Wind turbine generator power characteristic analysis and low-efficiency unit automatic identification method

Abstract

The invention provides a wind turbine generator power characteristic analysis and low-efficiency unit automatic identification method. Relates to the field of wind turbine generator operation efficiency automatic identification, in particular to an automatic identification method for a part of wind turbine generators in a low-efficiency state for a long time caused by factors such as non-uniformity of wind resources, unit aging, equipment faults and wake effects. A physical information neural network (PINN) is adopted for SCADA data and data obtained after voiceprint and vibration signal feature fusion, data driving and a physical model are combined, and the method is used for creating a wind speed-power curve, so that the real operation characteristics of a wind turbine generator are better reflected. Comparing and analyzing the power curve of the wind turbine generator set, and effectively finding out a low-efficiency generator set with obvious power characteristic difference; the low-efficiency wind turbine generator set automatic identification algorithm based on fuzzy comprehensive evaluation is provided, comprehensive evaluation of the performance of the wind turbine generator set is realized, and the low-efficiency wind turbine generator set is successfully found out through the representation of a visual quantitative value and an evaluation grade.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind power generation technology, in particular to a wind turbine power characteristic analysis and low-efficiency unit automatic identification method. Specifically, the present application analyzes and identifies the performance of wind turbine power characteristics through collecting different sensor data and combining data processing technology, which can timely analyze and identify low-efficiency wind turbines and provide strong support for larger-scale wind power technology access to the power grid. BACKGROUND

[0002] In recent years, wind power generation technology has developed rapidly, and the problems of wind turbine power characteristic deterioration and operation performance decline have become increasingly prominent. This not only affects the overall power generation efficiency and economy of the wind farm, but also the intermittency and volatility of wind energy conversion, which will have a significant impact on the stability of the power system. Therefore, it is extremely important to carry out wind turbine power characteristic analysis and low-efficiency unit identification to help the economic and efficient operation of the wind farm.

[0003] At present, the research on wind turbine power characteristics has become a relatively mature field, and many scholars and researchers at home and abroad have conducted a lot of research on this problem and proposed some new research methods and technologies, but how to analyze and identify low-efficiency wind turbines still needs to be studied, and the analysis method only relies on a single data source, and cannot effectively integrate other data in the SCADA system. For example, problems such as sensor failure, communication interference, and equipment aging can cause SCADA data to have missing values, abnormal values, repeated data, and inconsistent problem data. The existence of these problem data seriously affects the accuracy and reliability of data analysis, which is not conducive to the operation management and decision support of the wind farm. Therefore, the existing method cannot effectively combine multiple data sources, and cannot comprehensively analyze the power characteristics of wind turbines and accurately identify low-efficiency units.

[0004] Power curve is essential for wind farm operation and economic benefit analysis, and can intuitively present the power output level of the unit. The power curve of the wind turbine in good operating condition should approach the theoretical expectation. In terms of operation practice, there is generally a gap between the actual operating power and the theoretical value when approaching the rated wind speed. Due to the intermittency and uncertainty of environmental conditions such as wind speed and wind direction in SCADA data, the power generation of wind turbines and various monitoring temperatures have volatility. In order to accurately evaluate the actual operating state of the wind turbine, a low-efficiency wind turbine automatic identification method needs to be designed to monitor and deeply analyze the operating data of the wind turbine in real time, and quickly and accurately identify the wind turbines with unsaturated operation in the wind farm. SUMMARY

[0005] The purpose of the present application is to provide a wind turbine power characteristic analysis and low-efficiency unit automatic identification method to solve the problems raised in the background art.

[0006] To solve the above technical problems, the present application provides the following technical solution: a wind turbine power characteristic analysis and low-efficiency unit automatic identification method. The automatic identification method of the present application mainly includes the following contents: first, the pre-processing of wind turbine SCADA data, then the power curve fitting, the performance evaluation and comparison, to realize the accurate discrimination of low-efficiency units. The method includes the following steps: S1 First, obtain the SCADA (Supervisory Control and Data Acquisition) data of the main parts of the wind turbine cabin (main shaft, gear box, generator, blade, etc.) and the voiceprint, vibration signal. Second, the SCADA data is cleaned and standardized, including denoising, standardization, missing data filling and outlier detection, to ensure the data quality and provide a reliable basis for subsequent analysis. Then, the voiceprint and vibration signal are denoised, standardized and feature extracted, mainly extracting frequency domain features, time domain features and statistical features to reflect the operating characteristics of the wind turbine under different working conditions and extract the key information in the data. Finally, the processed SCADA data and voiceprint, vibration signal are fused, mainly using weighted average, dimension reduction mechanism and other fusion methods to provide reliable data for subsequent analysis.

[0007] S2 The accurate data obtained by S1 feature fusion is used to adopt physical information neural network (PINN), combined with data-driven and physical model, to ensure physical consistency and uncertainty quantization, to obtain a dynamic power curve model. The model is the core parameter reflecting the power generation characteristics of the wind turbine, and expresses the mapping relationship between wind speed and power in a visual way. When analyzing the characteristics of the wind turbine, the power curve directly reflects the performance level of the unit, and is also the key benchmark for determining low-efficiency units.

[0008] S3 carries out performance evaluation and comparison of the power curve model obtained in S2 on multiple directions of the wind turbine, establishes an evaluation system, which can accurately reflect the performance of the wind farm at each time period and can accurately reflect the power of each wind turbine in the wind farm. The evaluation indexes of the wind turbine power generation, power standard deviation average, maximum power coefficient below the rated wind speed, etc. are determined, a decision matrix is formed by calculating the degradation degree corresponding to each index, and the membership degree of each index of the wind turbine and the wind farm with respect to each state grade is determined by combining the normal membership cloud generation algorithm and the principle of seeking the maximum gray correlation degree between the subjective and objective preference values and the decision value. The combination weight is obtained, the combination weight and the membership degree matrix are multiplied to perform fuzzy transformation, and the limit value matrix is used to evaluate the matrix after fuzzy transformation. Specifically, the differences between the wind turbines are clearly displayed in the form of intuitive numbers and evaluation levels, and the low-efficiency units are identified.

[0009] Further, the SCADA data in S1 mainly includes gear box oil temperature, gear box bearing temperature, average wind speed, generator bearing temperature, generator power, rotating speed, etc.; the voiceprints and vibration signals include main shaft vibration, gear box vibration, generator vibration, cabin voiceprint, blade voiceprint, etc.

[0010] Further, the frequency domain features, time domain features and statistical features of the voiceprints and vibration signals in S1 are extracted. The time domain features include mean, root mean square, kurtosis, etc., which are used to reflect the volatility, amplitude and change of the signal; the frequency domain features include frequency spectrum, main frequency, frequency bandwidth, etc., which can reveal the vibration mode, frequency component and potential fault frequency of the equipment; the statistical features include peak factor, frequency domain entropy, signal energy, etc., which can reveal the stability, complexity and potential fault mode of the equipment.

[0011] Further, the features from different signal sources (SCADA data and voiceprints and vibration signals) in S1 are fused to improve the accuracy and robustness of model prediction. Common fusion methods include weighted average and KPCA (Kernel Principal Component Analysis) dimension reduction. The weighted average assigns weights to different signal sources, combines their importance, and generates a comprehensive feature vector, which is suitable for the case where the contributions of different signal sources are uneven; and the KPCA dimension reduction maps the data from high-dimensional space to low-dimensional space through nonlinear mapping, extracts the most representative features, reduces redundant information, and improves the calculation efficiency. These fusion methods can effectively combine the advantages of multi-source data, provide comprehensive and accurate feature representation, and enhance the performance of subsequent analysis and prediction.

[0012] Further, the SCADA data and voiceprints, vibration signals in S1 are time-aligned to ensure consistency in the time dimension. This process is achieved by synchronizing the timestamps of different data sources, thereby generating time-aligned multi-source fusion data, providing a unified time reference for subsequent deep learning models.

[0013] Further, the dynamic wind speed and power curve model in S2 is constructed as follows: A physical information neural network (PINN) is used to combine data-driven and physical models, ensuring physical consistency and uncertainty quantification. The physical law (RMSE) of wind speed and power is used as a constraint term. A fully connected neural network is constructed with wind speed as input and predicted power as output. The model is embedded into the neural network loss function, and finally a wind speed and power curve model that is both accurate and consistent with physical common sense is obtained.

[0014] Further, in constructing the wind turbine evaluation model in S3, the proper selection of wind turbine power performance indicators directly affects the quality of the evaluation system. Based on the wind turbine operation standards and literature review analysis, the scatter plot of wind speed and power normal value is used to select the evaluation indicators, including power generation, mean of power standard deviation at rated wind speed, mean of power standard deviation above rated wind speed, and peak value of power coefficient at rated wind speed.

[0015] Further, in constructing the wind turbine evaluation model in S3, the relative degradation degree of evaluation indicators and the determination of state evaluation indicator membership are calculated. The degradation degree calculation of each indicator of the wind turbine can normalize the data of different dimensions and orders of magnitude to the interval [0, 1], which facilitates subsequent membership calculation and comprehensive evaluation. The membership function is created using the normal cloud model. By specifying the cloud parameters of different evaluation levels (excellent, good, medium, poor) corresponding to the expected value, entropy, and hyper-entropy, the membership of each indicator of each wind turbine for each evaluation level is obtained, and then a complete membership matrix is obtained.

[0016] The cloud model is defined by three characteristic parameters: expected value Ex, entropy En, and hyper-entropy He. The Ex value represents the spatial distribution center of the cloud model, the En value directly corresponds to the dispersion level of the cloud model, and the He value measures the dispersion degree of the entropy value. These parameters together determine the shape and distribution of the entire cloud model. In this paper, we set four evaluation levels: excellent, good, medium, and poor. And for each evaluation level, we set a corresponding cloud model characteristic parameter.

[0017] Furthermore, when constructing the wind turbine evaluation model described in S3, the calculation schemes for determining the weights of each indicator can be summarized into three types: subjective, objective, and a combination of both. Subjective weighting methods, which rely on expert experience, use qualitative expert judgment to determine weight values, such as expert scoring, the Cole method, and the Delphi method. Unlike subjective weighting methods based on expert judgment, objective weighting methods calculate the weights of each indicator by leveraging the correlation between indicators, such as the entropy weight method and the coefficient of variation method. Subjective weighting methods are highly subjective and easily influenced by human factors, but they can fully utilize professional judgment to improve the realism of the evaluation results. Objective weighting methods directly relate to the initial information content of each indicator and use mathematical methods to establish the weight foundation. We comprehensively utilize both subjective and objective weighting methods, employing grey relational analysis to calculate combined weights to promote the organic integration of subjective and objective evaluations.

[0018] Furthermore, during the identification of inefficient wind turbine units in S3, a fuzzy transformation is performed by multiplying the combined weights with the membership matrix. Then, a limit matrix is ​​used to evaluate the fuzzy transformed matrix, yielding a quantitative comprehensive evaluation result for each wind turbine unit. Furthermore, during the identification of inefficient wind turbines in S3, the quantitative value of the comprehensive evaluation result for each wind turbine is labeled with a corresponding evaluation level. This evaluation level visually displays the performance differences between wind turbines, thus successfully identifying inefficient turbines. This provides a scientific basis for the operation, maintenance, and optimized management of wind farms.

[0019] The beneficial effects of this patent are as follows: It employs a Physical Information Neural Network (PINN), combined with data-driven approaches and physical models, to create wind speed-power curves, thereby better reflecting the actual operating characteristics of wind turbines. By comparing and analyzing the power curves of wind turbines, it effectively identifies inefficient turbines with significantly different power characteristics. Furthermore, it proposes an automatic identification algorithm for inefficient wind turbines based on fuzzy comprehensive evaluation. This algorithm combines multiple comprehensive evaluation indicators, such as power generation, average power standard deviation, and maximum power coefficient below rated wind speed, to evaluate wind turbines. Through steps such as deterioration degree calculation, membership degree calculation, and combination weight calculation, it achieves a comprehensive evaluation of wind turbine performance, which is then reflected through intuitive quantitative values ​​and evaluation levels, successfully identifying inefficient wind turbines.

[0020] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are provided below. It should be understood that the above general description and the following detailed description are merely exemplary and do not limit the invention. Attached Figure Description

[0021] To more clearly illustrate the technical solutions implemented in this invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0022] Figure 1 The flowchart below illustrates the operational process of the wind turbine power characteristic analysis and inefficient unit automatic identification method described in this invention. Figure 2 This is a flowchart illustrating the data cleaning and preprocessing process described in this invention. Figure 3 This is a structural diagram of the wind power SCADA system described in this invention. Figure 4 The flowchart of the fuzzy comprehensive evaluation of wind turbine generators described in this invention is as follows. Detailed Implementation

[0023] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0024] Wind turbine power characteristic analysis and automatic identification method for inefficient units, such as Figure 1 As shown, through data collection, preprocessing, fusion, and analysis, combined with a wind turbine evaluation model, the performance differences between turbines can be accurately determined, and inefficient turbines can be precisely identified to promptly detect problems. This allows for appropriate maintenance of the wind farm, improving its power generation efficiency and economic benefits. The specific implementation plan is as follows: S1. First, acquire SCADA (Supervisory Control and Data Acquisition) data, as well as acoustic and vibration signals, from the major components of the wind turbine nacelle (main shaft, gearbox, generator, blades, etc.). Second, perform data cleaning and standardization on the SCADA data, specifically including denoising, standardization, missing data imputation, and outlier detection, to ensure data quality and provide a reliable foundation for subsequent analysis. Then, perform denoising, standardization, and feature extraction on the acoustic and vibration signals, primarily extracting frequency domain features, time domain features, and statistical features to reflect the operating characteristics of the wind turbine under different operating conditions and to extract key information from the data. Finally, perform feature fusion on the processed SCADA data, acoustic and vibration signals, mainly using weighted averaging and dimensionality reduction mechanisms to provide reliable data for subsequent analysis.

[0025] S2. Using the precise data obtained in S1, a physical information neural network is used to combine data-driven and physical models to ensure physical consistency and uncertainty quantification, resulting in a dynamic power curve model. This model is the core parameter reflecting the power generation characteristics of wind turbine units. It expresses the wind speed-power mapping relationship in a visual way. When conducting characteristic analysis of wind turbine units, the power curve directly reflects the performance level of the unit and is also a key benchmark for judging inefficient units.

[0026] S3. The power curve model obtained in S2 is used to evaluate and compare the performance of wind turbines in multiple directions, establishing an evaluation system that accurately reflects the performance of the wind farm at various times and the power of each wind turbine within the wind farm. Evaluation indicators such as wind turbine power generation, average power standard deviation, and maximum power coefficient below rated wind speed are determined. A decision matrix is ​​constructed by calculating the degradation degree corresponding to each indicator. A normal membership cloud generation algorithm is used to determine the membership degree of each indicator of the wind turbine and wind farm relative to each state level. The combined weights are calculated based on maximizing the grey relational degree between subjective and objective preference values ​​and decision values. The combined weights are multiplied by the membership degree matrix to perform a fuzzy transformation, and then a limit matrix is ​​used to evaluate the fuzzy transformed matrix. Specifically, this clearly displays the differences between wind turbines in an intuitive numerical and evaluation level format, and identifies inefficient turbines.

[0027] Furthermore, the sampling interval of the SCADA data in S1 is 1 minute, which mainly includes gearbox oil temperature, gearbox drive end bearing temperature, gearbox non-drive end bearing temperature, average wind speed, generator front bearing temperature, generator rear bearing temperature, generator stator winding temperature, average generator power, etc.; the sampling frequency of sound, vibration and speed signals is 12.8 kHz, which mainly includes: horizontal vibration of the main shaft front bearing, horizontal vibration of the main shaft rear bearing, axial vibration of the gearbox low-speed shaft, axial vibration of the gearbox high-speed shaft, vertical vibration of the generator drive end, vertical vibration of the generator non-drive end, nacelle sound pattern, blade sound pattern, etc.

[0028] Furthermore, in S1, DBSCAN is a density-based clustering algorithm that can automatically remove noise points and abnormal clusters. The density clustering algorithm uses two key neighborhood parameters. To assess the density of data points, and These represent the neighborhood radius and the minimum number of samples within the neighborhood, respectively. The neighborhood parameters help the DBSCAN algorithm distinguish between core points, boundary points, and outliers in the dataset. Assume the sample set... For any point Using Euclidean distance metric Define a radius as Clusters (called of (Neighborhood). For any point ,if ,but Considered the core point; if And at a certain core point of Within the neighborhood, A point is considered a boundary point; if a point does not meet the above two conditions, it is called a noise point.

[0029] Furthermore, frequency domain features, time domain features, and statistical features are extracted from the acoustic signature and vibration signal in S1. The time domain features include mean, root mean square, and kurtosis, which reflect the signal's volatility, amplitude, and variation. The frequency domain features include spectrum, dominant frequency, and bandwidth, which reveal the equipment's vibration modes, frequency components, and potential fault frequencies. The statistical features include peak factor, frequency domain entropy, and signal energy, which reveal the equipment's stability, complexity, and potential fault modes.

[0030] Furthermore, features from different signal sources (SCADA data, acoustic signatures, and vibration signals) in S1 are fused to improve the accuracy and robustness of model predictions. Common fusion methods include weighted averaging and KPCA dimensionality reduction. Weighted averaging generates a comprehensive feature vector by assigning weights to different signal sources and combining their respective importance; it is suitable for situations where the contributions of signal sources are uneven. KPCA dimensionality reduction, on the other hand, maps data from a high-dimensional space to a low-dimensional space through nonlinear mapping, extracting the most representative features, reducing redundant information, and improving computational efficiency. These fusion methods effectively combine the advantages of multi-source data, providing comprehensive and accurate feature representations and enhancing the performance of subsequent analysis and prediction.

[0031] Furthermore, the SCADA data fused from the features in S1, along with the voiceprint and vibration signals, undergo time alignment processing to ensure the consistency of these data in the time dimension. This process is achieved by synchronizing the timestamps of different data sources, thereby generating time-aligned multi-source fused data, providing a unified time reference for subsequent deep learning models.

[0032] Further, in step S2, the wind speed-power curve is obtained and the root mean square error (RMSE) is calculated. RMSE is an important indicator used to evaluate the goodness of fit of a model, showing the difference between the values ​​predicted by the model and the actual observed values. Its specific formula is as follows:

[0033] in, This represents the actual power output value. This represents the power output value predicted by the model. This represents the total number of data points. A lower RMSE indicates a smaller deviation between the model's calculated results and the actual results, suggesting a better model fit. RMSE can be used to compare and analyze the advantages and disadvantages of various modeling methods, allowing for the selection of the most appropriate modeling path.

[0034] Furthermore, in S2, a Physical Information Neural Network (PINN) is used, which combines data-driven and physical models to ensure physical consistency and uncertainty quantification. The physical laws of wind speed and power are used as constraints. A fully connected neural network is constructed, with wind speed as the input and predicted power as the output, and is embedded in the loss function of the neural network. Finally, a wind speed and power curve model that is both accurate and in line with physical common sense is obtained.

[0035] The physical laws employed are: For each interval, find all wind speed and power data points within that interval, and calculate the mean wind speed and mean power for these data points. The specific calculation formula is as follows:

[0036]

[0037] in, Indicates the first The average wind speed of the interval represents the first interval. Average power of each interval It is the first The number of data points within each interval and They are the first The first interval Each wind speed and power value is calculated. Based on this, the average wind speed and power for each interval can be obtained, serving as representative points for that interval, ultimately forming wind speed and power curves.

[0038] The data error term in the loss function is the root mean square error (RMSE) mentioned above.

[0039] The physical error term is the residual calculated from the wind speed and power curves.

[0040] The above design visually represents the performance level of wind turbine generators and the key benchmarks for inefficient units during characteristic analysis. Analysis of the power curves clearly shows which units output significantly less power than others at the same wind speed. This provides data support for optimizing wind farm operation and improving power generation efficiency. Accurately constructing and analyzing wind speed-power curves has a significant impact on the automatic identification of inefficient wind farm units and the overall operational level of the wind farm.

[0041] Furthermore, in constructing the fuzzy comprehensive evaluation model described in S3, firstly, a first-level and multi-level evaluation model is introduced. The operation process of the first-level fuzzy comprehensive evaluation can be divided into: The set of 'a' evaluation indicators is defined as the factor set U. These indicators collectively determine the influence of the evaluation object, and are generally denoted as:

[0042] in, For each influencing factor, =1,2,…,a.

[0043] Evaluation Set It is the set of possible evaluation results that evaluators may make for the object being evaluated; the evaluation set. It can be represented as:

[0044] In the formula, For various evaluation results, =1, 2, ..., b. For example: Excellent, Good, Average, Poor. These levels provide clear standards for evaluation, making the evaluation results more intuitive and easier to understand.

[0045] Membership degree representation factor Compared to the evaluation results The probability of [the event] is denoted as [the probability]. Its value ranges from [0,1], with a value closer to 1 indicating a higher membership degree. Therefore, for evaluating i factors, their membership vector relative to the evaluation set is:

[0046] in =1,2,…,b. The matrix formed by combining the membership vectors of each factor in the factor set U relative to the evaluation set W is called the membership matrix R, i.e.

[0047] The importance of each influencing factor relative to the evaluation set is called the weight ω, denoted as ω.

[0048] in, , .

[0049] Based on fuzzy transform,

[0050] In the formula, To assess the likelihood of pooling various rating results, =1,2,…,b.

[0051] Furthermore, in constructing the wind turbine evaluation model described in S3, the subjective and objective weight coefficients are determined by introducing the comparative scoring method and the entropy weight method, respectively. The combined weights are calculated by maximizing the grey relational degree between subjective and objective preferences and decision values. Based on the calculation system of relative degradation degree and normal membership cloud, the membership degree value of the evaluation set corresponding to each evaluation index is obtained. Following the above method (e.g.... Figure 3 As shown in the figure, a fuzzy comprehensive evaluation system for the power characteristics of the wind turbine was finally established. The relevant operation procedures are shown in [the figure]. Figure 4 During the project implementation phase, a fuzzy transformation is completed by multiplying the combined weights and membership matrices. Then, a limit matrix is ​​used to quantitatively evaluate the transformation results, thereby obtaining the final analysis results, namely:

[0052] in, For combined weights; This is the membership matrix; It is a fuzzy matrix; It is a limit matrix; This is the quantitative value of the final fuzzy comprehensive evaluation result.

[0053] The limit matrix specified for the evaluation is as follows:

[0054] In order to quantify values This is converted into specific evaluation levels, evaluation thresholds are set, and the evaluation results are obtained based on the relationship between these thresholds and quantitative values. When, the evaluation level is "Excellent"; when When the evaluation level is "good", when When, the evaluation level is "medium"; when At that time, the evaluation level was "poor". This grading method allows quantitative values ​​to intuitively reflect the performance level of the unit.

[0055] In the wind turbine evaluation model, based on the wind turbine operation standards and literature review analysis, a scatter plot of normal wind speed and power values ​​is used. The evaluation index system is constructed by selecting the power generation, the mean standard deviation of power at rated wind speed, the mean standard deviation of power at rated wind speed, and the peak power coefficient at rated wind speed.

[0056] Power generation is a direct indicator of the energy conversion efficiency of wind turbines. It can be obtained by calculating the power output of a wind turbine over a unit of time as the integral of time. In this paper, a time interval of 3600 seconds (1 hour) is used for calculation, and the specific formula is as follows:

[0057] This indicator reflects the energy output capacity of wind turbines within a specific time period and is an important basis for evaluating their economy and efficiency.

[0058] The power standard deviation needs to be calculated separately for the two wind speed ranges. The derivation of the power standard deviation for the ranges below and above the rated wind speed is the same for both, and the standard deviation calculation is completed using the Bessel formula. The calculation logic is as follows: After outlier handling, the first The power samples for each wind speed-power range are denoted as , and the power samples corresponding to different wind speed ranges are denoted as . , The number of power points within each wind speed-power interval. For the preprocessed dataset, wind speed is segmented at intervals of 0.5 m / s, and N is defined as the total number of intervals. The number of bin intervals not exceeding the rated wind speed is [number missing]. The number of bin intervals with wind speeds below the rated wind speed is The average wind speed and output power within each segment are calculated using the following formula:

[0059]

[0060] For the Power data set within a wind speed range The average power data is The residual is defined as the difference between the sample value and the mean, i.e. , ,…, Then the residual set is Then, according to Bessel's formula, the average standard deviation of power below the rated wind speed is:

[0061]

[0062] In the formula, For the first The number of power points within a wind speed range ; For residuals; For wind speeds below the rated wind speed The standard deviation of power in each interval.

[0063] The formula for calculating the average standard deviation of power above the rated wind speed is:

[0064]

[0065] in, For the first The number of power points within a wind speed range ; For residuals; For wind speeds above the rated wind speed The standard deviation of power in each interval.

[0066] Power coefficient It is a key indicator for measuring the energy conversion efficiency of wind turbine units. The formula for calculating the power coefficient within each bin interval is as follows:

[0067] In the formula, and For respectively the first Average wind speed and output power within each interval; and The first The j-th wind speed and power value within each interval.

[0068] The formula for calculating the maximum power coefficient below the rated wind speed is:

[0069] The following formula is used when calculating the degree of degradation: The calculation expressions for the two benefit parameters, AEP and the maximum value of the power coefficient at rated wind speed, are as follows:

[0070] For the two inferior indicators, the mean standard deviation of power at wind speeds below the rated speed and the average standard deviation of power at the rated wind speed, the following formula is used to calculate the degree of degradation:

[0071] in, This is the current measured value. This represents the normal range for the indicator parameters.

[0072] When determining the membership degree of the state evaluation index, a normal cloud model is used to create the membership function. The cloud model is defined by three feature parameters: expected value Ex, entropy En, and hyperentropy He. The value of Ex reflects the spatial distribution center of the cloud model, the value of En directly corresponds to the discreteness of the cloud model, and the value of He measures the dispersion of the entropy value. These parameters collectively determine the shape and distribution of the entire cloud model. In this paper, we set four evaluation levels: excellent, good, average, and poor. A corresponding cloud model feature parameter is set for each evaluation level. Based on the defined feature parameters, a cloud model for each evaluation level is generated.

[30] The cloud model consists of a series of cloud droplets ( Composed of, among which These are randomly generated values. yes The corresponding membership degree. The specific generation steps are as follows: Generate random values ​​of entropy ( ):according to and Generate a random entropy value. It follows a normal distribution: .

[0073] The number of cloud droplets generated ( ):according to and the generated entropy value ( Generate a random cloud droplet value. It follows a normal distribution: .

[0074] Calculate the membership degree of cloud droplets ( Based on cloud droplet values and expected value Calculate its membership degree The formula is as follows:

[0075] When determining the weights of each indicator, the original data is normalized to adjust each indicator data to [0,1] to eliminate the influence of unit differences. The normalization process uses the following formula:

[0076] in, Indicates the first The first sample Individual indicator values, and Let represent the minimum and maximum values ​​of the j-th indicator, respectively. Normalization ensures comparability between different indicators.

[0077] Next, the normalized data is normalized column-wise to obtain the proportion matrix. :

[0078] in, This represents the sample size. To avoid zero-value issues in logarithmic calculations, the zero values ​​in the proportion matrix are replaced with a local minimum. (like This ensures that no mathematical errors occur when calculating the logarithm. Then, the entropy value of each index is calculated based on the proportion matrix. :

[0079] in, It is a normalization factor that ensures the entropy value remains within the range of [0,1]. This indicates that the greater the numerical fluctuation of this indicator, the stronger its criticality in the overall evaluation.

[0080] Calculate the weight of each indicator based on the entropy value. The calculation formula is as follows:

[0081] in, Number of indicators. Weight. Reflects the first The importance of each indicator in the comprehensive evaluation.

[0082] When determining the weights of each indicator, the degradation data of the wind turbine units needs to be normalized to obtain a normalized degradation matrix. ,in This refers to the number of wind turbine units. This refers to the number of evaluation indicators. The purpose of normalization is to eliminate the influence of different dimensions between indicators and ensure data comparability. The specific formula is as follows:

[0083] in, Indicates the first The first wind turbine unit Individual indicator values, and They represent The minimum and maximum values ​​of each indicator.

[0084] Subjective weights obtained from the comparative rating method And the objective weights obtained by the entropy weight method Calculate subjective preference values Z ′ and objective preference value Z The specific calculation formula is as follows:

[0085]

[0086] Determine the grey relational degree between subjective and objective preference values ​​and decision values. and The grey relational coefficient reflects the decision-maker's subjective and objective preferences and decision values. The consistency level is calculated using the following formula from grey relational analysis:

[0087] In the formula, The resolution coefficient, Generally take ; Defined as the grey relational coefficient, it reflects the decision-maker's perception of the indicator. The degree of closeness between subjective and objective preferences and decision values This indicates that the decision-makers The consistency between subjective and objective preferences and decision values ​​is higher; =1,2,…,g, =1,2,…,4; This represents the difference matrix, which is the absolute difference between the decision value and the corresponding indicator preference value.

[0088]

[0089] The above formula can be used to calculate... Dehe .

[0090] To find the optimal combination weights, construct the following objective optimization model:

[0091] Through the above design, an automatic identification method for inefficient wind turbine units was developed. The differences between wind turbine units are clearly displayed in the form of intuitive numbers and evaluation levels, and inefficient units are identified. This method presents the comprehensive evaluation of each unit directly and clearly, which helps the operation and maintenance of wind farms.

[0092] 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 them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A method for analyzing the power characteristics of wind turbine units and automatically identifying inefficient units, characterized in that: include: S1: Obtain SCADA data and acoustic and vibration signals of major components (main shaft, gearbox, generator, blades, etc.) of wind turbine nacelle under inefficient conditions; perform data cleaning and standardization on the SCADA data using the DBSCAN algorithm; perform noise reduction, standardization and feature extraction on the acoustic and vibration signals; and perform feature fusion on the processed SCADA data and acoustic and vibration signals. S2: The fused data obtained by fusing the features of S1 is processed using a Physical Information Neural Network (PINN). By combining data-driven and physical models, physical consistency and uncertainty quantification are ensured, resulting in a dynamic power curve model that can be automatically updated to express the wind speed-power mapping relationship and is used to reflect the unit performance level. S3: Based on the dynamic power curve model of S2, an evaluation index system is established. Relying on the calculation system of relative degradation degree and normal membership cloud, the membership degree value of the evaluation set corresponding to each evaluation index is obtained. The subjective and objective weight coefficients are determined by comparing the scoring method and the entropy weight method respectively. By maximizing the gray relational degree between subjective and objective preferences and decision values, the combined weight is calculated. The combined weight and membership matrix are multiplied to complete the fuzzy transformation. Then, the transformation result is quantitatively evaluated through the limit matrix. Finally, the analysis results are obtained to achieve accurate identification of inefficient units.

2. The method for analyzing the power characteristics of wind turbine units and automatically identifying inefficient units according to claim 1, characterized in that: The SCADA data in step S1 includes gearbox oil temperature, gearbox bearing temperature, average wind speed, generator bearing temperature, generator power, and rotational speed; the acoustic and vibration signals include main shaft vibration, gearbox vibration, generator vibration, nacelle acoustic signal, and blade acoustic signal.

3. The method for analyzing the power characteristics of wind turbine units and automatically identifying inefficient units according to claim 1, characterized in that: Step S1 involves data cleaning and standardization of the SCADA data, including noise reduction, standardization, missing data imputation, and outlier detection; the features extracted from the voiceprint and vibration signals include time-domain features, frequency-domain features, and statistical features.

4. The method for analyzing the power characteristics of wind turbine units and automatically identifying inefficient units according to claim 1, characterized in that: In S1, the feature fusion method includes weighted averaging and KPCA (Kernel Principal Component Analysis) dimensionality reduction; the SCADA data after feature fusion, as well as the voiceprint and vibration signals, are time-aligned to provide a unified time reference.

5. The method for analyzing the power characteristics of wind turbine units and automatically identifying inefficient units according to claim 1 is characterized as follows: In S2, a Physical Information Neural Network (PINN) is used, combined with data-driven and physical models, to ensure physical consistency and uncertainty quantification. The physical law of wind speed and power (RMSE) is used as a constraint term. A fully connected neural network is constructed, with wind speed as the input and predicted power as the output, and is embedded in the loss function of the neural network to finally obtain a dynamic power curve model that can be automatically updated.

6. The method for analyzing the power characteristics of wind turbine units and automatically identifying inefficient units according to claim 1, characterized in that: Step S3 selects the power generation, the mean standard deviation of power below the rated wind speed, the mean standard deviation of power above the rated wind speed, and the peak value of the power coefficient below the rated wind speed to form an evaluation index system.

7. The method for power characteristic analysis and automatic identification of inefficient wind turbine units according to claim 1, characterized in that: In S3, the deterioration calculation of each evaluation index of the wind turbine can normalize the index data of different dimensions and magnitudes to the [0,1] interval.

8. The method for analyzing the power characteristics of wind turbine units and automatically identifying inefficient units according to claim 1, characterized in that, In S3, a membership function is created using a normal cloud model. By specifying the expected values, entropy, and hyperentropy of cloud parameters corresponding to different evaluation levels (excellent, good, medium, poor), the degradation degree of each evaluated indicator is transformed into the corresponding membership value.

9. The method for power characteristic analysis and automatic identification of inefficient wind turbine units according to claim 1, characterized in that, In step S3, a multiplication operation is performed on the combined weights and membership matrices to complete the fuzzy transformation. Then, the transformation result is quantitatively evaluated through the limit matrix to obtain the final analysis result.

10. The method for analyzing the power characteristics of wind turbine units and automatically identifying inefficient units according to claim 1, characterized in that, In S3, the differences between wind turbine units are clearly displayed in the form of intuitive numbers and evaluation levels, and inefficient units are successfully identified.

Images