Predictive maintenance method and system for wind turbine

Abstract

The invention discloses a predictive maintenance method and system for a wind turbine, and the method comprises the following steps: 1), collecting the multi-source data of the wind turbine, including operation data, environment data and historical maintenance data; 2) preprocessing the collected data, extracting high-frequency features and low-frequency trends of vibration signals by adopting wavelet transform, performing principal component analysis to reduce dimensions and redundant data, and eliminating noise through Kalman filtering; 3) constructing a deep learning prediction model fused with multi-source data, inputting the model including time sequence data, environmental parameters and equipment image data, and outputting health state scores and potential fault categories of equipment; 4) performing fault precursor analysis based on a local anomaly clustering algorithm; according to the method and the system, the aspects of data comprehensiveness, model prediction precision, real-time performance, maintenance automation and the like are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of wind power generation equipment maintenance and fault prediction, and in particular relates to a predictive maintenance method and system for a wind turbine. Background Art

[0002] Wind turbines operate in a complex environment. They are exposed to adverse conditions such as high wind speeds, temperature and humidity changes, and dust for a long time. Key components are susceptible to wear, fatigue, and failure. Once a wind turbine fails, it will not only lead to a decrease in power generation efficiency, but may also cause equipment damage and economic losses. Therefore, how to efficiently and reliably predict and maintain wind turbine failures has become a technical problem that needs to be solved in the industry.

[0003] In the prior art, wind turbine maintenance methods mainly include periodic maintenance and condition-based maintenance. Traditional periodic maintenance is usually carried out according to a fixed cycle, which fails to fully consider the real-time status and environmental impact of the equipment, and easily leads to resource waste or untimely maintenance.

[0004] In addition, existing technologies have obvious deficiencies in data collection, fault prediction, maintenance strategy and intelligence. Therefore, there is an urgent need for an intelligent predictive maintenance method that can integrate multi-source data, improve fault prediction accuracy, and dynamically optimize maintenance strategies to improve the operational reliability and operation and maintenance efficiency of wind turbines. Summary of the invention

[0005] The object of the present invention is to provide a predictive maintenance method and system for wind turbines, which method and system have significant improvements in data comprehensiveness, model prediction accuracy, real-time performance and maintenance automation.

[0006] The technical solution adopted by the present invention is as follows: A predictive maintenance method for a wind turbine comprises the following steps: 1) Collect multi-source data of wind turbines, including operation data, environmental data and historical maintenance data; 2). Preprocess the collected data, use wavelet transform to extract the high-frequency characteristics and low-frequency trends of the vibration signal, use principal component analysis to reduce the dimension and reduce redundant data, and use Kalman filtering to eliminate noise; 3) Construct a deep learning prediction model that integrates multi-source data, the model input includes time series data, environmental parameters and equipment image data, and outputs the health status score and potential fault category of the equipment; 4) Fault precursor analysis based on local anomaly clustering algorithm, including: Divide the real-time data into fixed time windows and extract statistical features within the sliding windows; Perform cluster analysis on outliers based on K-Means or DBSCAN algorithm; The dynamic time warping algorithm is used to match abnormal clusters with historical failure patterns to determine the possibility of potential failures; 5). Dynamically adjust the maintenance cycle, the adjustment formula is: ;in, For the adjusted maintenance cycle, As the basic maintenance cycle, and is the adjustment factor, is the current operating load, Score health status; 6). Optimize the health status score by combining the dynamic environmental influencing factors. The optimization formula is: ;in, is the adjusted health score, is the initial health score, is the environmental dynamic weight, which is determined by the principal component weight and the characteristic value of the environmental feature; 7) Automatically generate maintenance plans based on prediction results and historical maintenance records, including maintenance parts, required tools and operation steps, and sort tasks according to priority; 8). Send the maintenance plan to the operation and maintenance personnel or the automated maintenance system for execution; Among them, in step 1), the operating data includes real-time status information of vibration, current and speed components; environmental data includes wind speed, wind direction, temperature and humidity, and dust concentration environmental parameters; historical maintenance data includes fault category, cause and maintenance record; the operating data in the multi-source data also obtains the temperature distribution of key components through thermal imaging equipment, combines image processing algorithms to detect abnormal hot spots, and optimizes fault prediction results; Among them, the local anomaly clustering algorithm in step 4) uses the sliding window technology to extract the mean, standard deviation and peak value of the time series data as clustering features, and the clustering results of the anomaly points are matched with the historical fault data through the dynamic time warping algorithm; Among them, the running load in step 5) The relationship between wind speed and power generation is represented by the power curve model, which is calculated by real-time monitoring of power generation, rotation speed and wind speed; Among them, the environmental dynamic weight in step 6) Calculated by the following formula: ;in, is the environmental dynamic weight, is the principal component weight of environmental characteristics, is the environmental characteristic value, is the number of environmental parameters; j is the environmental parameter index, indicating the jth environmental parameter; Wherein, in step 7), the task priority of the maintenance plan is based on the health status score and the environmental risk score, and the wind turbines with low scores or high risks are processed first; Among them, the deep learning prediction model in step 3) adopts a multimodal Transformer model to process time series data, environmental parameters and image data respectively, and the fusion results are used to output the health status score; Wherein, an edge computing node is deployed in the control cabinet of the wind turbine, and the edge node processes sensor data in real time and uploads abnormal detection results to the cloud; Among them, in step 7), the automatically generated maintenance plan combines dynamic priorities to optimize maintenance paths and task scheduling, thereby reducing operation and maintenance costs; Among them, in step 5), the real-time environmental data is dynamically updated and parameters to adapt to changes in environmental conditions.

[0007] A wind turbine predictive maintenance system, comprising: A data acquisition module for collecting operation data, environmental data and historical maintenance data of the wind turbine; Edge computing module, used to denoise the collected operation data and environmental data and extract key feature values; A multimodal deep learning prediction module that fuses time series data, environmental parameters, and image data to generate health status scores and potential fault categories for wind turbines; Dynamic maintenance optimization module, which is used to adjust the maintenance cycle in real time based on health status score, environmental dynamic influencing factors and operating load, and automatically generate maintenance plans including maintenance parts, tools and operation steps; Central monitoring and management platform for visual monitoring of wind turbines’ real-time operating status, health scores and maintenance plans, as well as allocation and scheduling of maintenance tasks.

[0008] Wherein, the multimodal deep learning prediction module includes: Time series data processing unit, used to extract time series characteristics of vibration, current and speed; Environmental data processing unit, used to analyze environmental characteristics such as wind speed, temperature and humidity, and dust concentration; An image data processing unit that processes thermal images of key components through a convolutional neural network to detect areas of abnormal temperature; The feature fusion and prediction unit fuses the features of the time series data, environmental data, and image data through a Transformer model, and outputs a health status score and a potential fault category.

[0009] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: Multi-source data collection and integration: By integrating operational data, environmental data, and historical maintenance data, we can fully understand the operating status of wind turbines and avoid information loss that may be caused by a single data source.

[0010] Edge computing and real-time analysis: Deploy edge computing modules to process data in real time, reduce data transmission delays, and improve the real-time and efficiency of fault prediction.

[0011] Multimodal deep learning prediction model: Integrates time series, environmental parameters, and image data, and uses a multimodal Transformer model to achieve equipment health status scoring and fault category prediction, significantly improving prediction accuracy.

[0012] Local anomaly clustering analysis: Extract statistical features through sliding windows and combine with K-Means or DBSCAN algorithms to perform anomaly clustering analysis to further improve the sensitivity of fault precursor detection.

[0013] Dynamic maintenance cycle adjustment: Dynamically adjust the maintenance cycle based on operating load, health score and environmental influencing factors to improve the adaptability of the maintenance plan and resource utilization efficiency.

[0014] Automated maintenance plan generation: Based on prediction results and historical records, prioritized maintenance plans are automatically generated to optimize the allocation and scheduling of maintenance tasks and reduce operation and maintenance costs.

[0015] Optimize the health scoring model: Incorporate environmental dynamic influencing factors into the calculation of health scores to ensure that the scores are more applicable and accurate.

[0016] Comprehensive data storage and visual management: Use time series database and relational database to manage different types of data, and intuitively present the status of wind turbines through a visual platform to facilitate decision-making by operation and maintenance personnel.

[0017] The present invention has significant improvements in data comprehensiveness, model prediction accuracy, real-time performance, and maintenance automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the maintenance method of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0020] See also Figure 1 , a predictive maintenance method for a wind turbine, comprising the following steps: 1). Collect multi-source data of wind turbines, including operation data, environmental data and historical maintenance data; operation data includes real-time status information of vibration, current and speed components; environmental data includes wind speed, wind direction, temperature and humidity, and dust concentration environmental parameters; historical maintenance data includes fault category, cause and maintenance record; operation data in multi-source data also obtains temperature distribution of key components through thermal imaging equipment, combines image processing algorithms to detect abnormal hot spots, and optimizes fault prediction results.

[0021] Specifically, ① Operation data collection Data collection equipment, the operating data mainly reflects the real-time status information of key components of wind turbines, and the following sensors need to be deployed: Vibration sensor: installed on the main shaft bearing, gearbox and generator housing, used to collect vibration data of mechanical parts. Accelerometer or piezoelectric vibration sensor is used. Data collection frequency: 10-1000 Hz, the specific frequency depends on the collection component; Current sensor: Deployed in the output circuit of the generator to monitor current fluctuations and identify possible overloads or electrical faults. A Hall effect current sensor is used; Speed ​​sensor: installed at the end of the rotor shaft, used to measure the blade rotation speed in real time. Use photoelectric encoder or magnetic induction speed sensor; Data transmission and storage Edge computing module: Edge computing devices are deployed inside the turbine to preliminarily process the operating data, remove noise and calculate key characteristic values ​​(mean, standard deviation); Data protocol: MQTT (lightweight IoT protocol) is used to transmit operation data to the central server to ensure low latency and high efficiency; Storage method: Use a time series database to store running data to support subsequent time series analysis.

[0022] ②Environmental data collection Data collection equipment, environmental data reflects the operating conditions of the area where the turbine is located. The following sensors need to be deployed: Wind speed and direction sensor: deployed on the top of the turbine tower to monitor wind speed and direction changes in real time. Ultrasonic anemometer and wind vane are used; Temperature and humidity sensors: deployed in the middle and top of the tower, used to measure ambient temperature and humidity to determine the fan cooling efficiency and operating environment. Digital temperature and humidity sensors are used; Dust concentration sensor: deployed at the wind turbine base or nearby areas to monitor the concentration of particulate matter in the air and identify the impact that dust may have on the equipment. Laser scattering particle sensor is used; Rainfall / snowfall monitoring: Deploy rain and snow sensors to detect the impact of weather changes on turbine operation; Data transmission and storage Wireless transmission module: The data of environmental sensors is transmitted to edge computing devices via LoRa or NB-IoT wireless modules; Real-time monitoring: Dynamically visualize environmental data on a central server to provide an overall environmental monitoring view of the wind farm.

[0023] ③Historical maintenance data integration Data sources,Historical maintenance data needs to integrate information from the following sources: Operation and maintenance records: Turbine maintenance logs, including regular maintenance and fault repair records; Contains information: maintenance time, parts, fault type, repair measures, spare parts used, etc.; Fault data: Turbine fault alarm records and corresponding cause analysis; Contains information: fault type, occurrence time, faulty parts, downtime; Spare parts usage records: including the spare parts model, quantity and replacement reason used for each maintenance; Data Storage and Retrieval: Maintenance database: Design a maintenance data storage system based on a relational database. The table structure includes: Fault category table: records the classification of all known faults; Parts list: records the key parts information of wind turbines; Maintenance record table: associated with the fault category and parts table to record specific maintenance details; Data labeling: adding timestamps, component identification, and fault classification labels to maintenance records for quick retrieval and modeling; Data Integration: Correlate historical maintenance data with real-time operational data: Matching operating and maintenance data by turbine number; Perform cluster analysis on historical data to generate typical failure modes for training prediction models.

[0024] ④ Overall data integration architecture Data collection architecture: Perception layer: deploy various sensor devices to collect operation data and environmental data; Transport layer: Use edge computing nodes to pre-process data and upload it to the central server using wireless communication (LoRa, 4G / 5G); Data processing layer: The central server integrates operation data, environmental data and historical maintenance data to form a complete wind turbine status database; Data Management Platform: Time series analysis platform: used for trend analysis of operational data and environmental data; Historical record retrieval system: provides a quick retrieval and analysis interface for operation and maintenance personnel; Real-time monitoring platform: Displays the real-time status of the turbine through a dashboard, including health scores, environmental parameters and historical fault information.

[0025] 2). Preprocess the collected data, use wavelet transform to extract the high-frequency characteristics and low-frequency trends of the vibration signal, use principal component analysis to reduce the dimension and reduce redundant data, and use Kalman filtering to eliminate noise; Specifically, ① Wavelet transform extracts high-frequency features and low-frequency trends of vibration signals The basic principle of wavelet transform: Wavelet transform is a time-frequency analysis tool that can decompose the signal into multiple components, representing the high-frequency part (local mutations or abnormal features) and the low-frequency part (overall trend).

[0026] Implementation steps: Input signal: The vibration signal is collected from the key components of the turbine, such as the main shaft, gearbox or generator. The input signal is the time series data X(t); Select wavelet basis function: select Daubechies wavelet (db4 or db6) or Symlet wavelet, because it is suitable for feature extraction of mechanical vibration signals; Decomposition level setting: Set the number of decomposition levels according to the sampling frequency and target feature frequency band , ;in Sampling frequency. , The frequency range obtained by decomposition; Signal decomposition: Use discrete wavelet transform (DWT) to decompose the signal X(t) into multiple layers of high-frequency components and low frequency components . ; Where X(t): original signal (time series data), such as vibration signal, unit: (Acceleration). L: The number of wavelet decomposition layers, which determines the depth of decomposition and is usually set based on the sampling frequency and target frequency band. : The high-frequency component of the i-th layer represents the local mutation characteristics, and the unit is consistent with the original signal. : The Lth layer low-frequency component, indicating the overall trend, the unit is consistent with the original signal; High-frequency and low-frequency signal extraction: High-frequency components are used to extract features (such as abnormal signals), and low-frequency components are used to monitor the overall operating status.

[0027] ②Principal component analysis (PCA) reduces dimension and redundancy Implementation steps: Construct data matrix: Organize multi-source data (vibration, current, speed, etc.) into a matrix: ;in : number of features, n: length of time series; Standardization: Standardize the data matrix to have a mean of 0 and a variance of 1: ;in : The i-th feature data after standardization. Original i-th feature data, unit: specific feature unit (such as vibration , current is A, speed is rpm. μ: characteristic mean, unit is Same as σ characteristic standard deviation, unit is same as same; Calculate the covariance matrix: Covariance matrix: ; Where X: multi-source data matrix, dimension is ,in is the number of samples, is the number of features. C: covariance matrix, used to measure the correlation between features, dimension is ; Eigendecomposition: Finding the eigenvalues ​​of the covariance matrix and the eigenvector , The eigenvector of the covariance matrix represents the direction of the i-th principal component. Unit: dimensionless. Principal component selection: Select the first K principal components based on the cumulative contribution rate: ;in : The feature matrix after dimensionality reduction retains K principal components with higher information contribution rate, and the dimension is .

[0028] 3). Construct a deep learning prediction model that integrates multi-source data, wherein the model input includes time series data, environmental parameters and equipment image data, and outputs the health status score and potential fault category of the equipment; the deep learning prediction model adopts a multimodal Transformer model to process time series data, environmental parameters and image data respectively, and the fusion result is used to output the health status score.

[0029] Specifically, ①Time series data processing module Input format, input shape: ( , t, f), where: : number of samples; t number of time steps; f: time series feature number (vibration, current, speed, etc.); Embedding generation, formula: ;in, Time series embedding representation; Input time series data; Embedding weight matrix; Embedding bias; Position encoding, used to preserve time step information; Time series embedding dimension; Feature extraction: extracting time series features through Transformer Encoder .

[0030] ②Environmental parameter processing module Input format, input shape: ( , e), where: : number of samples; e: characteristic number of environmental parameters (wind speed, temperature and humidity, etc.); Embedding generation, formula: ;in, Embedded representation of environmental parameters; : Input environment parameters; Embedding weight matrix; Embedding bias; Environmental parameters are embedded in the dimension; Feature extraction: Extracting environmental features through Transformer Encoder .

[0031] ③Image data processing module Input format, input shape: ( , h, w, c), where: : number of samples; h: image height; w: image width; c: number of image channels; Embedding generation, the image generates embedded features through a convolutional neural network: ;in, Image embedding representation; Input image; Feature extraction: Extract image features through Transformer Encoder .in, Image embedding dimension.

[0032] ④Feature fusion module Fusion operation, concatenating and fusing the features of each modality: Formula: ;in, The fused feature vector; Fully connected fusion, formula: ;in The fused features; Fully connected weight matrix; Fully connected bias; Fusion feature dimensions.

[0033] ⑤Output module Health status score, output device health status score through regression layer: Formula: ;in, :Health status score; : weight matrix; Bias; Fault classification, output the probability distribution of equipment fault categories through the classification layer: Formula: ;in, : Fault category probability distribution; : weight matrix; : bias; C: number of fault categories.

[0034] ⑥Model training Loss function, health status score loss (mean square error): ;in, True Health Score; Predicted health scores; Fault classification loss (cross entropy): ;in, : Actual fault category label; Predict the probability of failure category; Total loss: ; Among them, α, β: weight factors.

[0035] ⑦Model output Health status score: scalar, ranging from [0,1], reflecting the overall health status of the device; Fault classification: A vector representing the probability distribution of potential fault categories of the device.

[0036] 4) Fault precursor analysis based on local anomaly clustering algorithm, including: Divide the real-time data into fixed time windows and extract statistical features within the sliding windows; Perform cluster analysis on outliers based on K-Means or DBSCAN algorithm; The dynamic time warping algorithm is used to match abnormal clusters with historical failure patterns to determine the possibility of potential failures. The local anomaly clustering algorithm uses sliding window technology to extract the mean, standard deviation and peak value of time series data as clustering features. The clustering results of abnormal points are matched with historical fault data through dynamic time warping algorithm.

[0037] The specific implementation steps are as follows: ① Statistical feature extraction within the sliding window: Data partitioning: real-time data is divided into fixed time windows and processed using sliding window technology: Input data shape: time series data Where n: number of data points; f: number of features (vibration, current, speed, etc.). Sliding window parameters: window size w: number of data points contained in each window; sliding step s: number of points the window moves each time; Feature extraction, within each window, calculate the following statistical features: Mean: ; Wherein, μ: mean value of data in the window; : the i-th data point; w: window size. Standard deviation: ;σ: standard deviation of the data in the window. Peak value: ; p: maximum value in the window; Feature matrix generation, the features (mean, standard deviation, peak) calculated for each window form a feature vector: F = [μ, σ, p]. The feature vectors of all windows are stacked into a feature matrix: ; m: number of windows; k=3f: feature dimension extracted by each window (3 statistical features for each original feature).

[0038] ② Cluster analysis of abnormal points K-Means clustering, use K-Means to cluster the feature matrix F, ;in, : The feature vector of the i-th window; Kth cluster; cluster The center of; K: the number of clusters (including one outlier cluster); DBSCAN clustering, using DBSCAN to cluster the feature matrix F based on density: define parameters: radius threshold ε: define the density range; minimum number of neighbor points minPts: specify the minimum density of core points. Marking results: points with sufficient density are classified as normal points; points in sparse areas are marked as abnormal points; Output, the clustering results mark each window as: normal point: belongs to a dense cluster; abnormal point: isolated point or sparse cluster.

[0039] ③ Dynamic Time Warping (DTW) matching Dynamic time warping formula, using dynamic time warping to match the abnormal point sequence with the historical failure mode: ; Where, D(i, j): cumulative matching distance; , : Feature vector of abnormal point sequence and historical failure mode; min: path selection of dynamic programming; Threshold judgment, calculate the cumulative matching distance D (m, n) between the abnormal point sequence and each historical fault mode: If D (m, n) < threshold, it is judged that the abnormal point sequence matches a certain fault mode and there may be a potential fault.

[0040] 5). Dynamically adjust the maintenance cycle, the adjustment formula is: ;in, For the adjusted maintenance cycle, As the basic maintenance cycle, and is the adjustment factor, is the current operating load, Score health status; run load The relationship between wind speed and power generation is represented by the power curve model. Dynamic update through real-time environmental data and parameters to adapt to changes in environmental conditions.

[0041] 6). Optimize the health status score by combining the dynamic environmental influencing factors. The optimization formula is: ;in, is the adjusted health score, is the initial health score, is the environmental dynamic weight, which is determined by the principal component weight and the characteristic value of the environmental feature; Environmental Dynamic Weight Calculated by the following formula: ;in, is the environmental dynamic weight, is the principal component weight of environmental characteristics, is the environmental characteristic value, and m is the number of environmental parameters.

[0042] 7) Automatically generate maintenance plans based on prediction results and historical maintenance records, including maintenance parts, required tools and operation steps, and sort tasks according to priority; The maintenance plan’s task priority is based on the health status score and environmental risk score, with wind turbines with low scores or high risks being given priority. Automatically generated maintenance plans are combined with dynamic priorities to optimize maintenance paths and task scheduling, reducing operation and maintenance costs.

[0043] 8). Send the maintenance plan to the operation and maintenance personnel or the automated maintenance system for execution; In support of the method, a wind turbine predictive maintenance system, comprising: A data acquisition module for collecting operation data, environmental data and historical maintenance data of the wind turbine; Edge computing module, used to denoise the collected operation data and environmental data and extract key feature values; A multimodal deep learning prediction module that fuses time series data, environmental parameters, and image data to generate health status scores and potential fault categories for wind turbines; Dynamic maintenance optimization module, which is used to adjust the maintenance cycle in real time based on health status score, environmental dynamic influencing factors and operating load, and automatically generate maintenance plans including maintenance parts, tools and operation steps; Central monitoring and management platform for visual monitoring of wind turbines’ real-time operating status, health scores and maintenance plans, as well as allocation and scheduling of maintenance tasks.

[0044] Furthermore, the multimodal deep learning prediction module includes: Time series data processing unit, used to extract time series characteristics of vibration, current and speed; Environmental data processing unit, used to analyze environmental characteristics such as wind speed, temperature and humidity, and dust concentration; An image data processing unit that processes thermal images of key components through a convolutional neural network to detect areas of abnormal temperature; The feature fusion and prediction unit fuses the features of the time series data, environmental data, and image data through a Transformer model, and outputs a health status score and a potential fault category.

[0045] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predictive maintenance of a wind turbine, characterized in that: The following steps are involved: 1) Collect multi-source data of wind turbines, including operation data, environmental data and historical maintenance data; 2). Preprocess the collected data, use wavelet transform to extract the high-frequency characteristics and low-frequency trends of the vibration signal, use principal component analysis to reduce the dimension and reduce redundant data, and use Kalman filtering to eliminate noise; 3) Construct a deep learning prediction model that integrates multi-source data, the model input includes time series data, environmental parameters and equipment image data, and outputs the health status score and potential fault category of the equipment; 4) Fault precursor analysis based on local anomaly clustering algorithm, including: Divide the real-time data into fixed time windows and extract statistical features within the sliding windows; Perform cluster analysis on outliers based on K-Means or DBSCAN algorithm; The dynamic time warping algorithm is used to match abnormal clusters with historical failure patterns to determine the possibility of potential failures. 5). Dynamically adjust the maintenance cycle, the adjustment formula is: ;in, For the adjusted maintenance cycle, As the basic maintenance cycle, and is the adjustment factor, is the current operating load, Score health status; 6). Optimize the health status score by combining the dynamic environmental influencing factors. The optimization formula is: ;in, is the adjusted health score, is the initial health score, is the environmental dynamic weight, which is determined by the principal component weight and the characteristic value of the environmental feature; 7) Automatically generate maintenance plans based on prediction results and historical maintenance records, including maintenance parts, required tools and operation steps, and sort tasks according to priority; 8). Send the maintenance plan to the operation and maintenance personnel or the automated maintenance system for execution.

2. A wind turbine predictive maintenance method according to claim 1, characterized in that: In step 1), the operating data includes real-time status information of vibration, current and speed components; the environmental data includes wind speed, wind direction, temperature, humidity and dust concentration environmental parameters; the historical maintenance data includes fault category, cause and maintenance record; the operating data in the multi-source data also obtains the temperature distribution of key components through thermal imaging equipment, combines image processing algorithms to detect abnormal hot spots, and optimizes fault prediction results.

3. The predictive maintenance method for a wind turbine according to claim 1, characterized in that: The local anomaly clustering algorithm in step 4) uses the sliding window technology to extract the mean, standard deviation and peak value of the time series data as clustering features. The clustering results of the anomaly points are matched with the historical fault data through the dynamic time warping algorithm.

4. The method for predictive maintenance of a wind turbine according to claim 1, characterized in that: Run the load in step 5) The relationship between wind speed and power generation is represented by the power curve model, which is calculated by real-time monitoring of power generation, rotation speed and wind speed. and parameters to adapt to changes in environmental conditions.

5. The method for predictive maintenance of a wind turbine according to claim 1, characterized in that: The dynamic weight of the environment in step 6) Calculated by the following formula: ;in, is the environmental dynamic weight, is the principal component weight of environmental characteristics, is the environmental characteristic value, is the number of environmental parameters; j is the environmental parameter index, indicating the jth environmental parameter.

6. A wind turbine predictive maintenance method according to claim 1, characterized in that: In step 7), the task priority of the maintenance plan is based on the health status score and environmental risk score, and wind turbines with low scores or high risks are given priority; the automatically generated maintenance plan combines dynamic priorities to optimize maintenance paths and task scheduling, thereby reducing operation and maintenance costs.

7. The method for predictive maintenance of a wind turbine according to claim 1, characterized in that: The deep learning prediction model in step 3) uses a multimodal Transformer model to process time series data, environmental parameters, and image data respectively, and the fusion results are used to output the health status score.

8. The method for predictive maintenance of a wind turbine according to claim 1, characterized in that: An edge computing node is deployed in the control cabinet of the wind turbine, and the edge node processes sensor data in real time and uploads abnormal detection results to the cloud.

9. A wind turbine predictive maintenance system, characterized in that: include: A data acquisition module for collecting operation data, environmental data and historical maintenance data of the wind turbine; Edge computing module, used to denoise the collected operation data and environmental data and extract key feature values; A multimodal deep learning prediction module that fuses time series data, environmental parameters, and image data to generate health status scores and potential fault categories for wind turbines; Dynamic maintenance optimization module, which is used to adjust the maintenance cycle in real time based on health status score, environmental dynamic influencing factors and operating load, and automatically generate maintenance plans including maintenance parts, tools and operation steps; Central monitoring and management platform for visual monitoring of wind turbines’ real-time operating status, health scores and maintenance plans, as well as allocation and scheduling of maintenance tasks.

10. A wind turbine predictive maintenance system according to claim 9, characterized in that: include: The multimodal deep learning prediction module includes: Time series data processing unit, used to extract time series characteristics of vibration, current and speed; Environmental data processing unit, used to analyze environmental characteristics such as wind speed, temperature and humidity, and dust concentration; An image data processing unit that processes thermal images of key components through a convolutional neural network to detect areas of abnormal temperature; The feature fusion and prediction unit fuses the features of the time series data, environmental data, and image data through a Transformer model, and outputs a health status score and a potential fault category.

Images