Load reconstruction method for wind generating set

By fitting Johnson distribution and modeling p-box variables, and combining double-layer sampling and slice inverse regression techniques, a sparse multinomial chaotic expansion model is constructed. This solves the problems of high cost and low accuracy in wind turbine load monitoring, and enables rapid and accurate load calculation and confidence prediction, thereby improving the safety and stability of wind turbines.

CN118934502APending Publication Date: 2024-11-12HUNAN SHIYOU ELECTRIC PUBLIC
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411308501.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-19
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

Existing methods for monitoring and reconfiguring wind turbine loads are costly and inaccurate, making it difficult to effectively cope with complex operating conditions and uncertainties, leading to difficulties in safe operation and maintenance.

Method used

Johnson distribution fitting technique and p-box variable modeling are used, combined with double-layer sampling and slice inverse regression techniques, to construct a sparse multinomial chaotic expansion model for rapid calculation and uncertainty analysis of wind turbine loads.

Benefits of technology

It improves the accuracy and reliability of load calculation, provides prediction results at different confidence levels, enhances the operational safety and stability of wind turbine generators, and solves the computational efficiency problem of traditional methods in high-dimensional input spaces.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118934502A_ABST
    Figure CN118934502A_ABST
Patent Text Reader

Abstract

The invention relates to a load reconstruction method for a wind generating set, and is applied to the technical field of online monitoring of wind generating sets. Comprising the following steps: acquiring fan blade operation data from an SCADA system, and identifying to obtain uncertain factors; the method comprises the following steps: performing classification and modeling on uncertainty input in operation of a wind turbine based on SCADA data, and constructing a p-box variable; carrying out random sampling on the generated p-box variable to obtain sample data with interval and probability uncertainty; on the basis of an SIR technology, dimension reduction processing is carried out on uncertainty input data randomly selected through double-layer sampling; constructing a slice inverse regression technology-sparse polynomial chaos expansion model and performing accuracy analysis; and according to the optimization model, carrying out uncertainty analysis on the load of the wind generating set to obtain a confidence prediction result of the load. According to the invention, the load of the wind generating set can be comprehensively monitored and evaluated, and the operation safety and stability are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of online monitoring of wind turbine generator sets, and in particular to a method for reconstructing a load of a wind turbine generator set. Background Art

[0002] In the field of online monitoring of wind turbines, the load changes of wind turbines are crucial to the safe operation of the units. Existing load monitoring and reconstruction methods have not been widely used due to their high cost and complex equipment. In actual operation, the load will be affected by a variety of uncertain factors, which makes the regular changes of the load very complicated. Therefore, monitoring and analysis of the load is of great significance to ensure the safe operation and maintenance of the unit.

[0003] The existing technology mainly uses two load reconstruction methods: one is the prediction method that combines measured loads with machine learning methods; the other is the simulation calculation method that combines physical models and SCADA data. The former faces the problems of data scarcity and high cost, while the latter, although it has a large amount of SCADA system operation data, lacks accurate operation conditions and operation boundary data (such as wind profiles and turbulent wind speeds) due to the complexity of the operating conditions, resulting in the reconstruction accuracy being affected by uncertainty.

[0004] Therefore, it is necessary to develop a new type of wind turbine load reconstruction method that can effectively realize fast load calculation and consider multi-source uncertainty. Summary of the invention

[0005] The purpose of the present invention is to provide a wind turbine load reconstruction method to achieve rapid calculation of the wind turbine load and predict the confidence of the result, and to apply it to the online monitoring and operation and maintenance of the wind turbine.

[0006] According to a first aspect of the present invention, a method for reconfiguring a load of a wind turbine generator set is provided, characterized in that it comprises the following steps:

[0007] Obtaining wind turbine blade operation data from a SCADA system, and identifying uncertainty factors that affect the load of a wind turbine generator set during wind turbine operation based on the operation data;

[0008] By using Johnson distribution fitting technology, the uncertainty input in wind turbine operation is classified and modeled based on SCADA data, and p-box variables are constructed to accurately describe the impact of uncertainty factors on wind turbine loads;

[0009] Based on the double-layer sampling technology, the generated p-box variables are randomly sampled to obtain sample data with interval and probability uncertainty, and the sample data includes wind speed, wind direction angle, temperature, turbulent wind speed and wind profile;

[0010] Based on the slice inverse regression technology, the uncertain input data randomly selected by double-layer sampling is subjected to dimensionality reduction processing, retaining its key features, thereby obtaining the reduced-dimensional data;

[0011] Based on the uncertain input data with key features retained after dimension reduction, the optimal polynomial basis function is selected to construct a sparse polynomial chaos expansion model.

[0012] Based on the constructed sparse polynomial chaos expansion model, its coefficients are expressed as quadratic polynomial function forms of interval distribution parameters, and the polynomial coefficients are recalculated to construct a slice inverse regression technology-sparse polynomial chaos expansion model;

[0013] Considering MSE and R 2 The numerical value of is used to analyze the accuracy of the slice inverse regression technology-sparse polynomial chaos expansion model, and the best balance between the accuracy and computational efficiency of the model is optimized by adjusting the polynomial order p and the number of basis functions of the model.

[0014] According to the optimized slice inverse regression technology-sparse polynomial chaos expansion model, uncertainty analysis is performed on the load of the wind turbine generator set to obtain a confidence prediction result of the load.

[0015] According to the above aspects and any possible implementation, there is further provided an implementation, wherein the operation data includes wind speed, wind direction, temperature, and rotation speed;

[0016] The uncertain factors affecting the load of the wind turbine generator set include steady-state wind speed, turbulent wind speed, wind direction angle and temperature.

[0017] According to the above aspects and any possible implementation, an implementation is further provided, wherein the generated p-box variables are randomly sampled based on the double-layer sampling technology to obtain sample data with interval and probability uncertainty, wherein the sample data includes wind speed, wind direction angle, temperature, turbulent wind speed and wind profile, including:

[0018] Generated using the Latin Hypercube Sampling (LHS) method Outer layer samples θ (w) =(θ1 (w) ,θ2 (w) ,...,θ n (w) ),in The subscripts 1, 2, ... n represent the number of uncertainty input variables, and each sample represents the interval distribution parameter of the uncertainty input;

[0019] For each sample θ in the outer layer (w), the sparse interpolation collocation point method is used to select valid sample points and generate inner layer samples Ξ. Specifically, the sparse interpolation of the sparse interpolation value collocation point method is used:

[0020]

[0021] in, represents the tensor product, ι represents the configuration level, |i|=i1+i2+...i n Represents the sum of multiple indicators, Indicates the i k dimensional collocation points, which can be obtained by Gaussian quadrature or Clenshaw-Curtis rule;

[0022] Then select M sample points X from the sparse interpolation point set Ξ (v) =(X1 (v) ,X2 (v) ,...,X n (v) ), where v = 1, 2, ..., M is used for linear regression calculation; in order to obtain stable results, the number of sample points M must be greater than or equal to twice the polynomial coefficient, that is, M ≥ 2q. .

[0023] According to the above aspects and any possible implementation, an implementation is further provided, wherein the key calculation using the slice inverse regression technique is as follows:

[0024] The sample data generated by double-layer sampling is input as uncertain input parameters into the gas-elastic-mechanical-electrical coupling model to calculate the output response y;

[0025] Standardize each input variable so that its mean is zero and its variance is 1, and calculate the standardized input matrix X (v) The covariance matrix of

[0026]

[0027] X1 (i) For X1 (v) input variables, is the mean vector of the input data, M is the number of samples;

[0028] Divide the output response y into H slices, and the number of samples in each slice is n h (v) , for each slice, calculate the slice mean m of the input variable h (v) :

[0029]

[0030] Among them, s h (v) represents the sample set belonging to the hth slice, j represents the slice S h The sample index of

[0031] Compute the covariance matrix Γ using the slice mean (v) , which reflects the differences between slices:

[0032]

[0033] By solving the generalized eigenvalue problem Γβ=λξβ, the eigenvalue λ and eigenvector β are solved;

[0034] Among them, the eigenvector β is the effective dimensional regression direction of the slice inverse regression method, reflecting the most representative low-dimensional feature direction;

[0035] Select the first q eigenvectors β1, β2, ..., β1 with the largest eigenvalues. q , the high-dimensional input data X (v) Project to low-dimensional space to get the reduced-dimensional data matrix Z (v) =(Z1 (v) ,Z2 (v) ,...,Z q (v) ), where each Z i are the independent variables:

[0036] Z=XB

[0037] Where B = [β1,β2,...,β q ] is the matrix consisting of selected eigenvectors.

[0038] According to the above aspects and any possible implementation, an implementation is further provided, wherein the constructed sparse polynomial chaos expansion model is expressed as:

[0039]

[0040] Among them, Y(t) represents the change of wind turbine load with time t, Z represents the input variable after processing based on SIR algorithm, Indicates that X i The most relevant polynomial basis functions, Represents polynomial coefficients.

[0041] According to the above aspects and any possible implementation, an implementation is further provided, and the slice inverse regression technology-sparse polynomial chaos expansion model finally constructed is:

[0042]

[0043] in, represents the polynomial coefficients, Represents Z i The most relevant polynomial basis function, i.e., the optimal polynomial basis function, q is the polynomial expansion order, n is the total number of uncertain input variables, θ represents the outer layer distribution variable, and a i,0 、b i,j d i,j are the coefficients of the quadratic polynomial.

[0044] According to the above aspects and any possible implementation, a further implementation is provided, wherein the comprehensive consideration of MSE and R 2 The numerical value of is used to analyze the accuracy of the slice inverse regression technology-sparse polynomial chaos expansion model. By adjusting the polynomial order p and the number of basis functions of the model, the best balance between the accuracy and computational efficiency of the model is optimized, including:

[0045] calculating a mean value of the first output load;

[0046] The uncertainty input data after dimension reduction generated by the double-layer sampling strategy and slice inverse regression technology is input into the slice inverse regression technology-sparse polynomial chaos expansion model to obtain the second output vector

[0047] ^

[0048] Load Y;

[0049] The mean square error MSE and determination coefficient R are calculated based on the following formulas: 2 :

[0050]

[0051] in, is the second output load, Y i is the first output load, is the mean of the first output loads.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] (1) Traditional methods often rely on accurate input data, but in actual operation, it is usually difficult to obtain accurate data due to the complexity of various factors such as environmental conditions and equipment status. The present invention successfully solves the problem of difficult to accurately quantify input parameters in actual simulation analysis by introducing uncertainty modeling technology and using Johnson distribution fitting technology and p-box modeling. By combining slice inverse regression (SIR) technology and sparse chaotic polynomial expansion (SPCE) technology, the present invention can effectively reduce the dimension and retain the key features of the input data, while quantifying these uncertainty factors, significantly improving the accuracy and reliability of load calculation;

[0054] (2) During the load reconstruction process, the present invention can provide confidence intervals of prediction results at different confidence levels, so that decision makers can make more reliable operation and maintenance decisions according to different confidence levels in practical applications, thereby effectively improving the safety and stability of wind turbine operation;

[0055] (3) The present invention successfully solves the "curse of dimensionality" problem often encountered by traditional PCE methods when dealing with high-dimensional input spaces by combining slice inverse regression (SIR) and sparse chaotic polynomial expansion (SPCE) technology. With the increase of input variables, traditional methods usually lead to a sharp increase in the number of polynomial terms and excessive computational burden. Through the dimensionality reduction of SIR technology and the selection of key features, the present invention constructs a more compact and efficient proxy model, which significantly improves the computational efficiency and predictive ability of the model. Therefore, the method of the present invention breaks through the limitations of traditional PCE and provides an innovative and efficient solution for uncertainty analysis in the wind power field.

[0056] It should be understood that the contents described in the summary of the invention are not intended to limit the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The above and other features, advantages and aspects of the embodiments of the present invention will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are provided for a better understanding of the present invention and do not constitute a limitation of the present invention. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, wherein:

[0058] Figure 1 A flow chart of a method for reconfiguring a load of a wind turbine generator set according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0061] Please refer to the attached Figure 1 The present invention provides a method for reconfiguring a load of a wind turbine generator set, comprising the following steps:

[0062] S101. Acquire operation data from a SCADA system of a wind turbine generator set, and identify uncertainty factors that affect the load of the wind turbine generator set during wind turbine operation.

[0063] The SCADA (Supervisory Control and Data Acquisition) system of a wind turbine is a real-time monitoring system used to collect and store data such as the operating status and environmental parameters of the wind turbine. These data are the basic input for load reconstruction analysis.

[0064] The operating data of this embodiment are selected from wind speed, wind direction, temperature, and rotation speed data, and the obtained uncertainty factors include steady-state wind speed, turbulent wind speed, wind direction angle, temperature, and wind profile factors.

[0065] S102. By using Johnson distribution fitting technology, the uncertain inputs in wind turbine operation (such as wind speed, wind direction angle, temperature, turbulent wind speed and wind profile, etc.) are classified and modeled based on SCADA data, and p-box variables are constructed to accurately describe the impact of uncertain factors on wind turbine loads.

[0066] Since wind turbines work in the complex and changeable atmospheric boundary layer all year round, factors such as the operating wind speed and direction, turbulence, and geometric manufacturing errors will affect the aerodynamic characteristics of the wind turbine blades, thereby causing fluctuations in the wind turbine aerodynamic force and affecting the load of the wind turbine generator set.

[0067] In order to accurately quantify these uncertainties, especially non-parametric data (such as wind profiles and other interval data), the present invention adopts Johnson distribution fitting, which can flexibly adapt to different types of data distributions, including SB (bounded distribution), SU (unbounded distribution), SN (normal distribution) and SL distribution (lognormal distribution). Through the quantile matching method, the Johnson distribution parameters (γ, ζ, δ, λ) are fitted to establish an accurate probability distribution model for the uncertain input data.

[0068] Determining uncertainty inputs is one of the key steps in conducting uncertainty analysis of wind turbine loads. These factors need to be used as input parameters for statistical modeling. The steps are as follows:

[0069] Extract uncertainty inputs (e.g., steady-state wind speed, turbulent wind speed, wind direction angle, temperature, wind profile) from SCADA datasets and classify them into interval data (e.g., wind profile) and probabilistic data (e.g., wind speed, temperature);

[0070] Select the appropriate Johnson distribution type based on the data characteristics: SB is used for bounded interval data, SU is used for unbounded data, SL is used for lognormal distribution data, and SN is used for data close to normal distribution;

[0071] Normalize the data to fit the selected Johnson distribution type: SB distribution data is normalized to the interval 0 to 1, SU distribution data is unbounded normalized, SL distribution data is logarithmically transformed, and SN distribution data usually does not need further normalization;

[0072] By standardizing the data, the quantile matching equation is established and the Johnson distribution parameters are fitted to obtain the best matching distribution model to describe the uncertainty of the data;

[0073] Based on the fitting results, the lower and upper bound cumulative distribution functions (LCDF and UCDF) are defined for the interval data, and the p-box variables are constructed to characterize the possible distribution range of the interval data;

[0074] For probabilistic data, statistical analysis methods are used to determine their probability distribution and generate p-box variables to fully describe the uncertainty of the input data.

[0075] Among them, the p-box (Probability Box) variable is a tool for representing uncertainty information, especially suitable for situations where both probability uncertainty and interval uncertainty exist. It defines the range of uncertainty through the upper and lower bounds of the two cumulative distribution functions (CDFs), so the p-box variable can simultaneously describe the uncertainty range and possible distribution characteristics of the input data.

[0076] S103. Based on the double-layer sampling technology, randomly sample the generated p-box variables to obtain sample data with interval and probability uncertainty. These data include uncertain input parameters such as wind speed, wind direction angle, temperature, turbulent wind speed and wind profile.

[0077] The specific steps are as follows:

[0078] Generated using the Latin Hypercube Sampling (LHS) method Outer layer samples θ (w) =(θ1 (w) ,θ2 (w) ,...,θ n (w) ),in The subscripts 1, 2, ... n represent the number of uncertainty input variables, and each sample represents the interval distribution parameter of the uncertainty input;

[0079] For each sample θ in the outer layer (w) , the sparse interpolation collocation point method is used to select valid sample points and generate inner layer samples Ξ. Specifically, the sparse interpolation of the sparse interpolation value collocation point method is used:

[0080]

[0081] in, represents the tensor product, ι represents the configuration level, |i|=i1+i2+...i n Represents the sum of multiple indicators, Indicates the i k dimensional collocation points, which can be obtained by Gaussian quadrature or Clenshaw-Curtis rule. Then select M sample points X from the sparse interpolation point set Ξ (v) =(X1 (v) ,X2 (v) ,...,X n (v) ), where v = 1, 2, ..., M is used for linear regression calculation. In order to obtain stable results, the number of sample points M must be greater than or equal to twice the polynomial coefficient, that is, M ≥ 2q.

[0082] S104. Based on the slice inverse regression (SIR) technology, the uncertain input data randomly selected by double-layer sampling is subjected to dimensionality reduction processing, and its key features are retained, thereby obtaining the data after dimensionality reduction;

[0083] In the process of wind turbine load reconstruction, high-dimensional uncertain input data (such as wind speed, wind direction, temperature wind profile, etc.) will lead to a significant increase in computational complexity, thereby affecting the efficiency and prediction ability of the model. In order to effectively deal with this problem, the present invention introduces the Sliced ​​Inverse Regression (SIR) technology to reduce the dimensionality of high-dimensional uncertain input data. The SIR technology can retain the most influential feature directions, simplify the data structure, and maintain important information for the model output.

[0084] Specifically, this example uses the SIR technology, and its key calculations are as follows:

[0085] The sample data generated by double-layer sampling is input as uncertain input parameters into the gas-elastic-mechanical-electrical coupling model to calculate the output response y;

[0086] Standardize each input variable so that its mean is zero and its variance is 1, and calculate the standardized input matrix X (v) The covariance matrix ξ (v) =(ξ1 (v) ,ξ2 (v) ,...,ξ n (v) ):

[0087]

[0088] X1 (i) For X1 (v) input variables, is the mean vector of the input data, and M is the number of samples.

[0089] Divide the output response y into H slices, and the number of samples in each slice is n h (v) , for each slice, calculate the slice mean m of the input variable h (v) :

[0090]

[0091] Among them, s h (v) represents the sample set belonging to the hth slice, j represents the slice S h The sample index of .

[0092] Compute the covariance matrix Γ using the slice mean (v) , which reflects the differences between slices:

[0093]

[0094] By solving the generalized eigenvalue problem Γβ=λξβ, the eigenvalue λ and eigenvector β are solved;

[0095] The eigenvector β is the effective dimension regression (EDR) direction of the SIR method, reflecting the most representative low-dimensional feature direction.

[0096] Select the first q eigenvectors β1, β2, ..., β1 with the largest eigenvalues. q , the high-dimensional input data X (v) Project to low-dimensional space to get the reduced-dimensional data matrix Z (v) =(Z1 (v) ,Z2 (v) ,...,Z q (v) ), where each Z i are the independent variables:

[0097] Z=XB

[0098] Where B = [β1,β2,...,β q ] is the matrix consisting of selected eigenvectors.

[0099] S105. Based on the uncertain input data with key features retained after dimensionality reduction, the optimal polynomial basis function is selected to construct the SPCE proxy model.

[0100] In this embodiment, the following steps are specifically included:

[0101] S1051, inputting the uncertainty input data after dimension reduction generated by the double-layer sampling strategy and the slice inverse regression (SIR) technology into the gas-elastic-mechanical-electrical coupling model to calculate the output response Y, i.e., the first output load;

[0102] S1052, constructing SPCE model;

[0103] In the uncertainty analysis of wind turbine loads, the sparse polynomial chaos expansion (SPCE) model is a key step to improve computational efficiency and accuracy. Through sparse decomposition, a simplified polynomial proxy model can be constructed in the high-dimensional input space to quickly calculate the load response of the wind turbine at different time points.

[0104] As the dimension n and order p increase, the total number of polynomial terms s0 will grow exponentially. For complex systems, due to the large number of input variables, the computational efficiency of the standard chaotic polynomial expansion is greatly reduced, which is called the "curse of dimensionality". Therefore, in order to reduce the influence of the total number of terms s0 on the variables n and order p, the sparse decomposition chaotic polynomial expansion theory is proposed.

[0105] Sparse decomposition is a data dimensionality reduction technique that aims to approximate high-dimensional functions by selecting a small number of important basis functions. For high-dimensional problems, directly constructing a polynomial chaos expansion (PCE) model may lead to problems such as high computational complexity and large data volume. Sparse decomposition significantly reduces the complexity and computational amount of the model by selecting sparse polynomial basis functions.

[0106] In this embodiment, the process of constructing the SPCE model is as follows:

[0107] PCE is a technique for uncertainty quantification and propagation that approximates the input-output relationship through polynomial basis functions. Let the output response Y be the input variables X1, X2, ..., X n The function can be expressed as:

[0108]

[0109] in, represents the i-th polynomial coefficient to be estimated, α i ={α i,1 ,α i,2 ,...,α i,n} represents a multi-index set, which is used to determine the i-th multivariate chaotic polynomial It can be obtained by tensor product of univariate polynomials:

[0110]

[0111] in, and α i,j They represent the order of univariate polynomial and polynomial respectively; Table 1 gives the univariate polynomial families corresponding to common random distribution types.

[0112] Table 1 Univariate polynomial forms corresponding to different random variables

[0113]

[0114] The chaotic polynomial multidimensional orthogonal basis satisfies the following orthogonal properties:

[0115]

[0116] Where, f(X)=f1(X1)*f2(X2)*...*f n (X n ) represents the joint probability density function of the input variables, f i (X i ), i=1,2,...,n represents the i-th marginal probability density function, Ω n represents the domain of vector X, when i=j,δ ij =1, i≠j when δij =0.

[0117] To simplify the calculation, The most common method is to keep the polynomial chaos expansion up to p order:

[0118]

[0119] Among them, s0 represents the total number of chaotic polynomial basis elements, which has the following relationship with dimension n and order p:

[0120]

[0121] Specifically, first initialize the chaotic polynomial coefficient c0 = 0, and define the candidate set as all basis functions Define the activity set as an empty set, and define the initial residual equal to the response function observation value (sample value) r0 = Y;

[0122] Calculate the inner product of each basis function and the current residual <r0,Ψ α >, find the basis function that is most relevant to the current residual That is, the optimal polynomial basis function, where α i represents the index of the i-th most relevant basis function;

[0123] Move all coefficients of the current active set toward their least squares values ​​until their regressors and residuals are equally correlated with the other regressors in the candidate set;

[0124] Calculate and store the error estimate ∈ j,LOO , and update all active coefficients and Move from the candidate set to the active set;

[0125] Repeat the previous steps until the size of the active set is equal to m = min(P, N-1), where m is the maximum capacity of the active set, P is the total number of basis functions, and N is the number of sample points;

[0126] Using the optimal polynomial basis functions selected in the active set, a sparse factorization polynomial chaos expansion (SPCE) model is constructed to calculate the final output response.

[0127] Specifically, the SPCE model constructed in this embodiment can be expressed as:

[0128]

[0129] Among them, Y(t) represents the change of wind turbine load with time t, Z represents the input variable after processing based on SIR algorithm, Represents Z iThe most relevant polynomial basis function is the optimal basis function. Represents polynomial coefficients.

[0130] S106. Based on the constructed SPCE model, its coefficients are expressed as quadratic polynomial function forms of interval distribution parameters, and the polynomial coefficients are recalculated to construct a multi-source uncertainty proxy model (SIR-SPCE) combining slice inverse regression technology and sparse chaotic polynomial expansion.

[0131] In this embodiment, for the uncertain input in the SPCE (sparse polynomial chaos expansion) model, especially the p-box variable, the model coefficients are expressed as quadratic polynomial functions of interval distribution parameters to more accurately describe the impact of input uncertainty on the output. This method is particularly suitable for handling transmission problems involving probability and interval multi-source uncertainty.

[0132] For the p-box variable, the coefficient c in the SPCE method is αi , i=1,2,...,q is expressed as a quadratic polynomial function of interval distribution parameters. Since there is no correlation between the distribution parameters of the p-box variables, the quadratic polynomial function does not contain interaction terms, which is specifically defined as:

[0133]

[0134] where θ j is the interval distribution parameter, a i,0 、b i,j d i,j are the coefficients of the quadratic polynomial, q is the order of the polynomial expansion, and the subscript j represents the dimension of the uncertain input variable.

[0135] Substituting the above coefficients into the original formula of the SPCE model, we get the SIR-SPCE model:

[0136]

[0137] In order to calculate the quadratic polynomial function, the input matrix and the corresponding output response are constructed based on the uncertainty input data after dimension reduction generated by the double-layer sampling strategy and the slice inverse regression (SIR) technique, and then the linear regression solution is performed using the least squares method.

[0138] Specifically, for the wth distribution parameter sample θ in the outer layer (w) , using the inner M sample points Z (v) =(Z1 (v) ,Z2 (v) ,...,Z q (v)), v = 1, 2, ..., M, and the SIR-SPCE coefficient is calculated by linear regression:

[0139] c s (w) =Y(ψ(Z) T ψ(Z) -1 ψ(Z) T

[0140] where Z = [Z (1) ,Z (2) ,...,Z (M) ] represents M sample points in n-dimensional space, Y=[Y(Z (1) ),Y(Z (2) ),...,Y(Z (M) )] represents the response function vector at the sample point, is the SIR-SPCE coefficient vector corresponding to the w-th distribution parameter sample in the outer layer, ψ(Z) represents the transformation matrix composed of the orthogonal polynomials at the sample point, and its form is:

[0141]

[0142] Repeat the above steps until all the outer layer parameter samples are executed, all the corresponding coefficients are obtained, and finally the SIR-SPCE model is obtained:

[0143]

[0144] in, represents the polynomial coefficients, Represents Z i The most relevant polynomial basis function, i.e. the optimal polynomial basis function, q is the polynomial expansion order, n is the total number of uncertain input variables, θ represents the outer layer distribution variable, a i,0 、b i,j d i,j are the coefficients of the quadratic polynomial.

[0145] S107. Comprehensive consideration of MSE and R 2 The numerical size of the SIR-SPCE model is then used to analyze the accuracy of the SIR-SPCE model. The polynomial order q and the number of basis functions of the model are adjusted to optimize the best balance between the accuracy and computational efficiency of the model.

[0146] In this embodiment, the specific steps are as follows:

[0147] calculating a mean value of the first output load;

[0148] The uncertainty input data after dimension reduction generated by the double-layer sampling strategy and slice inverse regression (SIR) technique is input into SIR-SPCE to obtain the second output load

[0149] The mean square error MSE and determination coefficient R are calculated based on the following formulas: 2 :

[0150]

[0151] in, is the second output load, Y i is the first output load, is the mean of the first output loads.

[0152] MSE represents the average of the squared differences between the actual value and the predicted value. It measures the degree of deviation between the model's predicted value and the actual value. The smaller the value, the better the model's prediction effect. 2 It indicates the proportion of total variance explained by the model and measures the degree of fit of the model. 2 The value range is between 0 and 1. The closer it is to 1, the better the predictive ability of the model.

[0153] Considering MSE and R 2 The numerical value of the SIR-SPCE model is then analyzed for accuracy. The polynomial order q and the number of basis functions of the model are adjusted to ensure the best balance between model accuracy and computational efficiency. Specifically, if the MSE is less than 0.1 and R 2 If it is greater than 0.9, the output result of the SIR-PCE model is judged to be accurate.

[0154] S108. According to the optimized SIR-SPCE model, uncertainty analysis is performed on the load of the wind turbine generator set to obtain a confidence prediction result of the load.

[0155] In this embodiment, based on a large number of simulation results, the confidence interval of the load is calculated by a statistical analysis method. For example, based on the distribution characteristics of the simulation samples, a 90% confidence interval is calculated, that is, there is a 90% probability that the actual load value will fall within this range.

[0156] The confidence interval is usually calculated using the quantile method or other statistical methods, and the corresponding confidence interval is output according to different confidence levels (such as 90%, 95%, etc.).

[0157] Finally, the prediction result of load reconstruction is output in the form of confidence. The typical result expression is "load value ± error range (confidence)", for example, "5000 ± 200N (90% confidence)", which means that in 90% of cases, the actual load falls between 4800N and 5200N.

[0158] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0159] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution disclosed in the present invention can be achieved, and this document does not limit this.

[0160] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for reconfiguring a load of a wind turbine generator set, characterized in that: The following steps are involved: Obtaining wind turbine blade operation data from a SCADA system, and identifying uncertainty factors that affect the load of a wind turbine generator set during wind turbine operation based on the operation data; By using Johnson distribution fitting technology, the uncertainty input in wind turbine operation is classified and modeled based on SCADA data, and p-box variables are constructed to accurately describe the impact of uncertainty factors on wind turbine loads; Based on the double-layer sampling technology, the generated p-box variables are randomly sampled to obtain sample data with interval and probability uncertainty, and the sample data includes wind speed, wind direction angle, temperature, turbulent wind speed and wind profile; Based on the slice inverse regression technology, the uncertain input data randomly selected by double-layer sampling is subjected to dimensionality reduction processing, retaining its key features, thereby obtaining the reduced-dimensional data; Based on the uncertain input data with key features retained after dimension reduction, the optimal polynomial basis function is selected to construct a sparse polynomial chaos expansion model. Based on the constructed sparse polynomial chaos expansion model, its coefficients are expressed as quadratic polynomial function forms of interval distribution parameters, and the polynomial coefficients are recalculated to construct a slice inverse regression technology-sparse polynomial chaos expansion model; Considering MSE and R 2 The numerical value of is used to analyze the accuracy of the slice inverse regression technology-sparse polynomial chaos expansion model, and the best balance between the accuracy and computational efficiency of the model is optimized by adjusting the polynomial order p and the number of basis functions of the model. According to the optimized slice inverse regression technology-sparse polynomial chaos expansion model, uncertainty analysis is performed on the load of the wind turbine generator set to obtain a confidence prediction result of the load.

2. The method for reconfiguring the load of a wind turbine generator set according to claim 1, characterized in that: The operating data includes wind speed, wind direction, temperature and rotation speed; The uncertain factors affecting the load of the wind turbine generator set include steady-state wind speed, turbulent wind speed, wind direction angle, temperature and wind profile.

3. The method for reconfiguring the load of a wind turbine generator set according to claim 1, characterized in that: Based on the double-layer sampling technology, the generated p-box variables are randomly sampled to obtain sample data with interval and probability uncertainty, and the sample data includes wind speed, wind direction angle, temperature, turbulent wind speed and wind profile, including: Generated using the Latin Hypercube Sampling (LHS) method Outer layer samples in The subscripts 1, 2, ... n represent the number of uncertainty input variables, and each sample represents the interval distribution parameter of the uncertainty input; For each sample θ in the outer layer (w) , the sparse interpolation collocation point method is used to select valid sample points and generate inner layer samples Ξ. Specifically, the sparse interpolation of the sparse interpolation value collocation point method is used: in, represents the tensor product, ι represents the configuration level, |i|=i1+i2+...i n Represents the sum of multiple indicators, Indicates the i k dimensional collocation points, which can be obtained by Gaussian quadrature or Clenshaw-Curtis rule; Then select M sample points X from the sparse interpolation point set Ξ (v) =(X1 (v) ,X2 (v) ,...,X n (v) ), where v = 1, 2, ..., M is used for linear regression calculation; in order to obtain stable results, the number of sample points M must be greater than or equal to twice the polynomial coefficient, that is, M ≥ 2q. .

4. The method for reconfiguring the load of a wind turbine generator set according to claim 1, characterized in that: in, The key calculations using the slice inverse regression technique are as follows: The sample data generated by double-layer sampling is input as uncertain input parameters into the gas-elastic-mechanical-electrical coupling model to calculate the output response y; Standardize each input variable so that its mean is zero and its variance is 1, and calculate the standardized input matrix X (v) The covariance matrix ξ (v) =(ξ1 (v) ,ξ2 (v) ,...,ξ n (v) ): X1 (i) For X1 (v) input variables, is the mean vector of the input data, M is the number of samples; Divide the output response y into H slices, and the number of samples in each slice is n h (v) , for each slice, calculate the slice mean m of the input variable h (v) : Among them, s h (v) represents the sample set belonging to the hth slice, j represents the slice S h The sample index of Compute the covariance matrix Γ using the slice mean (v) , which reflects the differences between slices: By solving the generalized eigenvalue problem Γβ=λξβ, the eigenvalue λ and eigenvector β are solved; Among them, the eigenvector β is the effective dimensional regression direction of the slice inverse regression method, reflecting the most representative low-dimensional feature direction; Select the first q eigenvectors β1, β2, ..., β1 with the largest eigenvalues. q , the high-dimensional input data X (v) Project to low-dimensional space to get the reduced-dimensional data matrix Z (v) =(Z1 (v) ,Z2 (v) ,...,Z q (v) ), where each Z i are the independent variables: Z=XB Where B = [β1,β2,...,β q ] is the matrix consisting of selected eigenvectors.

5. The method for reconfiguring the load of a wind turbine generator set according to claim 4, characterized in that: The constructed sparse polynomial chaos expansion model is expressed as: Among them, Y(t) represents the change of wind turbine load with time t, Z represents the input variable after processing based on SIR algorithm, Indicates that X i The most relevant polynomial basis functions, Represents polynomial coefficients.

6. The method for reconfiguring the load of a wind turbine generator set according to claim 4, characterized in that: Finally, a slice inverse regression technology-sparse polynomial chaos expansion model is constructed as follows: in, represents the polynomial coefficients, Represents Z i The most relevant polynomial basis function, i.e., the optimal polynomial basis function, q is the polynomial expansion order, n is the total number of uncertain input variables, θ represents the outer layer distribution variable, and a i,0 , b i,j ,d i,j are the coefficients of the quadratic polynomial.

7. The method for reconfiguring the load of a wind turbine generator set according to claim 6, characterized in that: The comprehensive consideration of MSE and R 2 The numerical value of is used to analyze the accuracy of the slice inverse regression technology-sparse polynomial chaos expansion model. By adjusting the polynomial order q and the number of basis functions of the model, the best balance between the accuracy and computational efficiency of the model is optimized, including: calculating a mean value of the first output load; The uncertainty input data after dimension reduction generated by the double-layer sampling strategy and the slice inverse regression technology is input into the slice inverse regression technology-sparse polynomial chaos expansion model to obtain the second output vector. Load Y; The mean square error MSE and determination coefficient R are calculated based on the following formulas: 2 : in, is the second output load, Y i is the first output load, is the mean of the first output loads.

Citation Information

Cited By

  • Method for rapidly forecasting hydrodynamic load of navigation body under influence of internal waves

    CN120724851A