Fan blade icing diagnosis method considering sample imbalance condition

Abstract

The application provides a fan blade icing diagnosis method considering sample imbalance conditions, comprising the following steps: S1. Obtain wind turbine operation data as initial sample data, and pretreat the initial sample data; S2. Process the unbalanced initial sample data to form balanced samples; S3. Extract icing identification features from the balanced samples; S4. Input the icing identification features into a LightGBM model to train the LightGBM model; S5. Obtain operation data of a to-be-tested wind turbine as real-time samples, pretreat the real-time samples, extract features, and then input the extracted features into the trained LightGBM model to obtain an icing grade. Through the above method, when a small amount of samples exist in the wind turbine, the sample data is unbalancedly processed, so that the final balanced data is obtained, and the accuracy and reliability of the fan blade icing diagnosis are effectively improved, and accurate data support is provided for subsequent deicing operation and maintenance.

Description

A method for diagnosing wind turbine blade icing considering imbalanced sample conditions Technical Field

[0001] This invention relates to a method for icing diagnosis, and more particularly to a method for icing diagnosis of wind turbine blades that takes into account sample imbalance conditions. Background Technology

[0002] Wind power generation, as an important component of clean energy, has been widely used. However, in cold regions, icing of wind turbine blades is a common problem, seriously affecting the operating efficiency and safety of wind turbine units.

[0003] Blade icing not only reduces power generation efficiency and increases blade load, but can also cause equipment damage due to imbalance. Current mainstream methods for diagnosing blade icing primarily rely on human experience, physical modeling, or machine learning techniques based on balanced datasets. However, in practical applications, the number of blade icing events is usually far fewer than the number of non-icing events, resulting in a severely imbalanced data sample. This data characteristic makes it difficult for traditional algorithms to effectively capture the characteristics of the minority class of samples, leading to reduced reliability of diagnostic results.

[0004] Therefore, in order to solve the above-mentioned technical problems, it is urgent to propose a new technical approach. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a wind turbine blade icing diagnosis method that takes into account sample imbalance conditions. When there are a small number of samples in the wind turbine, the sample data is processed to remove the imbalance, thereby obtaining the final balanced data. This effectively improves the accuracy and reliability of wind turbine blade icing diagnosis and provides accurate data support for subsequent de-icing and maintenance.

[0006] This invention provides a method for diagnosing icing on wind turbine blades considering sample imbalance, comprising the following steps:

[0007] S1. Obtain wind turbine operating data as initial sample data and preprocess the initial sample data;

[0008] S2. Process the imbalanced initial sample data to form a balanced sample;

[0009] S3. Extract ice-covering identification features from balanced samples;

[0010] S4. Input the icing recognition features into the LightGBM model to train the LightGBM model;

[0011] S5. Implement the acquisition of the operating data of the wind turbine under test as real-time samples, perform feature extraction after preprocessing the real-time samples, and then input the extracted features into the trained LightGBM model to obtain the icing level.

[0012] Furthermore, in step S2, the process of processing the imbalanced sample data to form a balanced sample specifically includes:

[0013] Select m samples from the initial sample data as the baseline sample X. m :

[0014] X m =(x1,x2, … ,x i ,…,x m );

[0015] Calculate the Euclidean distance between the baseline sample and the remaining sample data in the initial sample data, and select the k samples with the smallest distance in the remaining sample data as the nearest neighbor samples;

[0016] Randomly select M nearest neighbor samples from the k nearest neighbor samples as auxiliary samples Y. mk :

[0017] Y mk =(y m1 ,y m2 , … ,y mj ,…,y mk );

[0018] Reconstruct new samples as balanced samples:

[0019] X new =x i +γ(y mj -x i ), where γ is a random coefficient between [0,1].

[0020] Furthermore, the initial sample data and real-time sample data include temperature, humidity, wind speed, and vibration signals of the wind turbine blades.

[0021] Furthermore, preprocessing of the initial and real-time samples includes:

[0022] Box plot detection method is used to remove outliers from the sample data;

[0023] Linear interpolation was used to process the difference in the sample data after outliers were removed.

[0024] The interpolated sample data is then normalized using Z-score.

[0025]

[0026] Where z represents the normalized sample data, x represents the sample data before normalization, μ is the mean of the sample data, and σ represents the standard deviation of the sample data.

[0027] Furthermore, the extraction of icing identification features includes:

[0028] Determine the mean, standard deviation, and peak value of the sample data;

[0029] Principal component analysis was used to obtain the final icing characteristics from the mean, standard deviation, and peak values ​​of the sample data.

[0030] The beneficial effects of this invention are as follows: When there are a small number of samples in a wind turbine, the sample data is processed to obtain the final balanced data, thereby effectively improving the accuracy and reliability of wind turbine blade icing diagnosis and providing accurate data support for subsequent de-icing and maintenance. Attached Figure Description

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0032] Figure 1 is a flowchart of the present invention. Detailed Implementation

[0033] The present invention will now be described in further detail:

[0034] This invention provides a method for diagnosing icing on wind turbine blades considering sample imbalance, comprising the following steps:

[0035] S1. Obtain wind turbine operating data as initial sample data and preprocess the initial sample data;

[0036] S2. Process the imbalanced initial sample data to form a balanced sample;

[0037] S3. Extract ice-covering identification features from balanced samples;

[0038] S4. Input the icing recognition features into the LightGBM model to train the LightGBM model;

[0039] S5. Implement the acquisition of operating data of the wind turbine under test as real-time samples, and perform feature extraction after preprocessing the real-time samples. Then, input the extracted features into the trained LightGBM model to obtain the icing level. Through the above method, when there are a small number of samples in the wind turbine, the sample data is imbalanced to obtain the final balanced data, thereby effectively improving the accuracy and reliability of wind turbine blade icing diagnosis and providing accurate data support for subsequent de-icing operation and maintenance.

[0040] In this embodiment, step S2, which involves processing the imbalanced sample data to form a balanced sample, specifically includes:

[0041] Select m samples from the initial sample data as the baseline sample X. m :

[0042] X m =(x1,x2,…,x i ,…,x m );

[0043] Calculate the Euclidean distance between the baseline sample and the remaining sample data in the initial sample data, and select the k samples with the smallest distance in the remaining sample data as the nearest neighbor samples;

[0044] Randomly select M nearest neighbor samples from the k nearest neighbor samples as auxiliary samples Y. mk :

[0045] Y mk =(y m1 ,y m2 ,…,y mj ,…,y mk );

[0046] Reconstruct new samples as balanced samples:

[0047] X new =x i +γ(y mj -x i ), where γ is a random coefficient between [0,1]. By using the above method, the balance of sample data can be effectively ensured, thereby ensuring the accuracy of the model parameter update during the training of the prediction model LightGBM, and thus ensuring the accuracy of the final prediction result.

[0048] In this embodiment, the initial sample data and real-time sample data include temperature, humidity, wind speed, and vibration signals of wind turbine blades.

[0049] In this embodiment, the preprocessing of the initial samples and real-time samples includes:

[0050] Box plot detection method is used to remove outliers from the sample data;

[0051] Linear interpolation was used to process the difference in the sample data after outliers were removed.

[0052] The interpolated sample data is then normalized using Z-score.

[0053]

[0054] Where z represents the normalized sample data, x represents the sample data before normalization, μ is the mean of the sample data, and σ represents the standard deviation of the sample data; the box plot detection method and the linear interpolation method are both existing technologies, and their details will not be elaborated here.

[0055] In this embodiment, the extraction of icing identification features includes:

[0056] Determine the mean, standard deviation, and peak value of the sample data; where:

[0057] The formulas for mean and standard deviation are:

[0058]

[0059] Principal component analysis was used to obtain the final icing characteristics from the mean, standard deviation, and peak values ​​of the sample data. Principal component analysis is an existing technique and will not be elaborated upon here.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for diagnosing icing on wind turbine blades considering sample imbalance, characterized in that: Includes the following steps: S1. Obtain wind turbine operating data as initial sample data and preprocess the initial sample data; The initial sample data and real-time sample data include temperature, humidity, wind speed, and vibration signals of wind turbine blades. Preprocessing of the initial and real-time samples includes: removing outliers using box plot detection; performing interpolation on the outlier-removed sample data; and performing Z-score normalization on the interpolated sample data. ;in, This represents the sample data after normalization. This represents the sample data before normalization. The mean of the sample data. S1 represents the standard deviation of the sample data; S2. Processing the imbalanced initial sample data to form a balanced sample; Processing the imbalanced sample data to form a balanced sample specifically includes: selecting m samples from the initial sample data as the reference sample. ： Calculate the Euclidean distance between the baseline sample and the remaining samples in the initial sample data, and select the k samples with the smallest distance from the remaining samples as nearest neighbors; randomly select M nearest neighbors from the k nearest neighbors as auxiliary samples. ： Reconstruct new samples as balanced samples: ,in, for S3. Extract icing identification features from the balanced samples; S4. Input the icing identification features into the LightGBM model to train the LightGBM model; S5. Obtain the operating data of the wind turbine under test as real-time samples, preprocess the real-time samples and extract features, and then input the extracted features into the trained LightGBM model to obtain the icing level.

2. The wind turbine blade icing diagnosis method considering sample imbalance conditions according to claim 1, characterized in that: The extraction of icing identification features includes: determining the mean, standard deviation, and peak value of the sample data; and using principal component analysis to obtain the final icing features from the mean, standard deviation, and peak value of the sample data.

Images