Wind turbine generator operation performance detection method and system based on power generation working condition

Abstract

The invention relates to the technical field of electric power detection, and discloses a wind turbine generator operation performance detection method and system based on a power generation working condition, and the method comprises the following steps: dynamically adjusting the data collection frequency and range according to the risk level; real-time operation data and real-time image data of the wind turbine generator are obtained, and working condition division is carried out; fusing the real-time operation data and the real-time image features to obtain real-time fusion features; based on a working condition division result, combining a dynamic weight distribution strategy to adjust the weights of the operation data and the image features in the real-time fusion features; and predicting a real-time power generation performance index of the wind turbine generator by using a pre-constructed multi-modal deep learning model, and analyzing the operation performance of the wind turbine generator based on the predicted real-time power generation performance index. The method overcomes the defects that a traditional method is single in data, insufficient in feature fusion and lagged in risk perception, and provides important technical support for efficient operation and maintenance and intelligent development of the wind turbine generator.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power detection, and in particular to a method and system for detecting the operating performance of a wind turbine generator set based on power generation conditions. Background Art

[0002] As a clean and renewable energy, wind power generation has attracted widespread attention worldwide due to its environmental friendliness and resource sustainability, and plays an important role in optimizing the energy structure and achieving carbon reduction targets. In recent years, with the rapid growth of global demand for green energy, wind power installed capacity has continued to expand, and wind turbines have gradually become an indispensable part of the power system. However, the operating performance of wind turbines is affected by a variety of complex factors, including changes in the external natural environment and the mechanical and electrical conditions inside the unit.

[0003] During operation, wind turbines need to adapt to changing environmental conditions, such as changes in wind speed, fluctuations in temperature, changes in air humidity, and extreme weather conditions such as rain, snow, and freezing. These environmental factors directly affect the power generation efficiency and operational stability of the unit. For example, a sharp change in wind speed may cause load fluctuations on the blades, while ice accumulation in a low-temperature environment may increase mechanical resistance and further reduce operational performance. In addition, sandstorms and corrosive sea breezes in the natural environment may also cause wear on key components of the unit, such as blades and bearings, shortening the life of the equipment.

[0004] Not only that, the operation of the unit itself is also full of challenges. Wind turbines usually contain complex mechanical structures and electrical systems, such as gearboxes, generators, and power conversion equipment. During long-term operation, these components may experience performance degradation due to fatigue, aging, or improper maintenance. For example, wear of the gearbox may cause a decrease in transmission efficiency, while heat dissipation problems of the generator may cause overheating or even failure. In addition, the vibration and noise generated during the operation of the wind turbine may also indicate abnormalities in internal components or potential failure hazards.

[0005] Faced with such a complex operating environment and diverse influencing factors, traditional wind turbine performance detection methods seem to be unable to cope with the situation. Most existing detection methods rely on single operating data analysis, such as power output, speed, vibration amplitude, etc. However, these methods usually ignore the interaction with the external environment of the unit, and fail to make comprehensive use of other types of data such as image features. This limitation of data singleness makes it difficult to accurately diagnose the operating status and potential faults of the unit under complex working conditions.

[0006] Especially under high-risk conditions, relying solely on operational data analysis may not capture critical information in a timely manner. For example, abnormal phenomena such as blade cracks or surface ice accumulation, although they may seriously affect power generation performance, cannot be directly discovered through conventional operating parameter detection. Therefore, the analysis method that relies solely on a single data source shows obvious limitations when faced with multimodal data fusion and complex risk environments, and it is difficult to meet the needs of efficient operation and reliable diagnosis of wind turbines. Therefore, a more comprehensive, intelligent and efficient operation performance detection method is urgently needed to meet the diverse challenges in the operation of modern wind turbines. Summary of the invention

[0007] In view of the problems in the related art, the present invention proposes a method and system for detecting the operating performance of a wind turbine generator set based on power generation conditions, so as to overcome the above technical problems existing in the existing related art.

[0008] To this end, the specific technical solution adopted by the present invention is as follows:

[0009] According to one aspect of the present invention, a method for detecting the operating performance of a wind turbine generator system based on power generation conditions is provided, comprising the following steps:

[0010] S1. Based on historical operation data and environmental data, predict the risk level of the wind turbine at the current moment, and dynamically adjust the frequency and scope of data collection according to the risk level;

[0011] S2. Acquire the real-time operation data and real-time image data of the wind turbine generator set according to the adjusted data collection frequency and range, and divide the current operation status of the wind turbine generator set into working conditions according to the real-time operation data;

[0012] S3, using image processing technology to extract features from real-time image data, and fusing real-time operation data with real-time image features to obtain real-time fusion features;

[0013] S4. Based on the working condition division results, the weights of the operating data and image features in the real-time fusion features are adjusted in combination with the dynamic weight allocation strategy;

[0014] S5. Use the pre-built multimodal deep learning model in combination with the adjusted real-time fusion features to predict the real-time power generation performance indicators of the wind turbine, and analyze the operating performance of the wind turbine based on the predicted real-time power generation performance indicators.

[0015] Furthermore, the method of predicting the risk level of the wind turbine at the current moment based on the historical operation data and environmental data, and dynamically adjusting the data collection frequency and range according to the risk level includes the following steps:

[0016] S11. Obtain historical operation data, environmental data, and fault annotation data of the wind turbine generator set, and preprocess the acquired historical data; construct and train a risk prediction model based on the preprocessed historical data;

[0017] S12. Obtain the operating data and environmental data of the wind turbine at the previous moment, and use the trained risk prediction model to predict the risk level of the wind turbine at the current moment; dynamically adjust the data collection frequency and range according to the risk level of the wind turbine at the current moment.

[0018] Furthermore, dynamically adjusting the data collection frequency and range according to the risk level of the wind turbine at the current moment includes:

[0019] When the risk level is low risk, the acquisition method is low-frequency acquisition, and the acquisition scope includes key indicators in the operating data and images acquired at the first resolution;

[0020] When the risk level is medium risk, the acquisition method is medium frequency acquisition, and the acquisition range includes all indicators in the operating data and images acquired at the second resolution;

[0021] When the risk level is high risk, the acquisition method is high-frequency acquisition, and the acquisition scope includes all indicators in the operating data, images acquired at the third resolution, and thermal imaging images of key components;

[0022] Among them, the first resolution, the second resolution and the third resolution increase in sequence.

[0023] Furthermore, the method of extracting features from real-time image data using image processing technology and fusing the real-time operation data with the real-time image features to obtain real-time fusion features includes the following steps:

[0024] S31, cleaning and standardizing the real-time collected image data, extracting key features of the wind turbine operating status from the real-time image based on image processing technology, and preprocessing the real-time operating data and extracting key dynamic features;

[0025] S32, performing time synchronization and space matching processing on the image features and the operation data features, and directly splicing the image features and the operation data features to form a real-time fusion feature vector.

[0026] Furthermore, the method of adjusting the weights of the operating data and image features in the real-time fusion features based on the working condition division results and combining the dynamic weight allocation strategy includes the following steps:

[0027] S41, determining an initial weight allocation strategy according to the working condition division result, and presetting a default weight ratio of real-time operation data and real-time image features according to the working condition type;

[0028] S42, calculating the importance scores of the real-time operation data and the real-time image features respectively based on the change rate, volatility and abnormal intensity of the real-time operation data and the real-time image features;

[0029] S43. Dynamically modify the default weights according to the importance scores of the real-time operation data and the real-time image features, and use the modified weights to fuse the operation data features and the image features.

[0030] Furthermore, the default weight ratios of the real-time operation data and the real-time image features are preset according to the working condition type, including:

[0031] When the working condition type is a low-risk working condition, the real-time operating parameters are dominant, and the weight of the real-time operating parameters is greater than the weight of the real-time image features;

[0032] When the working condition type is a medium-risk working condition, the real-time operating parameters and real-time image features are equally important, and the weight of the real-time operating parameters is equal to the weight of the real-time image features;

[0033] When the working condition type is a high-risk working condition, the real-time image features are dominant, and the weight of the real-time operation parameters is less than the weight of the real-time image features.

[0034] Furthermore, the calculation formula for the importance score of real-time operation data is:

[0035]

[0036] The calculation formula for the importance score of real-time image features is:

[0037]

[0038] In the formula, S r represents the importance score of real-time operation data, S i represents the importance score of the real-time image feature, ω b ,ω v ,ω a Respectively represent the weights of the rate of change, volatility, and abnormal intensity, R r 、V r , A r They represent the rate of change, volatility and abnormal intensity of real-time operation data, R i 、V i , A i They represent the rate of change, volatility and abnormal intensity of real-time image features, respectively, and x t ,y t Respectively represent the running data value and image feature value at the current moment, x t-1 ,y t-1They represent the running data value and image feature value of the previous moment, ε represents the number to avoid the denominator being zero, k represents the length of the sliding window, and x n ,y n They represent the nth running data value and image feature value in the sliding window, μ and ν represent the mean of the running data and image features in the sliding window, respectively. b ,y b Represent the normal baseline values ​​of operating data and image features respectively.

[0039] Furthermore, the calculation formula of the weight after real-time operation data correction is:

[0040] ω r ′=ω r +λ·S r

[0041] The calculation formula of the weight after real-time image feature correction is:

[0042] ω i ′=ω i +λ·S i

[0043] The feature fusion formula is:

[0044] F=ω i ′·F r +ω i ′·F i

[0045] In the formula, ω r ′ represents the weight after real-time operation data correction, ω r represents the default weight of real-time running data, ω i ′ represents the weight after real-time image feature correction, ω i represents the default weight of the real-time image feature, λ represents the correction coefficient, F represents the adjusted real-time fusion feature, and F r Indicates the operating data characteristics, F i Represents image features.

[0046] Furthermore, the analysis of the operating performance of the wind turbine generator system based on the predicted real-time power generation performance index includes the following steps:

[0047] Obtain a preset threshold range of the power generation performance indicator, and determine whether the predicted real-time power generation performance indicator meets the preset power generation performance indicator threshold range. If so, determine that the operating performance of the wind turbine is normal; if not, determine that the operating performance of the wind turbine is abnormal.

[0048] According to another aspect of the present invention, there is provided a wind turbine operating performance detection system based on power generation conditions, comprising a risk level analysis module, an operating condition division module, a feature fusion module, a weight allocation module, a performance index prediction module and an operating performance analysis module;

[0049] The risk level analysis module is used to predict the risk level of the wind turbine at the current moment based on historical operation data and environmental data, and dynamically adjust the data collection frequency and range according to the risk level;

[0050] The operating condition classification module is used to obtain the real-time operating data and real-time image data of the wind turbine according to the adjusted data acquisition frequency and range, and to classify the current operating state of the wind turbine according to the real-time operating data;

[0051] The feature fusion module is used to extract features from real-time image data using image processing technology, and fuse the real-time operation data with the real-time image features to obtain real-time fusion features;

[0052] The weight allocation module is used to adjust the weights of the operating data and image features in the real-time fusion features based on the working condition division results and in combination with the dynamic weight allocation strategy;

[0053] The performance indicator prediction module is used to predict the real-time power generation performance indicator of the wind turbine using a pre-built multi-modal deep learning model combined with the adjusted real-time fusion features;

[0054] The operation performance analysis module is used to analyze the operation performance of the wind turbine generator set based on real-time power generation performance indicators, locate the cause of the fault in combination with image features, and generate an operation status detection report.

[0055] The beneficial effects of the present invention are:

[0056] 1) The present invention realizes high-precision detection of the operating performance of wind turbines by dynamically adjusting data collection strategies, introducing multimodal feature fusion, dynamically allocating weights based on operating risk levels, and predicting performance based on deep learning. This method overcomes the defects of traditional methods such as single data, insufficient feature fusion, and delayed risk perception, and provides important technical support for the efficient operation and maintenance and intelligent development of wind turbines.

[0057] 2) The present invention can dynamically adjust the frequency and range of data collection according to the risk level of the wind turbine at the current moment. Low-frequency collection reduces redundant data under low-risk conditions; high-frequency collection covers key indicators and image features under high-risk conditions to ensure comprehensive capture of abnormal conditions.

[0058] 3) The present invention constructs a real-time fusion feature vector through time synchronization and spatial matching of operating data and image features, and combines it with a dynamic weight allocation strategy to achieve deep fusion between features, thereby improving the performance detection accuracy under complex working conditions.

[0059] 4) Based on the real-time working condition classification results, the present invention dynamically calculates the feature importance score by combining the change rate, volatility and abnormal intensity, and dynamically corrects the weights of the operating data and image features. This method focuses on operating data when the risk is low and on image features when the risk is high, thus improving the flexibility of multimodal feature utilization.

[0060] 5) The present invention uses a multimodal deep learning model to predict the real-time power generation performance indicators of wind turbines, combines preset thresholds to analyze the operating performance, locates the cause of the fault through image features, and generates a visual operating status detection report, thereby improving the operation and maintenance efficiency and fault diagnosis capabilities of wind turbines. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0062] Figure 1 The present invention is a flowchart of a method for detecting the operating performance of a wind turbine generator system based on power generation conditions according to an embodiment of the present invention. DETAILED DESCRIPTION

[0063] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in the field should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0064] According to an embodiment of the present invention, a method and system for detecting the operating performance of a wind turbine generator system based on power generation conditions are provided.

[0065] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to one embodiment of the present invention, a method for detecting the operating performance of a wind turbine generator system based on power generation conditions is provided, comprising the following steps:

[0066] S1. Based on historical operation data and environmental data, predict the risk level of the wind turbine at the current moment, and dynamically adjust the frequency and scope of data collection according to the risk level;

[0067] The method of predicting the risk level of the wind turbine at the current moment based on the historical operation data and environmental data, and dynamically adjusting the frequency and scope of data collection according to the risk level includes the following steps:

[0068] S11. Obtain historical operation data, environmental data and fault annotation data of wind turbines, and preprocess the acquired historical data; construct and train a risk prediction model based on the preprocessed historical data; specifically include:

[0069] 1. Historical data preparation and preprocessing:

[0070] Objective: Prepare the operating and environmental data of wind turbines to build the basis for risk prediction models.

[0071] 1) Data collection: Collect historical operating parameter data (wind speed, rotation speed, vibration intensity, power output, etc.) and historical environmental data (external conditions such as temperature, humidity, and atmospheric pressure), as well as fault annotation data, including abnormal events that occurred in the past and their risk levels (such as low risk, medium risk, and high risk).

[0072] 2) Data cleaning: fill in missing values ​​(such as interpolation method) and remove outliers (such as 3 times standard deviation method).

[0073] 3) Feature Engineering: Construct time series features, such as the moving average and rate of change of operating parameters, and extract interactive features, such as the ratio of wind speed to power output and the correlation between vibration and temperature.

[0074] 4) Data partitioning: Divide the data into training set, validation set and test set.

[0075] 2. Constructing a risk prediction model:

[0076] Objective: Use historical data to train the model and predict the risk level at the current moment in real time.

[0077] 1) Select the model:

[0078] Time series models (such as LSTM or GRU): suitable for processing continuous time series data.

[0079] Machine learning models (such as random forest and XGBoost): suitable for processing multi-dimensional feature data.

[0080] Hybrid models: Combining the advantages of time series and classification models.

[0081] 2) Model input and output:

[0082] Input: multi-dimensional time series features (such as wind speed, power, vibration, etc. in the past 10 minutes), external environment features (such as temperature, humidity).

[0083] Output: Risk level at the current moment (low, medium, high).

[0084] 3) Training and optimization:

[0085] Loss function: cross entropy for classification or mean squared error for regression.

[0086] Optimization method: such as Adam optimizer.

[0087] Model evaluation indicators: Accuracy, Recall and F1 score.

[0088] S12. Obtain the operating data and environmental data of the wind turbine at the previous moment, and use a trained risk prediction model (such as a time series model or a deep learning model) to predict the risk level of the wind turbine at the current moment; dynamically adjust the data collection frequency and range according to the risk level of the wind turbine at the current moment.

[0089] Specifically, according to the risk level of the wind turbine at the current moment, the frequency and scope of data collection are dynamically adjusted, including:

[0090] 1) Low risk (normal working conditions):

[0091] Collection frequency: low frequency collection (such as once every 10 minutes).

[0092] Collection range:

[0093] Operating parameters: wind speed, power, rotation speed and other key indicators.

[0094] Images: Low resolution or periodic acquisition.

[0095] 2) Medium risk (potential abnormal conditions):

[0096] Collection frequency: Medium frequency collection (such as once every 1-2 minutes).

[0097] Collection range:

[0098] Operating parameters: comprehensive collection (vibration, temperature and other additional data).

[0099] Image: Improve resolution, increase angles or close-ups.

[0100] 3) High risk (abnormal working conditions):

[0101] Acquisition frequency: high-frequency acquisition (such as continuous acquisition in seconds).

[0102] Collection range:

[0103] Operating parameters: full range collection, including rotation speed, wind speed, vibration, temperature, etc.

[0104] Image: Call high-definition equipment to collect multi-angle and panoramic images; dispatch drones for detailed inspection.

[0105] Thermal imaging: real-time acquisition of thermal imaging data of key components.

[0106] In addition, step S1 further comprises:

[0107] Feedback on test results: Compare the real-time collected data with the risk prediction results; update the parameters of the risk prediction model to improve prediction accuracy.

[0108] Strategy optimization: Adjust the frequency and scope of low, medium and high risk collection strategies; optimize features and data collection patterns for specific risks (such as ice accumulation and sandstorms).

[0109] S2. Acquire the real-time operation data and real-time image data of the wind turbine generator set according to the adjusted data collection frequency and range, and divide the current operation status of the wind turbine generator set into working conditions according to the real-time operation data;

[0110] Specifically, the collection of real-time operating data and real-time image data of wind turbines is achieved through operating data sensors (such as wind speed sensors, vibration monitors) and image acquisition equipment, such as using cameras or thermal imaging devices to capture real-time images of wind turbine blades and equipment, and performing noise reduction, feature extraction (identifying key areas such as cracks and ice accumulation areas) and quantitative feature (extracting key indicators in the image, such as crack length and ice accumulation area) processing on the images.

[0111] S3, using image processing technology to extract features from real-time image data, and fusing real-time operation data with real-time image features to obtain real-time fusion features;

[0112] The method of extracting features from real-time image data using image processing technology and fusing real-time operation data with real-time image features to obtain real-time fusion features includes the following steps:

[0113] S31. Clean and standardize the real-time collected image data, extract key features of the wind turbine operating status from the real-time image based on image processing technology, and pre-process the real-time operating data and extract key dynamic features, specifically including:

[0114] 1. Preprocessing of real-time image data: Clean and standardize the real-time collected image data to prepare for subsequent feature extraction.

[0115] 1) Denoising: Use median filtering, Gaussian filtering or bilateral filtering to remove noise from the image to ensure clear data. For thermal imaging data, apply pseudo color mapping (such as Jet or Parula color mapping) to enhance contrast.

[0116] 2) Image correction: Geometric correction, correcting image distortion caused by the camera shooting angle; Light compensation, adjusting uneven lighting through histogram equalization.

[0117] 3) Resolution adjustment: Dynamically adjust image resolution according to risk level:

[0118] Low risk: low resolution processing to reduce computational burden.

[0119] High Stakes: Maintain high resolution to ensure detail integrity of critical features.

[0120] 4) ROI (Region of Interest) extraction: Extract the key parts of the wind turbine (such as blades, gearboxes); use edge detection (Canny, Sobel) or deep learning models to automatically identify key areas.

[0121] 2. Preprocessing and feature extraction of real-time operation data: Extract key dynamic features from real-time operation data and match and fuse them with image features.

[0122] 1) Run data preprocessing:

[0123] Normalization: Standardize the operating parameters (such as wind speed, vibration intensity, and power output).

[0124] Time series construction: Use sliding windows to generate time series features (such as the rate of change of wind speed in the past 10 minutes).

[0125] 2) Run data feature extraction:

[0126] Statistical features: Calculate the mean, maximum value, rate of change, etc. of real-time data.

[0127] Time series features: Use deep learning models such as LSTM and GRU to extract embedded features of time series.

[0128] Physical characteristics: Use formulas to calculate specific physical quantities (such as power coefficient, tip speed ratio).

[0129] Result: Output the operating data feature vector (such as wind speed change rate, power fluctuation, vibration signal spectrum characteristics, etc.).

[0130] S32, performing time synchronization and space matching processing on the image features and the operation data features, and directly splicing the image features and the operation data features to form a real-time fusion feature vector.

[0131] S4. Based on the working condition division results, the weights of the operating data and image features in the real-time fusion features are adjusted in combination with the dynamic weight allocation strategy;

[0132] The method of adjusting the weights of the operating data and image features in the real-time fusion features based on the working condition division results and combining the dynamic weight allocation strategy includes the following steps:

[0133] S41. Determine an initial weight allocation strategy based on the working condition classification result (low risk, medium risk, high risk or specific abnormality), and preset a default weight ratio of real-time operation data and real-time image features based on the working condition type;

[0134] Specifically, the default weight ratios of real-time operation data and real-time image features are preset according to the working condition type, including:

[0135] When the working condition type is a low-risk working condition, the real-time operating parameters are dominant, and the weight of the real-time operating parameters is greater than the weight of the real-time image features. Preferably, the weight of the real-time operating parameters is 0.7, and the weight of the real-time image features is 0.3;

[0136] When the working condition type is a medium-risk working condition, the real-time operating parameters and the real-time image features are equally important, and the weight of the real-time operating parameters is equal to the weight of the real-time image features. Preferably, the weight of the real-time operating parameters is 0.5, and the weight of the real-time image features is 0.5;

[0137] When the working condition type is a high-risk working condition, the real-time image feature is dominant, and the weight of the real-time operation parameter is less than the weight of the real-time image feature. Preferably, the weight of the real-time operation parameter is 0.3, and the weight of the real-time image feature is 0.7;

[0138] In addition, when the operating condition type is a specific abnormal condition, further adjustments are made according to the abnormal type:

[0139] Ice accumulation: mainly based on image features (thermal imaging), the weight of real-time operation parameters is 0.4, and the weight of real-time image features is 0.6.

[0140] Abnormal vibration: mainly based on operating data (vibration signal), the weight of real-time operating parameters is 0.6, and the weight of real-time image features is 0.4.

[0141] S42, calculating the importance scores of the real-time operation data and the real-time image features respectively based on the change rate, volatility and abnormal intensity of the real-time operation data and the real-time image features;

[0142] Specifically, the calculation formula for the importance score of real-time operation data is:

[0143]

[0144] The calculation formula for the importance score of real-time image features is:

[0145]

[0146] In the formula, S r Indicates the importance score of real-time running data, ranging from [0,1], S i represents the importance score of the real-time image feature, ranging from [0,1], ω b ,ω v ,ω a Respectively represent the weights of the rate of change, volatility, and abnormal intensity, R r 、V r , A r They represent the rate of change of real-time operating data (i.e., the short-term change range of operating data), volatility (i.e., the fluctuation range of operating data, such as vibration intensity fluctuation) and abnormal intensity (i.e., the degree of deviation of operating data from the normal state), R i 、V i , A i They represent the rate of change of real-time image features (i.e., the short-term change amplitude of image features), volatility (i.e., the local change amplitude of image features), and abnormal intensity (i.e., the deviation degree of image features, such as crack area and ice accumulation area, from the normal state), respectively. t ,y t Respectively represent the running data value and image feature value at the current moment, x t-1 ,y t-1 They represent the running data value and image feature value of the previous moment respectively, and ε represents the number to avoid the denominator being zero, and its value is 10 -6 , k represents the length of the sliding window, x n ,y n They represent the nth running data value and image feature value in the sliding window, μ and ν represent the mean of the running data and image features in the sliding window, respectively. b ,y b Represent the normal baseline values ​​of operating data and image features respectively.

[0147] S43. Dynamically modify the default weights according to the importance scores of the real-time operation data and the real-time image features, and use the modified weights to fuse the operation data features and the image features.

[0148] Specifically, the calculation formula of the weight after real-time operation data correction is:

[0149] ω r ′=ω r +λ·S r

[0150] The calculation formula of the weight after real-time image feature correction is:

[0151] ω i ′=ω i +λ·S i

[0152] The feature fusion formula is:

[0153] F=ω i ′·F r +ω i ′·F i

[0154] In the formula, ω r ′ represents the weight after real-time operation data correction, ω r represents the default weight of real-time running data, ω i ′ represents the weight after real-time image feature correction, ω i represents the default weight of the real-time image feature, λ represents the correction coefficient, F represents the adjusted real-time fusion feature, and F r Indicates the operating data characteristics, F i Represents image features.

[0155] In addition, the feature fusion step also includes:

[0156] Feature dimension alignment: If the dimensions of the running data features and image features are different, you can first use dimensionality reduction techniques (such as PCA) or dimensionality expansion (such as fully connected layers) to align the features.

[0157] Fusion feature performance evaluation: Verify the effect of fusion features on target tasks (such as working condition classification and anomaly detection) and use indicators such as accuracy and recall to evaluate.

[0158] Feedback optimization: If the fusion effect is not good, adjust the dynamic correction parameter λ or redesign the initial weights.

[0159] S5. Use the pre-built multimodal deep learning model in combination with the adjusted real-time fusion features to predict the real-time power generation performance indicators of the wind turbine, and analyze the operating performance of the wind turbine based on the predicted real-time power generation performance indicators.

[0160] The analysis of the operating performance of the wind turbine generator system based on the predicted real-time power generation performance index comprises the following steps:

[0161] Obtain a preset threshold range of the power generation performance indicator, and determine whether the predicted real-time power generation performance indicator meets the preset power generation performance indicator threshold range. If so, determine that the operating performance of the wind turbine is normal; if not, determine that the operating performance of the wind turbine is abnormal.

[0162] Specifically, the setting of the preset threshold range of the power generation performance indicator is determined according to the design specifications and historical data of the unit.

[0163] According to another embodiment of the present invention, there is provided a wind turbine operating performance detection system based on power generation conditions, comprising a risk level analysis module, an operating condition division module, a feature fusion module, a weight allocation module, a performance index prediction module and an operating performance analysis module;

[0164] The risk level analysis module is used to predict the risk level of the wind turbine at the current moment based on historical operation data and environmental data, and dynamically adjust the data collection frequency and range according to the risk level;

[0165] The operating condition classification module is used to obtain the real-time operating data and real-time image data of the wind turbine according to the adjusted data acquisition frequency and range, and to classify the current operating state of the wind turbine according to the real-time operating data;

[0166] The feature fusion module is used to extract features from real-time image data using image processing technology, and fuse the real-time operation data with the real-time image features to obtain real-time fusion features;

[0167] The weight allocation module is used to adjust the weights of the operating data and image features in the real-time fusion features based on the working condition division results and in combination with the dynamic weight allocation strategy;

[0168] The performance indicator prediction module is used to predict the real-time power generation performance indicator of the wind turbine using a pre-built multi-modal deep learning model combined with the adjusted real-time fusion features;

[0169] The operation performance analysis module is used to analyze the operation performance of the wind turbine generator set based on real-time power generation performance indicators, locate the cause of the fault in combination with image features, and generate an operation status detection report.

[0170] In summary, with the help of the above technical solutions of the present invention, the present invention realizes high-precision detection of the operating performance of wind turbines by dynamically adjusting the data acquisition strategy, introducing multimodal feature fusion, dynamically allocating weights in combination with the risk level of working conditions, and predicting performance based on deep learning. This method overcomes the defects of single data, insufficient feature fusion, and delayed risk perception in traditional methods, and provides important technical support for the efficient operation and maintenance and intelligent development of wind turbines. In addition, the present invention can dynamically adjust the data acquisition frequency and range according to the risk level of the wind turbine at the current moment. Low-frequency acquisition reduces redundant data under low-risk working conditions; high-frequency acquisition covers key indicators and image features under high-risk working conditions to ensure comprehensive capture of abnormal states. In addition, the present invention constructs a real-time fusion feature vector by time synchronization and spatial matching of operating data and image features, and combines a dynamic weight allocation strategy to achieve deep fusion between features, thereby improving the performance detection accuracy under complex working conditions. In addition, based on the real-time working condition division results, the present invention dynamically calculates the feature importance score in combination with the change rate, volatility, and abnormal intensity, and dynamically corrects the weights of operating data and image features. This method focuses on operating data when the risk is low and on image features when the risk is high, which improves the flexibility of using multimodal features. In addition, the present invention uses a multimodal deep learning model to predict the real-time power generation performance indicators of wind turbines, analyzes the operating performance in combination with preset thresholds, locates the cause of the fault through image features, and generates a visual operating status detection report, which improves the operation and maintenance efficiency and fault diagnosis capabilities of wind turbines.

[0171] 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, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting the operating performance of a wind turbine generator system based on power generation conditions, characterized in that: The following steps are involved: S1. Based on historical operation data and environmental data, predict the risk level of the wind turbine at the current moment, and dynamically adjust the frequency and scope of data collection according to the risk level; S2. Acquire the real-time operation data and real-time image data of the wind turbine generator set according to the adjusted data collection frequency and range, and divide the current operation status of the wind turbine generator set into working conditions according to the real-time operation data; S3, using image processing technology to extract features from real-time image data, and fusing real-time operation data with real-time image features to obtain real-time fusion features; S4. Based on the working condition division results, the weights of the operating data and image features in the real-time fusion features are adjusted in combination with the dynamic weight allocation strategy; S5. Use the pre-built multimodal deep learning model in combination with the adjusted real-time fusion features to predict the real-time power generation performance indicators of the wind turbine, and analyze the operating performance of the wind turbine based on the predicted real-time power generation performance indicators.

2. A method for detecting the operating performance of a wind turbine generator system based on power generation conditions according to claim 1, characterized in that: The method of predicting the risk level of the wind turbine at the current moment based on the historical operation data and environmental data, and dynamically adjusting the data collection frequency and range according to the risk level includes the following steps: S11, obtaining historical operation data, environmental data and fault marking data of the wind turbine generator set, and preprocessing the obtained historical data; Build and train risk prediction models based on preprocessed historical data; S12, obtaining the operation data and environmental data of the wind turbine generator set at the previous moment, and using the trained risk prediction model to predict the risk level of the wind turbine generator set at the current moment; Dynamically adjust the data collection frequency and scope based on the current risk level of the wind turbine.

3. A method for detecting the operating performance of a wind turbine generator system based on power generation conditions according to claim 2, characterized in that: The dynamically adjusting the data collection frequency and range according to the risk level of the wind turbine at the current moment includes: When the risk level is low risk, the acquisition method is low-frequency acquisition, and the acquisition scope includes key indicators in the operating data and images acquired at the first resolution; When the risk level is medium risk, the acquisition method is medium frequency acquisition, and the acquisition range includes all indicators in the operating data and images acquired at the second resolution; When the risk level is high risk, the acquisition method is high-frequency acquisition, and the acquisition scope includes all indicators in the operating data, images acquired at the third resolution, and thermal imaging images of key components; Among them, the first resolution, the second resolution and the third resolution increase in sequence.

4. A method for detecting the operating performance of a wind turbine generator system based on power generation conditions according to claim 1, characterized in that: The method of extracting features from real-time image data using image processing technology and fusing real-time operation data with real-time image features to obtain real-time fusion features includes the following steps: S31, cleaning and standardizing the real-time collected image data, extracting key features of the wind turbine operating status from the real-time image based on image processing technology, and preprocessing the real-time operating data and extracting key dynamic features; S32, performing time synchronization and space matching processing on the image features and the operation data features, and directly splicing the image features and the operation data features to form a real-time fusion feature vector.

5. The method for detecting the operating performance of a wind turbine generator system based on power generation conditions according to claim 1, characterized in that: The method of adjusting the weights of the operating data and image features in the real-time fusion features based on the working condition division results and combining the dynamic weight allocation strategy includes the following steps: S41, determining an initial weight allocation strategy according to the working condition division result, and presetting a default weight ratio of real-time operation data and real-time image features according to the working condition type; S42, calculating the importance scores of the real-time operation data and the real-time image features respectively based on the change rate, volatility and abnormal intensity of the real-time operation data and the real-time image features; S43. Dynamically modify the default weights according to the importance scores of the real-time operation data and the real-time image features, and use the modified weights to fuse the operation data features and the image features.

6. A method for detecting the operating performance of a wind turbine generator system based on power generation conditions according to claim 5, characterized in that: The default weight ratios of real-time operation data and real-time image features are preset according to the working condition type, including: When the working condition type is a low-risk working condition, the real-time operating parameters are dominant, and the weight of the real-time operating parameters is greater than the weight of the real-time image features; When the working condition type is a medium-risk working condition, the real-time operating parameters and real-time image features are equally important, and the weight of the real-time operating parameters is equal to the weight of the real-time image features; When the working condition type is a high-risk working condition, the real-time image features are dominant, and the weight of the real-time operation parameters is less than the weight of the real-time image features.

7. A method for detecting the operating performance of a wind turbine generator system based on power generation conditions according to claim 5, characterized in that: The calculation formula for the importance score of real-time operation data is: The calculation formula for the importance score of real-time image features is: In the formula, S r represents the importance score of real-time operation data, S i represents the importance score of the real-time image feature, ω b ,ω v ,ω a Respectively represent the weights of the rate of change, volatility, and abnormal intensity, R r 、V r , A r They represent the rate of change, volatility and abnormal intensity of real-time operation data, R i 、V i , A i They represent the rate of change, volatility and abnormal intensity of real-time image features, respectively, and x t ,y t Respectively represent the running data value and image feature value at the current moment, x t-1 ,y t-1 They represent the running data value and image feature value of the previous moment, ε represents the number to avoid the denominator being zero, k represents the length of the sliding window, and x n ,y n They represent the nth running data value and image feature value in the sliding window, μ and ν represent the mean of the running data and image features in the sliding window, respectively. b ,y b Represent the normal baseline values ​​of operating data and image features respectively.

8. A method for detecting the operating performance of a wind turbine generator system based on power generation conditions according to claim 7, characterized in that: The calculation formula of the weight after real-time operation data correction is: oh r ′=ω r +λ·S r The calculation formula of the weight after real-time image feature correction is: oh i ′=ω i +λ·S i The feature fusion formula is: F=ω i ′·F r +oh i ′·F i In the formula, ω r ′ represents the weight after real-time operation data correction, ω r represents the default weight of real-time running data, ω i ′ represents the weight after real-time image feature correction, ω i represents the default weight of the real-time image feature, λ represents the correction coefficient, F represents the adjusted real-time fusion feature, and F r Indicates the operating data characteristics, F i Represents image features.

9. A method for detecting the operating performance of a wind turbine generator system based on power generation conditions according to claim 1, characterized in that: The analysis of the operating performance of the wind turbine generator set based on the predicted real-time power generation performance index comprises the following steps: Obtain a preset threshold range of the power generation performance indicator, and determine whether the predicted real-time power generation performance indicator meets the preset power generation performance indicator threshold range. If so, determine that the operating performance of the wind turbine is normal; if not, determine that the operating performance of the wind turbine is abnormal.

10. A wind turbine operating performance detection system based on power generation conditions, used to implement the steps of the wind turbine operating performance detection method based on power generation conditions as described in any one of claims 1 to 9, characterized in that: It includes risk level analysis module, working condition division module, feature fusion module, weight allocation module, performance index prediction module and operation performance analysis module; The risk level analysis module is used to predict the risk level of the wind turbine at the current moment based on historical operation data and environmental data, and dynamically adjust the data collection frequency and range according to the risk level; The operating condition classification module is used to obtain the real-time operating data and real-time image data of the wind turbine according to the adjusted data acquisition frequency and range, and to classify the current operating state of the wind turbine according to the real-time operating data; The feature fusion module is used to extract features from real-time image data using image processing technology, and fuse the real-time operation data with the real-time image features to obtain real-time fusion features; The weight allocation module is used to adjust the weights of the operating data and image features in the real-time fusion features based on the working condition division results and in combination with the dynamic weight allocation strategy; The performance indicator prediction module is used to predict the real-time power generation performance indicator of the wind turbine using a pre-built multi-modal deep learning model combined with the adjusted real-time fusion features; The operation performance analysis module is used to analyze the operation performance of the wind turbine generator set based on real-time power generation performance indicators, locate the cause of the fault in combination with image features, and generate an operation status detection report.

Images