN network and the meta-learning module to deeply fuse the latest wind turbine feature data, operational data, and component data, and outputting the deep fused features includes: Select and integrate the three-phase current, three-phase voltage, and power of the wind turbine generator set from the operational data, as well as the lubricating oil viscosity of the wind turbine generator's main shaft, gearbox, and generator from the component data, to obtain basic data for deep fusion features; The deep fusion feature base data and the latest wind turbine generator feature data are normalized; a multi-channel one-dimensional convolutional neural network is used to process the normalized data respectively, extract the deep features of each mode, and generate a joint fault diagnosis feature vector through feature fusion. The joint fault diagnosis feature vector is input into the adaptive multinomial KAN network. The deep nonlinear coupling feature is fitted using a multinomial basis function. The order of the adaptive multinomial is dynamically adjusted according to the operating condition label of the wind turbine generator, the weight of the adaptive multinomial KAN network is optimized, and the deep nonlinear coupling feature is output. Using deep nonlinear coupling features as the training set, a meta-learning module is constructed to output meta-optimized features. The joint fault diagnosis feature vector, deep nonlinear coupling features, and meta-optimized features are aggregated. A cross-modal attention mechanism is introduced to automatically assign weights to each modality feature. Through weighted integration, the joint fault diagnosis feature vector, deep nonlinear coupling features, and meta-optimized features are fused. First, the fused joint fault diagnosis feature vector, deep nonlinear coupling feature, and meta-optimization feature are linearly compressed, and then normalized to obtain the deep fused feature.

[0025] In S5, a neural network algorithm is used to construct a self-diagnosis model for wind turbine generator faults based on deep fusion features. The process of outputting the fault self-diagnosis coefficients includes: The deep fusion features are divided into training and test sets in a 7:3 ratio. A neural network algorithm is used, with the training set data as input data and the fault self-diagnosis coefficient as output data. The nonlinear relationship between the deep fusion features and the fault self-diagnosis coefficient is learned to obtain the trained model. The test set data is input into the trained model, the model parameters are adjusted, the model performance is optimized, and the final wind turbine generator fault self-diagnosis model is obtained. Combined with the current deep fusion features, the fault self-diagnosis coefficient is output.

[0026] In S6, the process of evaluating the fault level of wind turbine generators using a two-stage calibration evaluation method based on the fault self-diagnosis coefficient includes: Wind turbine generator set fault levels are classified into Level I, Level II, Level III, and Level IV faults. A Level I fault indicates that the fault does not affect the normal operation of the wind turbine generator set; a Level II fault indicates that the fault will cause minor abnormalities during operation and generally does not require intervention; a Level III fault indicates that the fault will cause general malfunctions during operation and requires intervention; and a Level IV fault indicates that the fault will cause serious malfunctions during operation, potentially even leading to immediate shutdown, and requires intervention. A two-stage calibration evaluation method is adopted. Based on the wind turbine generator fault self-diagnosis model and combined with the current deep fusion characteristics, the fault self-diagnosis coefficient of the past 30 days is output. The mean and standard deviation of the fault self-diagnosis coefficient of the past 30 days are calculated according to the mean formula and standard deviation formula. Based on the mean and standard deviation of the fault self-diagnosis coefficient over the past 30 days, the boundary thresholds between Level I and Level II faults, Level II and Level III faults, and Level III and Level IV faults were calculated. The calculation process includes: in, , and These are the thresholds for the distinction between Level I and Level II faults, the thresholds for the distinction between Level II and Level III faults, and the thresholds for the distinction between Level III and Level IV faults. and These are the mean and standard deviation of the fault self-diagnosis coefficient over the past 30 days, respectively. By using the boundary thresholds between Level I and Level II faults, Level II and Level III faults, and Level III and Level IV faults, a first-order assessment of the fault level of wind turbine generator sets is achieved, that is, a coarse classification of the fault level of wind turbine generator sets is realized. The specific coarse grading process includes: Q1. When the fault self-diagnosis coefficient is greater than or equal to 0 and less than the threshold between Level I and Level II faults, the first-order evaluation result of the fault level of the wind turbine generator set is Level I fault. Q2. When the fault self-diagnosis coefficient is greater than or equal to the boundary threshold between Level I and Level II faults and less than the boundary threshold between Level II and Level III faults, the first-order evaluation result of the fault level of the wind turbine generator set is Level II fault. Q3. When the fault self-diagnosis coefficient is greater than or equal to the boundary threshold between level II and level III faults and less than the boundary threshold between level III and level IV faults, the first-order evaluation result of the fault level of the wind turbine generator set is level III fault. Q4. When the fault self-diagnosis coefficient is greater than or equal to the threshold between Level III and Level IV faults and less than or equal to 1, the first-order evaluation result of the fault level of the wind turbine generator set is Level IV fault. From the fault self-diagnosis coefficients of the past 30 days, fault self-diagnosis coefficients for 5 consecutive days are retrieved. Confidence weights are assigned to each of these 5 retrieved fault self-diagnosis coefficients, and a weighted value for fault self-diagnosis is calculated. Then, the deviation rate between the current fault self-diagnosis coefficient and the weighted confidence level is calculated. The calculation formulas involved include: in, This is a weighted value for fault self-diagnosis. , , , and These are the fault self-diagnosis coefficients for five consecutive days, retrieved from the fault self-diagnosis coefficients of the past 30 days. , , , and These are the confidence weights of the fault self-diagnosis coefficients for five consecutive days, retrieved from the fault self-diagnosis coefficients of the past 30 days. This represents the current fault self-diagnosis coefficient; This represents the deviation rate between the current fault self-diagnosis coefficient and the weighted confidence level. Based on the deviation rate between the current fault self-diagnosis coefficient and the weighted confidence level, a second-order assessment of the fault level of the wind turbine generator set is achieved, that is, the accurate classification of the fault level of the wind turbine generator set is realized, and the final fault level of the wind turbine generator set is obtained. The specific second-order evaluation process includes: R1, If ​​the deviation rate between the current fault self-diagnosis coefficient and the weighted confidence level is less than or equal to... The first-order assessment result of the current wind turbine generator fault level will be used as the final wind turbine generator fault level. R2, If the deviation rate between the current fault self-diagnosis coefficient and the weighted confidence level is greater than... The coarse classification process of fault levels for wind turbine generators (Q1-Q4) involves a second-order evaluation until the deviation rate between the current fault self-diagnosis coefficient and the weighted confidence level is less than or equal to... Then stop the second-order evaluation process, set a threshold for the number of second-order evaluations, and if the actual number of second-order evaluations exceeds the set threshold, adjust the final fault level of the wind turbine generator set to Level IV fault.

[0027] In S6, the process of outputting a wind turbine generator fault early warning report using digital twin visualization technology includes: Build a digital twin 3D model to achieve synchronous mapping of wind turbine monitoring data, fault self-diagnosis coefficients, and the final fault level of the wind turbine. In the digital twin 3D model, based on the final fault level of the wind turbine generator, the corresponding early warning indicator signal is presented intuitively, and the wind turbine generator fault early warning report is output, realizing the hierarchical and visual early warning display of the fault level of the wind turbine generator. The specific process of displaying tiered, visualized early warnings includes: U1 and Level I faults: Green indicator displayed, no warning indicator; U2 and II level faults: A yellow flashing indicator is displayed, triggering a fault self-diagnosis coefficient fluctuation pop-up window; U3 and Level III faults: An orange highlighted indicator will be displayed, and a pop-up window will be shown to display the current monitoring data changes of the wind turbine generator set. U4 and IV level faults: display a red warning sign and audible and visual prompts, trigger a fault self-diagnosis coefficient fluctuation pop-up window, and simultaneously display a pop-up window showing the current wind turbine generator monitoring data changes.

[0028] Example 2, as Figure 2 As shown, based on Embodiment 1, the present invention provides a technical solution: a wind turbine generator fault self-diagnosis and early warning system based on multi-parameter fusion, used to implement the above-mentioned wind turbine generator fault self-diagnosis and early warning method based on multi-parameter fusion, including a monitoring data acquisition module, a feature extraction module, a working condition division module, a reverse optimization and deep fusion module, a fault self-diagnosis module and a visualization module, wherein each module is communicatively connected; The monitoring data acquisition module collects monitoring data from wind turbine generators and constructs a three-level physical information cleaning mechanism to preprocess the collected data to ensure the accuracy of the input data. The monitoring data of wind turbine generators includes operational data, component data, environmental data, and topology data. The feature extraction module constructs a multi-view adaptive spatiotemporal feature map based on operational data, component data, and topology data, comprehensively capturing the multi-scale dynamic characteristics of the unit's operating status and extracting high-fidelity wind turbine generator feature data. The operating condition classification module adopts a deep embedded clustering algorithm and source domain-target domain transfer adaptation technology. Through wind turbine generator characteristic data, operation data and environmental data, it realizes automated and refined identification of operating conditions, outputs operating condition labels of wind turbine generators, and realizes operating condition classification based on environment adaptive transfer learning. The reverse optimization and deep fusion module adjusts the multi-view adaptive spatiotemporal feature map based on the working condition label and extracts the latest wind turbine feature data. Combined with the adaptive multinomial KAN network and meta-learning module, the latest wind turbine feature data, operating data and component data are deeply fused to output deep fused features. The fault self-diagnosis module uses a neural network algorithm to construct a fault self-diagnosis model for wind turbine generators based on deep fusion features and outputs fault self-diagnosis coefficients. The visualization module employs a two-stage calibration evaluation method, assesses the fault level of wind turbine generators based on the fault self-diagnosis coefficient, and outputs a wind turbine generator fault early warning report by combining digital twin visualization technology.

[0029] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for self-diagnosis and early warning of faults in wind turbine generator sets based on multi-parameter fusion, characterized in that, Includes the following steps: Collect monitoring data of wind turbine generator sets, construct a three-level physical information cleaning mechanism, and preprocess the collected data. The monitoring data of wind turbine generator sets includes operation data, component data, environmental data, and topology data. Based on the operational data, component data, and topology data, a multi-view adaptive spatiotemporal feature map is constructed, and feature data of the wind turbine generator set is extracted. Using a deep embedded clustering algorithm and source-target domain migration adaptation technology, the operating condition label of the wind turbine generator is output through the characteristic data of the wind turbine generator, the operating data and the environmental data; Based on the operating condition labels, the multi-view adaptive spatiotemporal feature map is adjusted, and the latest wind turbine generator feature data is extracted. Combining the adaptive multinomial KAN network and meta-learning module, the latest wind turbine generator feature data, the operating data, and the component data are deeply fused to output deep fused features. Using a neural network algorithm, a fault self-diagnosis model for wind turbine generators is constructed based on the deep fusion features, and fault self-diagnosis coefficients are output. A two-stage calibration evaluation method is adopted to assess the fault level of the wind turbine generator set based on the fault self-diagnosis coefficient. Combined with digital twin visualization technology, a fault early warning report of the wind turbine generator set is output.

2. The method for self-diagnosis and early warning of wind turbine generator faults based on multi-parameter fusion according to claim 1, characterized in that, The process of collecting monitoring data from wind turbine generators and constructing a three-level physical information cleaning mechanism includes: Different types of data acquisition devices are deployed to collect monitoring data of the wind turbine generator set. These devices include power transmitters, speed sensors, current transformers, voltage transformers, grid frequency transmitters, vibration acceleration sensors, magnetoelectric velocity sensors, fiber optic grating sensors, patch temperature sensors, platinum resistance temperature sensors, online oil viscosity sensors, ultrasonic anemometers, ultrasonic wind vanes, atmospheric temperature sensors, humidity sensors, atmospheric pressure sensors, laser alignment devices, 3D laser scanners, wire-type displacement sensors, and articulated arm measuring machines. The operating data includes the power, speed, pitch angle, three-phase current, three-phase voltage, power, speed of the wind turbine generator set, and the grid frequency at the grid connection point; The component data includes the vibration acceleration, vibration velocity, vibration displacement and temperature of the wind turbine blades, hub, main shaft, gearbox, generator, converter and transformer, as well as the lubricating oil temperature and viscosity of the wind turbine main shaft, gearbox and generator. The environmental data includes wind speed, wind direction, temperature, humidity, air pressure, wind shear index, turbulence intensity, atmospheric stability, and air density of the environment in which the wind turbine is located. The topology data is the physical topology of the wind turbine drive train; A three-level physical information cleaning mechanism is constructed. Based on the wind turbine theoretical control curve, physical constraint boundaries are constructed. The collected operating data, environmental data, and topology data are compared, and abnormal data are automatically corrected and removed to achieve first-level physical information cleaning. An adaptive wavelet threshold denoising algorithm is used to dynamically adjust the denoising parameters to remove noise from the data after the first-level cleaning of physical information and the component data, thereby achieving the second-level cleaning of physical information. Based on an event-triggered asynchronous data fusion and alignment algorithm, the monitoring data of the wind turbine generator is processed in a spatiotemporal synchronization manner to achieve three-level cleaning of physical information.

3. The method for self-diagnosis and early warning of wind turbine generator faults based on multi-parameter fusion according to claim 2, characterized in that, The process of constructing a multi-view adaptive spatiotemporal feature map based on operational data, component data, and topology data, and extracting feature data of the wind turbine generator set includes: The wind turbine generator set characteristic data includes time-domain characteristic data, frequency-domain characteristic data, and time-frequency-domain characteristic data; Each component of the wind turbine generator is set as a graph node. Based on the physical topology of the wind turbine generator transmission chain, the connection weights between each graph node are initialized, and a static physical adjacency matrix is ​​constructed. The power, speed and pitch angle of the wind turbine generator set are extracted from the operating data. The vibration acceleration, vibration velocity, vibration displacement and temperature of the wind turbine blades, hub, main shaft, gearbox, generator, converter and transformer are extracted from the component data. The lubricating oil temperature of the wind turbine main shaft, gearbox and generator is also extracted. A multi-head attention mechanism is introduced, and data extracted from the running data and the component data is input in real time to identify abnormal graph nodes and their related nodes. The association between graph nodes is automatically learned through the graph network. The connection weights between abnormal graph nodes and their related nodes are dynamically adjusted by referring to the connection weights between each graph node after initialization. The real-time input data is converted into the node attributes of each graph node to realize the construction of the multi-view adaptive spatiotemporal feature map. Using the multi-view adaptive spatiotemporal feature map as a carrier, graph convolution and causal convolution are performed on the node attributes of each graph node to obtain spatial topological features and time series features, respectively. Finally, a three-dimensional feature tensor of time-space-frequency is output. Combined with Fourier transform, the time domain feature data, the frequency domain feature data and the time-frequency domain feature data are obtained.

4. The method for self-diagnosis and early warning of wind turbine generator faults based on multi-parameter fusion according to claim 3, characterized in that, The process of outputting the operating condition label of the wind turbine generator set by employing a deep embedded clustering algorithm and source-target domain migration adaptation technology, using the wind turbine generator set feature data, the operating data, and the environmental data, includes: The operating condition labels of the wind turbine generator set include low wind speed shear operating condition label, strong turbulence operating condition label, high temperature and high humidity operating condition label, power grid frequency fluctuation operating condition label, low air density operating condition label and atmospheric instability operating condition label. The three-phase current, three-phase voltage, power, speed, and grid frequency of the wind turbine generator set at the grid connection point are selected from the operating data. The selected data, the environmental data, and the characteristic data of the wind turbine generator set are standardized. Using the feature dimension unification technology, the standardized data are spliced ​​together according to the time dimension to construct a standardized feature matrix. The standardized feature matrix is ​​reduced in dimensionality using an autoencoder. A deep embedded clustering algorithm is then used to perform K-means clustering on the reduced-dimensional features to automatically divide the initial working condition clusters and output the initial working condition labels. Historical data of wind turbine generators are obtained from the wind turbine generator data acquisition and monitoring control system, the historical data of wind turbine generators are standardized, and a standardized historical feature matrix is ​​constructed. Using the standardized feature matrix and the standardized historical feature matrix as the target domain data and source domain data respectively, calculate the maximum mean difference between the feature distributions of the source domain and the target domain. Set a threshold for the difference in feature distribution between the source domain and the target domain, compare the maximum mean difference between the feature distributions of the source domain and the target domain, adjust the migration adaptation parameters, and output the operating condition label of the wind turbine generator set.

5. The method for self-diagnosis and early warning of wind turbine generator faults based on multi-parameter fusion according to claim 4, characterized in that, The process of adjusting the multi-view adaptive spatiotemporal feature map based on operating condition labels and extracting the latest wind turbine generator feature data includes: The operating condition labels of the wind turbine generators are encoded respectively. Based on the priority of the impact of different operating conditions on the wind turbine generators, basic weight coefficients are assigned to the encoding of each type of operating condition label to form an operating condition label weight vector. The correlation matrix between the operating condition label encoding of the wind turbine generators and the nodes of the multi-view adaptive spatiotemporal feature map is established. Read the current multi-view adaptive spatiotemporal feature map, and dynamically adjust the connection weights between nodes of the multi-view adaptive spatiotemporal feature map, driven by the correlation matrix and the working condition label weight vector. Based on the connection weights between nodes in the dynamically adjusted multi-view adaptive spatiotemporal feature map, the multi-view adaptive spatiotemporal feature map is reconstructed. The latest wind turbine generator feature data is extracted by combining the graph convolution and causal convolution of the node attributes of each graph node with Fourier transform.

6. The method for self-diagnosis and early warning of wind turbine generator faults based on multi-parameter fusion according to claim 5, characterized in that, The process of combining the adaptive multinomial KAN network and the meta-learning module to deeply fuse the latest wind turbine feature data, the operational data, and the component data, and outputting the deep fused features, includes: The three-phase current, three-phase voltage, and power of the wind turbine generator set in the operational data, as well as the lubricating oil viscosity of the wind turbine generator's main shaft, gearbox, and generator in the component data, are selected and integrated to obtain basic data of deep fusion features. The deep fusion feature base data and the latest wind turbine generator feature data are normalized; a multi-channel one-dimensional convolutional neural network is used to process the normalized data respectively, extract the deep features of each mode, and generate a joint fault diagnosis feature vector through feature fusion. The joint fault diagnosis feature vector is input into the adaptive multinomial KAN network, and the deep nonlinear coupling feature is fitted using multinomial basis functions. The order of the adaptive multinomial is dynamically adjusted according to the operating condition label of the wind turbine generator set to optimize the weight of the adaptive multinomial KAN network and output the deep nonlinear coupling feature. Using the deep nonlinear coupling features as a training set, a meta-learning module is constructed to output meta-optimized features. The joint fault diagnosis feature vector, the deep nonlinear coupling features, and the meta-optimized features are aggregated. A cross-modal attention mechanism is introduced to automatically allocate the weights of each modality feature. Through weighted integration, the joint fault diagnosis feature vector, the deep nonlinear coupling features, and the meta-optimized features are fused. First, the fused joint fault diagnosis feature vector, the deep nonlinear coupling feature, and the meta-optimization feature are linearly compressed, and then normalized to obtain the deep fusion feature.

7. The method for self-diagnosis and early warning of wind turbine generator faults based on multi-parameter fusion according to claim 6, characterized in that, The process of constructing a wind turbine generator fault self-diagnosis model based on the deep fusion features using a neural network algorithm and outputting fault self-diagnosis coefficients includes: The deep fusion features are divided into training and testing sets. The neural network algorithm is used, with the training set data as input data and the fault self-diagnosis coefficient as output data. The nonlinear relationship between the input data and the output data is learned to obtain the trained model. The test set data is input into the trained model, the model parameters are adjusted, the model performance is optimized, and the final wind turbine generator fault self-diagnosis model is obtained. Combined with the current deep fusion features, the fault self-diagnosis coefficient is output.

8. The method for self-diagnosis and early warning of wind turbine generator faults based on multi-parameter fusion according to claim 7, characterized in that, The process of evaluating the fault level of a wind turbine generator set using a two-stage calibration evaluation method based on a fault self-diagnosis coefficient includes: The fault levels of the wind turbine generator set include Level I fault, Level II fault, Level III fault and Level IV fault; Using the aforementioned two-stage calibration evaluation method, based on the aforementioned wind turbine generator fault self-diagnosis model and combined with the current deep fusion features, the fault self-diagnosis coefficients for the past 30 days are output, and the mean and standard deviation of the fault self-diagnosis coefficients for the past 30 days are calculated. Based on the mean and standard deviation of the fault self-diagnosis coefficients over the past 30 days, the boundary thresholds between Level I and Level II faults, Level II and Level III faults, and Level III and Level IV faults were calculated. The first-order assessment of the fault level of the wind turbine generator is achieved by using the boundary thresholds between Level I and Level II faults, Level II and Level III faults, and Level III and Level IV faults. From the fault self-diagnosis coefficients of the past 30 days, retrieve the fault self-diagnosis coefficients for 5 consecutive days, assign confidence weights to the retrieved fault self-diagnosis coefficients for 5 consecutive days, calculate the fault self-diagnosis weighted value, and then calculate the deviation rate between the current fault self-diagnosis coefficient and the weighted confidence level. Based on the deviation rate between the current fault self-diagnosis coefficient and the weighted confidence level, a second-order evaluation of the fault level of the wind turbine generator set is achieved, and the final fault level of the wind turbine generator set is obtained.

9. The method for self-diagnosis and early warning of wind turbine generator faults based on multi-parameter fusion according to claim 8, characterized in that, The process of using digital twin visualization technology to output a wind turbine generator fault early warning report includes: Build a digital twin 3D model to achieve synchronous mapping of wind turbine monitoring data, fault self-diagnosis coefficients, and the final fault level of the wind turbine. In the digital twin 3D model, based on the final fault level of the wind turbine generator set, a corresponding early warning indicator signal is presented, and a wind turbine generator set fault early warning report is output, realizing the hierarchical and visual early warning display of the fault level of the wind turbine generator set.

10. A wind turbine generator fault self-diagnosis and early warning system based on multi-parameter fusion, used to implement the wind turbine generator fault self-diagnosis and early warning method based on multi-parameter fusion as described in any one of claims 1-9, comprising a monitoring data acquisition module, a feature extraction module, an operating condition division module, a reverse optimization and deep fusion module, a fault self-diagnosis module, and a visualization module, wherein, The various modules are connected for communication, characterized in that, The monitoring data acquisition module collects monitoring data from the wind turbine generator set, constructs a three-level physical information cleaning mechanism, and preprocesses the collected data. The monitoring data from the wind turbine generator set includes operational data, component data, environmental data, and topology data. The feature extraction module constructs a multi-view adaptive spatiotemporal feature map based on the operating data, the component data, and the topology data, and extracts the feature data of the wind turbine generator set. The operating condition classification module uses a deep embedded clustering algorithm and source domain-target domain migration adaptation technology to output the operating condition label of the wind turbine generator set through the wind turbine generator set feature data, the operating data and the environmental data; The reverse optimization and deep fusion module adjusts the multi-view adaptive spatiotemporal feature map based on the operating condition label, extracts the latest wind turbine generator feature data, and combines the adaptive multinomial KAN network and meta-learning module to deeply fuse the latest wind turbine generator feature data, the operating data and the component data to output deep fusion features. The fault self-diagnosis module uses a neural network algorithm to construct a fault self-diagnosis model for wind turbine generators based on the deep fusion features, and outputs fault self-diagnosis coefficients. The visualization module uses a two-stage calibration evaluation method to assess the fault level of the wind turbine generator set based on the fault self-diagnosis coefficient, and outputs a wind turbine generator set fault early warning report by combining digital twin visualization technology.

Images