Method and device for screening dominant parameters of wind turbine generator

Abstract

This application relates to a method and apparatus for screening dominant parameters of wind turbine generators. The method includes: adopting a hierarchical decoupling screening strategy based on a physical excitation mechanism, dividing the parameters to be screened from the wind turbine generator into three subspaces: turbine-side electrical and control, grid-side conventional dynamics, and fault ride-through. For each subspace, a parameter sample matrix is ​​constructed using a sampling method. A trajectory error objective function is constructed based on the simulation output data obtained by substituting the parameters from the sample matrix into the simulation and the corresponding baseline transient output data, thereby calculating the full-order sensitivity index of each parameter to be screened. This method, through physical decoupling, splits the high-dimensional parameter space into three low-dimensional independent subspaces, reducing simulation time and thus improving screening efficiency. Furthermore, by determining the full-order sensitivity of the parameters to be screened and performing global sensitivity analysis, the independent contribution and coupling influence of parameters are accurately quantified, significantly enhancing the accuracy and physical reliability of the screening results.

Description

Technical Field

[0001] This application relates to the field of new energy power generation technology, and in particular to a method and apparatus for screening the dominant parameters of a wind turbine. Background Technology

[0002] With the high proportion of wind power connected to the grid, constructing high-precision wind turbine simulation models is crucial for optimizing turbine operation control and ensuring the safe and stable operation of the power grid. The accuracy of the simulation model highly depends on the precision of parameter identification. However, wind turbine simulation models contain a large number of electrical, control, and other parameters, forming a typical high-dimensional nonlinear space. Identifying all parameters would inevitably consume a significant amount of computation time and could easily lead to the curse of dimensionality, causing optimization algorithms to get stuck in local optima and failing to meet the simulation model's requirement for online parameter updates.

[0003] Traditional approaches either employ local sensitivity analysis, which estimates the partial derivatives of input variables at specific points to study the impact of small perturbations on the model output. This method is simple in principle and relatively computationally efficient, but it ignores the strong nonlinear cross-coupling between various parameters, easily leading to the omission of dominant parameters. Alternatively, global sensitivity analysis is used, which allows the parameters to be identified to undergo large-scale random fluctuations within a set physical space. This method can accurately quantify the independent contribution of a single parameter and the combined effect of multiple parameters. However, when dealing with complex nonlinear systems, it suffers from the curse of dimensionality, resulting in long computation time and sensitivity masking effects under single perturbation faults. Furthermore, the physical interpretability of the screening results is poor.

[0004] In the above-mentioned schemes, when faced with multi-parameter screening scenarios, it is impossible to simultaneously consider the speed, accuracy, and physical interpretability of parameter screening. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for selecting the dominant parameters of wind turbine units that can solve the above-mentioned technical problems.

[0006] Firstly, this application provides a method for selecting the dominant parameters of a wind turbine generator. The method includes: Based on the decoupling characteristics and control principle of the unit topology, multiple parameters to be screened for the wind turbine are divided into the machine-side electrical and control subspace, the grid-side conventional dynamic subspace, and the fault ride-through subspace, and the optimal excitation and disturbance conditions are matched for each subspace. For each subspace, two independent and highly uniform parameter sample matrices are generated based on the perturbation sequence, and a hybrid matrix is ​​constructed based on the two parameter sample matrices; wherein, each parameter sample matrix includes the physical values ​​of the corresponding parameter to be screened under different simulation states in the subspace; Each parameter sample matrix and the hybrid matrix are input into the simulation model of the wind turbine, and the simulation output datasets corresponding to each parameter sample matrix and the hybrid matrix are output; wherein, the simulation output datasets include the simulation output quantities of the active power, reactive power, DC bus voltage and rotor speed of the wind turbine. For each subspace, based on each simulation output and the corresponding baseline transient output, the parameter sample matrix and the mixture matrix are respectively determined to correspond to the trajectory error objective function of the subspace. Based on each function value and the total variance corresponding to each function value, the full-order sensitivity index corresponding to each parameter to be screened is determined, and the parameters to be screened whose full-order sensitivity index is greater than a preset threshold are determined as dominant parameters; wherein, the dominant parameters are used for parameter identification of the wind turbine.

[0007] Secondly, this application also provides a device for screening the dominant parameters of a wind turbine generator. The device includes: The parameter partitioning module is used to partition multiple parameters to be screened in the wind turbine into the turbine-side electrical and control subspace, the grid-side conventional dynamic subspace, and the fault ride-through subspace based on the decoupling characteristics and control principles of the turbine topology, and to match the optimal excitation and disturbance conditions for each subspace. The parameter processing module is used to generate two independent and highly uniform parameter sample matrices based on the perturbation sequence for each subspace, and to construct a hybrid matrix based on the two parameter sample matrices; wherein each parameter sample matrix includes the physical values ​​of the corresponding parameter to be screened under different simulation states in the subspace; The simulation module is used to input the parameter sample matrices and the hybrid matrix into the simulation model of the wind turbine, and output the simulation output datasets corresponding to each parameter sample matrix and the hybrid matrix; wherein, the simulation output datasets include the simulation output quantities of the active power, reactive power, DC bus voltage and rotor speed of the wind turbine. The function value determination module is used to determine, for each of the subspaces, the function values ​​of the trajectory error objective function corresponding to the parameter sample matrix and the mixture matrix respectively, based on each of the simulation output quantities and the reference transient output quantity corresponding to each of the simulation output quantities; The parameter filtering module is used to determine the full-order sensitivity index corresponding to each of the function values ​​and the total variance corresponding to each function value, and to determine the parameters to be filtered that have a full-order sensitivity index greater than a preset threshold as the dominant parameters; wherein, the dominant parameters are used for parameter identification of the wind turbine.

[0008] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the above method steps.

[0009] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the above method steps.

[0010] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the above method steps.

[0011] The aforementioned methods, devices, computer equipment, storage media, and computer program products for screening the dominant parameters of wind turbines, based on the decoupling characteristics of the turbine topology and control principles, divide multiple parameters to be screened into three subspaces: the turbine-side electrical and control subspace, the grid-side conventional dynamic subspace, and the fault ride-through subspace. Optimal excitation and disturbance conditions are then matched to each subspace. Firstly, by using hierarchical decoupling of physical excitations, the high-dimensional parameter space is split into three independent low-dimensional subspaces, reducing simulation time and effectively avoiding parameter masking effects and ineffective cross-coupling interference under single operating conditions, thus improving the resolution of dominant parameter screening. Secondly, by dividing a large number of parameters to be screened into three dedicated subspaces according to their physical meaning and dynamic characteristics, and matching optimal excitation and disturbance conditions (turbine-side, grid-side, and fault ride-through) to each subspace, structured grouping of parameters is achieved. This helps subsequent analysis to be more targeted, avoids mutual interference between parameters from different physical processes, and lays the foundation for sensitivity analysis by region and scenario.

[0012] For each subspace, two independent and highly uniform parameter sample matrices are generated based on the perturbation sequence, and a hybrid matrix is ​​constructed based on the two parameter sample matrices. Each parameter sample matrix includes the physical values ​​of the corresponding parameter to be screened under different simulation states within the subspace. Firstly, using the perturbation sequence to generate the parameter sample matrix reduces the exponential computational complexity to a linear level, lowering computational cost and making it more suitable for low-computational-capacity scenarios. Secondly, using the perturbation sequence to generate two parameter sample matrices is equivalent to introducing repeated or comparative samples of parameter perturbation within the same subspace. This effectively captures the impact of parameter changes on the system response, improves the statistical reliability and robustness of subsequent sensitivity estimation, and reduces random errors caused by single sampling.

[0013] The sample matrices and hybrid matrices of each parameter are input into the simulation model of the wind turbine, and the simulation output datasets corresponding to the hybrid matrix and each parameter sample matrix are output. The simulation output datasets include the simulated output quantities of the wind turbine's active power, reactive power, DC bus voltage, and rotor speed. By simultaneously inputting the original sample matrices and the hybrid matrix for simulation and outputting various key electrical and mechanical quantities, rich high-dimensional dynamic response data are obtained. This provides multi-dimensional information for comprehensively evaluating the impact of parameters on the transient behavior of the unit, and can more accurately reflect the global effect of parameters compared to observing only a single output quantity.

[0014] For each subspace, based on each simulation output and the corresponding baseline transient output, a parameter sample matrix and a mixing matrix are determined, each corresponding to the objective function value of the trajectory error for that subspace. By calculating the objective function values ​​of the trajectory error for both the mixing matrix and the sample matrix, the degree to which the system response deviates from the ideal behavior under different parameter perturbations can be quantified. The trajectory-based quantification provides a unified and comparable numerical basis for subsequent sensitivity ranking.

[0015] Based on each function value and its corresponding total variance, the full-order sensitivity index for each parameter to be screened is determined. This method, on the one hand, determines the full-order sensitivity of each parameter to be screened, performs global sensitivity analysis, and accurately quantifies the independent contribution and coupling effect of parameters, significantly enhancing the accuracy and physical reliability of the screening results. On the other hand, parameters with a full-order sensitivity index greater than a preset threshold are selected as dominant parameters. These dominant parameters are used for parameter identification of wind turbines, ensuring accurate screening while also considering subsequent identification accuracy and real-time online updates, reducing the difficulty of engineering applications. Furthermore, through physical decoupling, the high-dimensional parameter space is divided into three low-dimensional independent subspaces, reducing simulation time and thus improving screening efficiency. Attached Figure Description

[0016] Figure 1This is a flowchart illustrating a method for selecting the dominant parameters of a wind turbine in one embodiment. Figure 2 This is a flowchart illustrating a method for selecting the dominant parameters of a wind turbine in another embodiment. Figure 3 This is a schematic diagram of the subspace partitioning method for each parameter to be screened in one embodiment; Figure 4 This is a flowchart illustrating a method for generating a hybrid matrix in one embodiment; Figure 5 This is a structural block diagram of a device for screening the dominant parameters of a wind turbine in one embodiment; Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] In one embodiment, such as Figure 1 As shown, a method for selecting the dominant parameters of a wind turbine generator is provided, including: S101, based on the decoupling characteristics and control principle of the unit topology, divides multiple parameters to be screened for the wind turbine into the machine-side electrical and control subspace, the grid-side conventional dynamic subspace, and the fault ride-through subspace, and matches the optimal excitation and disturbance conditions for each subspace.

[0019] Among them, the decoupling characteristics of the unit topology utilize physical laws (such as Kirchhoff's laws in circuits) and power electronic control technology to artificially or naturally divide this unit topology into several physical links that do not interfere with each other (or have minimal interference).

[0020] The control principle of the turbine topology represents the hierarchical control principle of the wind turbine. Specifically, the hierarchical principle includes: Top layer (power / speed control): Controls the pitch angle and generator speed based on wind speed to capture maximum wind energy. Middle layer (generator-side converter control): Controls generator torque to implement speed loop commands. Its core is decoupling control, decomposing the stator current into excitation and torque components for separate control. Bottom layer (grid-side converter control): Stabilizes the DC bus voltage and controls the reactive power output to the grid. It also employs decoupling control, dividing the grid-connected current into active and reactive components.

[0021] Among them, the Wind Turbine Generator System (WTGS) is a complete system that converts wind energy into electrical energy.

[0022] The parameters to be screened represent all adjustable parameters in the wind turbine simulation model that require importance assessment. These parameters directly affect the dynamic response accuracy of the model. The number of parameters to be screened is large, potentially including dozens or even hundreds of electrical, mechanical, or control parameters.

[0023] A subspace refers to a subset of parameters obtained by dividing all parameters to be screened according to their physical module, dynamic response time scale, or functional role. A subspace can be a machine-side electrical and control subspace, a grid-side conventional dynamic subspace, or a fault-crossing subspace.

[0024] The generator-side electrical and control subspace can represent the space containing electrical and control parameters related to the generator body and its directly connected converter (generator-side converter).

[0025] Specifically, the machine-side electrical and control subspace The parameters to be screened include generator electrical parameters and generator-side converter control parameters. This space is mainly driven by mechanical torque; therefore, a step change in wind speed is adopted as the optimal excitation disturbance condition for the generator-side electrical and control subspace to maximize the contribution of the generator internal parameters and the trajectory variance of the speed loop.

[0026] The grid-side conventional dynamic subspace can represent a space containing parameters related to the grid-side converter and its interaction with the grid.

[0027] Specifically, the network-side conventional dynamic subspace The parameters to be screened include grid-side converter control parameters and phase-locked loop parameters. For these tracking and synchronization control parameters, a small disturbance of the grid voltage is adopted as the optimal excitation disturbance condition in the grid-side conventional dynamic subspace, avoiding the intervention of nonlinear fault logic and ensuring the accurate extraction of the dynamic characteristics of the conventional PI controller.

[0028] The fault ride-through subspace contains a space for protection logic parameters and transient control parameters specifically designed to handle severe grid faults such as undervoltage / overvoltage ride-through.

[0029] Specifically, fault traversal subspace The parameters to be screened include: DC unloading circuit control parameters, reactive power support coefficient, and other transient limiting and protection parameters. Since these parameters have strict dead-zone trigger thresholds, a deep grid-connected voltage drop disturbance is adopted as the optimal excitation disturbance condition for the fault ride-through subspace, in order to fully stimulate unbalanced power flow and accurately quantify the dominant role of fault ride-through characteristic parameters in the transient trajectory.

[0030] Specifically, each parameter to be screened can be divided into its own subspace, achieving physical decoupling while facilitating subsequent separate parameter sensitivity evaluation for different subspaces.

[0031] For example, such as Figure 2 As shown, the physical excitation layer decoupling characteristic can be used to further divide the high-dimensional parameter set to be screened (multiple parameters to be screened for wind turbines) into subspaces: machine-side electrical and control subspace. Network-side conventional dynamic subspace and fault traversal subspace .

[0032] For example, such as Figure 3 As shown, the parameter space is divided for each parameter to be screened, resulting in a generator-side electrical and control subspace, a grid-side conventional dynamic subspace, and a fault ride-through subspace. The generator-side electrical and control subspace corresponds to the generator electrical parameters and generator-side converter control parameters, matched to wind speed step change excitation; the grid-side conventional dynamic subspace corresponds to the grid-side converter control parameters and phase-locked loop parameters, matched to small grid voltage disturbance excitation; and the fault ride-through subspace corresponds to the DC unloading circuit control parameters and reactive power support coefficient, matched to deep voltage drop large disturbance excitation. These three subspaces are independent of each other. By using condition-specific excitation, the parameter masking effect under a single operating condition is avoided, achieving precise decomposition of the high-dimensional parameter space.

[0033] By using the above-mentioned decoupling strategy based on different working conditions, the sensitivity evaluation of the original high-dimensional parameter space is split into three independent low-dimensional evaluation tasks, which significantly reduces the total number of simulations required, while improving the resolution and physical reliability of the selection of dominant parameters.

[0034] S102, for each subspace, generate two independent and highly uniform parameter sample matrices based on the perturbation sequence, and construct a mixing matrix based on the two parameter sample matrices.

[0035] A perturbation sequence is a numerical sequence consisting of a series of artificially set changes with known statistical characteristics applied to the nominal values ​​of each parameter to be screened within a subspace. The purpose of processing parameters based on perturbation sequences is to simulate the uncertainty or variability of the parameters within a reasonable range, thereby stimulating differences in the system's dynamic response for sensitivity analysis. Perturbation sequences include Sobol sequences and Latin hypercube sampling sequences, and are used for global sensitivity analysis.

[0036] The parameter sample matrix is ​​a two-dimensional numerical table where each row represents an independent simulation experiment and each column represents a parameter to be screened within a subspace. Each element in the matrix represents the specific value assigned to the corresponding parameter in a given simulation. Both parameter sample matrices are independent and highly homogeneous, and their dimensions can be the same.

[0037] Each parameter sample matrix includes the simulation parameter values ​​corresponding to the parameters to be screened under different simulation conditions.

[0038] Simulation conditions refer to all settings or scenario descriptions that can affect the simulation results, except for the different values ​​of the parameters to be screened. Specifically, simulation conditions can include power grid fault scenarios (grid drop depth, fault duration, fault type), generator operating points (initial wind speed, initial active power, initial reactive power), simulation duration and step size, etc.

[0039] Simulation parameter values ​​refer to the specific numerical values ​​assigned to a particular parameter to be screened in the simulation model, which are used for calculations in this simulation experiment.

[0040] Specifically, for each subspace, the parameters to be screened in the subspace are processed based on the perturbation sequence to obtain two independent and highly uniform parameter sample matrices. This includes: determining the simulation conditions for each subspace; simulating the parameters to be screened in the subspace according to the simulation conditions and the perturbation sequence to obtain the simulation parameter values ​​corresponding to the parameters to be screened in the subspace; and constructing two parameter sample matrices based on the simulation parameter values. Both parameter sample matrices are independent and highly uniform matrices with the same matrix dimension.

[0041] Specifically, for each subspace, the parameters to be screened in the subspace are processed based on the perturbation sequence to obtain two independent and highly uniform parameter sample matrices, including: generating two independent and highly uniform parameter sample matrices using the perturbation Sobol sequence. N × d k dimensional parameter sample matrix A (k) and B (k) ,in N The base sample size.

[0042] The mixing matrix can be a new parameter sample matrix generated by cross-combining or column permutation of two independent parameter sample matrices. The parameter values ​​in different columns of the mixing matrix come from different original sample matrices, thus breaking the original joint distribution between the parameters.

[0043] First, by using a perturbation sequence to process the parameters to be screened, the exponential computational complexity is reduced to a linear level, thus lowering computational cost and making it more suitable for low-computational-capacity scenarios. Second, by generating two parameter sample matrices using the perturbation sequence, it is equivalent to introducing repeated or comparative samples of parameter perturbation within the same subspace. This effectively captures the impact of parameter changes on the system response, improves the statistical reliability and robustness of subsequent sensitivity estimation, and reduces random errors caused by single sampling.

[0044] S103 inputs the sample matrices of each parameter and the mixing matrix into the simulation model of the wind turbine, and outputs the mixing matrix and the simulation output datasets corresponding to each parameter sample matrix.

[0045] A simulation model can be a program or system that mathematically describes the dynamic behavior of a wind turbine and implements it on a computer. A simulation model can be used to receive a set of specific parameter values ​​defined in each row of a parameter sample matrix (or hybrid matrix), as well as fixed conditions such as external fault scenarios, and then solve differential-algebraic equations through numerical integration to output the electrical and mechanical quantities of the turbine that change over time (i.e., output simulation dataset).

[0046] A simulation dataset can be a collection of data obtained by summarizing the simulation results corresponding to all rows of a certain input matrix (parameter sample matrix or mixture matrix).

[0047] The simulation output dataset includes the simulation output values ​​of the wind turbine's active power, reactive power, DC bus voltage, and rotor speed. Specifically, the simulation output datasets corresponding to the hybrid matrix, simulation model, and simulation dataset are as follows: .in, , , , These represent the simulated output quantities of active power, reactive power, DC bus voltage, and rotor speed, respectively.

[0048] For example, such as Figure 2 As shown, in the machine-side electrical and control subspace In this process, Saltelli recombination matrices (parameter sample matrices and mixing matrices of subspaces) are generated based on Sobol low-difference sequences. This is done within the conventional dynamic subspace on the network side. The Saltelli recombination matrix (parameter sample matrix and mixing matrix of the subspace) is generated based on Sobol low-dissimilarity sequences. This is then applied in the fault-crossing subspace. In this study, Saltelli recombination matrices (parameter sample matrices and mixing matrices of subspaces) are generated based on Sobol low-difference sequences.

[0049] For example, such as Figure 4 The mixture matrix is ​​obtained from the parameter sample matrix, including: for the subspace The first in Parameters ( =1, 2, ..., Construct a mixture matrix . The column vector divided by the first The columns are fully inherited from matrix A (k)(parameter sample matrix) of the first Except for column B, all other columns are taken from matrix B. (k) (Parameter sample matrix). Subspace Just build and run N ×( d k By adding 2 sets of parameter matrices, the exponential computational complexity of the traditional Monte Carlo method can be reduced to the linear level, enabling efficient generation of wind turbine parameter samples.

[0050] Simulations were performed by simultaneously inputting the original sample matrix and the mixed matrix, and various key electrical and mechanical quantities were output, resulting in rich high-dimensional dynamic response data. This provides multi-dimensional information for comprehensively evaluating the impact of parameters on the unit's transient behavior, and can more accurately reflect the global effect of parameters compared to observing only a single output quantity.

[0051] S104, for each subspace, based on each simulation output and the corresponding baseline transient output, determine the mixing matrix and each parameter sample matrix, which respectively correspond to the function values ​​of the trajectory error objective function of the subspace.

[0052] The baseline transient output refers to the dynamic response trajectory of the wind turbine, used as a reference standard to measure the deviation between the simulated output and response under other parameter combinations. Essentially, it is a set of standard waveform data that varies over time. The baseline transient output can originate from a baseline transient output dataset, which is primarily obtained from a pre-built simulation model, recording its output data under the same operating conditions. This can be understood as setting corresponding operating conditions under the original simulation parameters and recording the corresponding output data.

[0053] The trajectory error objective function is a mathematical functional that maps the difference between two time series (simulation output and reference transient output) to a non-negative real number. The trajectory error objective function quantitatively describes the degree of deviation between the dynamic response and the reference response under the current parameter combination.

[0054] The objective function value of the trajectory error can be the root mean square error or the weighted root mean square error, which describes the degree of deviation mentioned above.

[0055] Specifically, for each subspace, based on each simulation output and the corresponding baseline transient output, a mixing matrix and parameter sample matrices are determined, each corresponding to the function value of the trajectory error objective function for the subspace. This includes: for each subspace, obtaining the baseline transient output of the simulation model under the baseline condition; the baseline condition is the same as the simulation condition of the simulation output; based on the baseline transient output and each simulation output, determining the mixing matrix and parameter sample matrices, each corresponding to the function value of the trajectory error objective function for the subspace.

[0056] Specifically, for each subspace, the weighting coefficients of the trajectory error objective function are obtained; based on the weighting coefficients, each simulation output quantity, and the baseline transient output quantity corresponding to each simulation output quantity, the function values ​​of the trajectory error objective function of each subspace corresponding to the mixing matrix and each parameter sample matrix are determined.

[0057] For example, the objective function for trajectory error in the machine-side electrical and control subspace is: ; The objective function for trajectory error in the conventional dynamic subspace on the network side is: ; The objective function for the trajectory error in the fault traversal subspace is: ; in, , , , , These are the weighting coefficients for active power, reactive power, rotor speed, DC bus voltage, and DC unloading circuit, respectively. The reference transient output dataset is... Simulation output dataset .

[0058] By calculating the objective function values ​​of the trajectory error of the mixing matrix and the sample matrix, the degree to which the system response deviates from the ideal behavior under different parameter perturbations can be quantified. The trajectory-based quantification index provides a unified and comparable numerical basis for subsequent sensitivity ranking.

[0059] S105, based on each function value and the total variance corresponding to each function value, determine the full-order sensitivity index corresponding to each parameter to be screened, and determine the parameters to be screened with a full-order sensitivity index greater than a preset threshold as the dominant parameters.

[0060] Among them, the dominant parameters are used for parameter identification of wind turbine units.

[0061] Among them, the full-order sensitivity index can be a quantitative indicator that comprehensively reflects the contribution of a single parameter to the total variance of the model output.

[0062] The preset threshold can be a pre-defined numerical threshold used to classify parameters to be filtered into dominant parameters and non-dominant parameters.

[0063] The dominant parameters can have a significant and non-negligible impact on the dynamic response of wind turbine units.

[0064] Parameter identification can be represented as an optimization problem that uses measured data or benchmark responses to infer the optimal parameter values ​​in a model.

[0065] Specifically, after obtaining the function values, the total variance corresponding to each parameter to be screened can be determined based on the function values, and the first-order sensitivity and full-order sensitivity index corresponding to each parameter to be screened can be determined based on the total variance.

[0066] Specifically, the preset threshold can be set to 0.05. Parameters with a full-order sensitivity index greater than 0.05 are identified as dominant parameters, which can then be fed into the subsequent high-dimensional optimization identification algorithm. In practical applications, unselected low-sensitivity parameters are fixed as typical empirical values, and these dominant parameters are then input into the high-dimensional optimization identification algorithm for accurate online or offline identification. This ensures the identification accuracy of core parameters while significantly improving the convergence speed of the optimization algorithm, meeting the real-time requirements of online updates.

[0067] For example, such as Figure 2 As shown, after obtaining the values ​​of each function and the total variance corresponding to each function value, the sensitivity index can be estimated based on each function and the total variance (variance decomposition principle), thus obtaining the first-order sensitivity. With full-order sensitivity index Furthermore, after obtaining the full-order sensitivity index of each parameter to be screened, a uniform threshold is set. After normalizing and sorting the parameters to be screened, the parameters with a full-order sensitivity index greater than the preset threshold are determined as dominant parameters, thereby obtaining a set of dominant parameters corresponding to multiple dominant parameters.

[0068] In the aforementioned method for selecting the dominant parameters of wind turbines, based on the decoupling characteristics of the turbine topology and the control principle, multiple parameters to be screened are divided into a turbine-side electrical and control subspace, a grid-side conventional dynamic subspace, and a fault ride-through subspace. Optimal excitation and disturbance conditions are then matched to each subspace. Firstly, by using hierarchical decoupling of physical excitations, the high-dimensional parameter space is split into three low-dimensional independent subspaces, reducing simulation time and effectively avoiding parameter masking effects and ineffective cross-coupling interference under single operating conditions, thus improving the resolution of dominant parameter screening. Secondly, by dividing a large number of parameters to be screened into three dedicated subspaces according to their physical meaning and dynamic characteristics, and matching optimal excitation and disturbance conditions (turbine-side, grid-side, fault ride-through) to each subspace, structured grouping of parameters is achieved. This helps subsequent analysis to be more targeted, avoids mutual interference between parameters of different physical processes, and lays the foundation for sensitivity analysis by region and scenario.

[0069] For each subspace, two independent and highly uniform parameter sample matrices are generated based on the perturbation sequence, and a hybrid matrix is ​​constructed based on the two parameter sample matrices. Each parameter sample matrix includes the physical values ​​of the corresponding parameter to be screened under different simulation states within the subspace. Firstly, using the perturbation sequence to generate the parameter sample matrix reduces the exponential computational complexity to a linear level, lowering computational cost and making it more suitable for low-computational-capacity scenarios. Secondly, using the perturbation sequence to generate two parameter sample matrices is equivalent to introducing repeated or comparative samples of parameter perturbation within the same subspace. This effectively captures the impact of parameter changes on the system response, improves the statistical reliability and robustness of subsequent sensitivity estimation, and reduces random errors caused by single sampling.

[0070] The sample matrices and hybrid matrices of each parameter are input into the simulation model of the wind turbine, and the simulation output datasets corresponding to the hybrid matrix and each parameter sample matrix are output. The simulation output datasets include the simulated output quantities of the wind turbine's active power, reactive power, DC bus voltage, and rotor speed. By simultaneously inputting the original sample matrices and the hybrid matrix for simulation and outputting various key electrical and mechanical quantities, rich high-dimensional dynamic response data are obtained. This provides multi-dimensional information for comprehensively evaluating the impact of parameters on the transient behavior of the unit, and can more accurately reflect the global effect of parameters compared to observing only a single output quantity.

[0071] For each subspace, based on each simulation output and the corresponding baseline transient output, a parameter sample matrix and a mixing matrix are determined, each corresponding to the objective function value of the trajectory error for that subspace. By calculating the objective function values ​​of the trajectory error for both the mixing matrix and the sample matrix, the degree to which the system response deviates from the ideal behavior under different parameter perturbations can be quantified. The trajectory-based quantification provides a unified and comparable numerical basis for subsequent sensitivity ranking.

[0072] Based on each function value and its corresponding total variance, the full-order sensitivity index for each parameter to be screened is determined. This method, on the one hand, determines the full-order sensitivity of each parameter to be screened, performs global sensitivity analysis, and accurately quantifies the independent contribution and coupling effect of parameters, significantly enhancing the accuracy and physical reliability of the screening results. On the other hand, parameters with a full-order sensitivity index greater than a preset threshold are selected as dominant parameters. These dominant parameters are used for parameter identification of wind turbines, ensuring accurate screening while also considering subsequent identification accuracy and real-time online updates, reducing the difficulty of engineering applications. Furthermore, through physical decoupling, the high-dimensional parameter space is divided into three low-dimensional independent subspaces, reducing simulation time and thus improving screening efficiency.

[0073] In one embodiment, the parameter sample matrices and the mixture matrix are input into the simulation model of the wind turbine, and the simulation output datasets corresponding to each parameter sample matrix and the mixture matrix are output, including: For each subspace, determine the simulation conditions for that subspace; According to the simulation conditions, the parameters in each parameter sample matrix and the corresponding hybrid matrix are input into the simulation model of the wind turbine, and the simulation output datasets corresponding to each parameter sample matrix and the hybrid matrix are output.

[0074] Simulation conditions can be settings or scenario descriptions that affect the simulation results. Specifically, simulation conditions can include grid fault scenarios (grid drop depth, fault duration, fault type), generator operating points (initial wind speed, initial active power, initial reactive power), simulation duration and step size, etc.

[0075] The simulation output dataset includes simulated outputs of the wind turbine's active power, reactive power, DC bus voltage, and rotor speed. By simultaneously inputting the original sample matrix and the hybrid matrix for simulation, and outputting various key electrical and mechanical quantities, rich high-dimensional dynamic response data was obtained. This provides multi-dimensional information for comprehensively evaluating the impact of parameters on the unit's transient behavior, and more accurately reflects the global effect of parameters compared to observing only a single output quantity.

[0076] The simulation conditions for each subspace are different. For example, the simulation conditions for the machine-side electrical and control subspace include step changes in wind speed, the simulation conditions for the grid-side conventional dynamic subspace include small disturbances in grid voltage, and the simulation conditions for the fault ride-through subspace include large drops in grid voltage.

[0077] Specifically, based on simulation conditions and the Sobol perturbation sequence, simulation sampling is performed on the parameters to be screened in the subspace to obtain the simulation parameter values ​​corresponding to the parameters to be screened in the subspace. After obtaining the simulation parameter values, two parameter sample matrices (A and B) can be constructed based on the simulation parameter values. Each parameter sample matrix has the same dimension (both parameter sample matrices are...). N × d k The parameters in the parameter sample matrix are input into the simulation model of the wind turbine, and the simulation output dataset of the parameter sample matrix is ​​output. Similarly, the parameters in the hybrid matrix are input into the simulation model of the wind turbine, and the simulation output dataset of the hybrid matrix is ​​output.

[0078] In this embodiment, firstly, a perturbation sequence is used to process the parameters to be screened, reducing the exponential computational complexity to a linear level, thus reducing computational power consumption and making it more suitable for low-computational-power scenarios. Secondly, two parameter sample matrices are generated using the perturbation sequence, which is equivalent to introducing repeated or comparative samples of parameter perturbation in the same subspace. This can effectively capture the impact of parameter changes on the system response, improve the statistical reliability and robustness of subsequent sensitivity estimation, and reduce random errors caused by single sampling. Thirdly, the simulation output dataset simultaneously contains four key output quantities (active power, reactive power, DC voltage, and rotor speed), enabling the subsequent trajectory objective function to comprehensively evaluate the impact of parameters on the dynamic coupling of multiple physical quantities of the unit, rather than focusing on only a single indicator. Fourthly, simulation output datasets for parameter sample matrices A, B, and the hybrid matrix are obtained. By analyzing the variance and covariance of the corresponding function values ​​of these three datasets, it is easy to isolate the main effects and interaction effects of individual parameters, and finally calculate the full-order sensitivity index.

[0079] In one embodiment, for each subspace, based on each simulation output and the corresponding baseline transient output, the function values ​​of the trajectory error objective function for the parameter sample matrix and the mixing matrix for each subspace are determined. This includes: for each subspace, obtaining the baseline transient output of the simulation model under the baseline condition; the baseline condition is the same as the simulation condition of the simulation output. Based on the baseline transient output and each simulation output, the function values ​​of the trajectory error objective function for the mixing matrix and each parameter sample matrix for each subspace are determined.

[0080] The baseline operating condition is the set of all external conditions and initial state of the model set when generating the baseline transient output (reference waveform). The baseline operating condition can include grid fault scenarios, wind turbine operating points, environmental and grid conditions, model structure and parameters, and simulation control settings.

[0081] The grid fault scenarios include fault type, drop depth, fault duration, fault start time, and fault recovery characteristics. The wind turbine operating points include initial wind speed, initial active power, initial reactive power, generator speed, and pitch angle. Environmental and grid conditions include grid short-circuit capacity, line impedance, transformer tap position, and ambient temperature (affecting resistance). The model structure and parameters include model topology (whether it includes detailed models of gearboxes and LCL filters) and some fixed parameters (such as generator rated capacity and rated voltage). The simulation control settings include total simulation duration, step size, solver type, and output sampling frequency.

[0082] A simulation test case refers to the set of all external conditions and initial state of the model set in each specific simulation experiment driven by a parameter sample matrix or a hybrid matrix. It shares all fixed elements with the baseline test case, and the only allowed change is the value of the parameter to be screened. Specifically, the simulation test case and the baseline test case are consistent in terms of external fault scenarios, initial operating points, model structure, and simulation step size / duration.

[0083] In this embodiment, firstly, by ensuring that the baseline operating condition and all simulation operating conditions are completely consistent in terms of external conditions such as fault scenarios and initial operating points, the trajectory deviations calculated subsequently are uniquely attributed to changes in the values ​​of the parameters to be screened. This eliminates the interference of external condition differences on the error, thereby ensuring the fairness and accuracy of parameter sensitivity analysis. Secondly, by quantitatively comparing each group of simulation outputs with the same baseline trajectory, the complex dynamic response differences corresponding to each parameter sample matrix (A, B) and the mixture matrix can be compressed into unified and comparable function values. This provides standardized input data for subsequent calculation of the full-order sensitivity index, enabling quantitative ranking of the importance of different parameters and supporting the automatic screening of dominant parameters.

[0084] In one embodiment, each function value includes a first function value corresponding to the mixing matrix and a second function value corresponding to each parameter sample matrix.

[0085] Based on each function value and the corresponding total variance, determine the full-order sensitivity index for each parameter to be screened, including: S301, based on the values ​​of each second function, determine the total variance corresponding to each parameter to be screened.

[0086] Among them, each function value includes the first function value corresponding to the mixing matrix. The second function value corresponding to each parameter sample matrix ( and k indicates which subspace it is in. For example, k=1 indicates the machine-side electrical and control subspace, k=2 indicates the grid-side conventional dynamic subspace, and k=3 indicates the fault traversal subspace.

[0087] Total variance, the statistical variance of all second function values ​​(from parameter sample matrices A and B), quantifies the overall fluctuation of the trajectory error objective function value when all the parameters to be screened change simultaneously and independently within their range of values.

[0088] Specifically, in determining each function value, including the first function value corresponding to the mixing matrix. And the second function value corresponding to each parameter sample matrix ( and After that, it can be based on the values ​​of each second function ( and ), determine the total variance corresponding to each parameter to be screened, total variance The formula is as follows: ; in, This represents the initial sample size, which is the number of initial random sample groups set when generating matrices A and B using Sobol low-discrepancy sequences. This represents the mean of the model's output response, which is obtained by averaging the output objective function values ​​from all N simulations.

[0089] It should be noted that before solving for the total variance, the total variance, the first-order sensitivity index, and the full-order sensitivity index can be derived based on the simulation model of the wind turbine and its error evaluation function. The derivation process is as follows: Assuming that the simulation model of the wind turbine and its error evaluation function can be abstracted into a nonlinear mapping ,in for The input parameter space is independent, and each parameter has been normalized to the closed interval [0, 1].

[0090] like Square-integrable, it can be decomposed into the sum of functional terms with increasing dimensions: ; in, It is a constant term. It is a parameter Independently varying effect terms, It is the interaction effect term between parameters.

[0091] Under the following conditions: ; For 1≤ k ≤ s ,1≤ s ≤ d We have the following formula: ; in Indicates exclusion of the first Variables of a subset of It is the corresponding support set. Assume... It is square-integrable, by applying... Square and in By integrating, we can obtain: ; Based on the expression for variance, the left side represents the variance of the model response, therefore we can obtain: ; Due to the orthogonality condition, the total variance of the model output V ( Y It can be uniquely decomposed into the sum of the variances of its subvariants: ; in, For parameters Output variance caused by individual variations; For parameters and Interaction variance caused by combined effects.

[0092] Based on the above formula, this invention focuses on evaluating two types of core sensitivity indicators: First-order sensitivity index Characterization parameters The contribution of individual variations to the total variance of the system output: ; Full-order sensitivity index Characterization parameters The sum of the independent effects of the parameter and the synergistic effects produced by its coupling with all other parameters: ; in Indicates except The variance caused by all other parameters besides the above.

[0093] S302, based on the first function value and each of the second function values, determine the first-order sensitivity index of each parameter to be screened.

[0094] The first-order sensitivity index measures the contribution of a single parameter to the output variance when the parameter changes independently, excluding the interaction effects between that parameter and other parameters. It reflects how much output fluctuation can be explained by a single parameter if only one parameter can be changed.

[0095] Specifically, in determining each function value, including the first function value corresponding to the mixing matrix. The second function value corresponding to each parameter sample matrix ( and ), and total variance Then, based on the first function value and the values ​​of each second function ( and ), determine the first-order sensitivity index for each parameter to be screened, the first-order sensitivity index The formula is as follows: ; Before calculating the computationally inexpensive full-order sensitivity exponent, the first-order exponent can serve as a rapid pre-screening tool. For parameters with extremely high first-order sensitivity exponents (such as...),... (>0.8), its interaction effect is usually not so low as to be negligible, and can be directly determined as the dominant parameter without the need to calculate the complex full-order exponent.

[0096] S303, for each parameter to be screened, based on the first function value, the first sensitivity index, and the total variance, determine the full-order sensitivity index corresponding to each parameter to be screened.

[0097] Specifically, in determining each function value, including the first function value corresponding to the mixing matrix. The second function value corresponding to each parameter sample matrix ( and ), and total variance Then, for each parameter to be filtered, based on the first function value... The second function value ( or ), and total variance Determine the full-order sensitivity index for each parameter to be screened. The formula for the full-order sensitivity index is as follows: ; In this embodiment, firstly, the full-order sensitivity corresponding to the parameters to be screened is determined, and global sensitivity analysis is performed to accurately quantify the independent contribution and coupling effect of the parameters, significantly enhancing the accuracy and physical reliability of the screening results. Secondly, an accurate first-order sensitivity index is determined to reflect the degree of contribution of each parameter individually to the dynamic response of the system. Parameter pre-screening is performed based on the first-order sensitivity index, balancing the accuracy and efficiency of parameter screening.

[0098] In one embodiment, the method for screening the dominant parameters of a wind turbine further includes: determining the parameters to be screened whose full-order sensitivity index is less than or equal to a preset threshold as fixed empirical values; and performing parameter identification for the wind turbine based on the fixed empirical values ​​and the dominant parameters to obtain the parameter identification results of the wind turbine.

[0099] Among them, fixed empirical values ​​refer to parameter values ​​that do not participate in the optimization iteration during the parameter identification process and are assigned constant values. These parameters have low overall sensitivity indices, meaning that their impact on the dynamic response of wind turbines is negligible. Therefore, there is no need to identify them through complex optimization algorithms; known and reliable values ​​can be used directly.

[0100] The parameter identification result can refer to the final set of parameter values ​​output after solving the parameter identification problem through an optimization algorithm. Specifically, the parameter identification result can include the optimal estimate of the dominant parameter (obtained through optimization calculation) and the original value of the fixed empirical parameter.

[0101] In this embodiment, firstly, a few dominant parameters are selected from dozens or even hundreds of parameters to be identified, and parameter identification is performed based on these dominant parameters, thus improving the efficiency of parameter identification. Secondly, the fixed values ​​(empirical values) of non-dominant parameters have good universality across different units and operating conditions; the identification results of the dominant parameters can be directly used for unit performance diagnosis, enhancing the interpretability and transferability of the parameter identification results.

[0102] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0103] Based on the same inventive concept, this application also provides a device for screening the dominant parameters of a wind turbine to implement the above-described method for screening the dominant parameters of a wind turbine. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the device for screening the dominant parameters of a wind turbine provided below can be found in the limitations of the method for screening the dominant parameters of a wind turbine described above, and will not be repeated here.

[0104] In one embodiment, such as Figure 5 As shown, a device for screening the dominant parameters of a wind turbine generator is provided, comprising: The parameter acquisition module 501 is used to divide multiple parameters to be screened in the wind turbine into the turbine-side electrical and control subspace, the grid-side conventional dynamic subspace, and the fault ride-through subspace based on the decoupling characteristics and control principle of the turbine topology, and to match the optimal excitation and disturbance conditions for each subspace. The parameter processing module 502 is used to generate two independent and highly uniform parameter sample matrices based on the perturbation sequence for each subspace, and to construct a hybrid matrix based on the two parameter sample matrices; wherein, each parameter sample matrix includes the physical values ​​of the corresponding parameter to be screened under different simulation states in the subspace. The simulation module 503 is used to input the sample matrices and the mixture matrix of each parameter into the simulation model of the wind turbine, and output the simulation output datasets corresponding to each parameter sample matrix and the mixture matrix. The simulation output datasets include the simulation output quantities of the active power, reactive power, DC bus voltage and rotor speed of the wind turbine. The function value determination module 504 is used to determine the function values ​​of the trajectory error objective function of the subspace corresponding to the parameter sample matrix and the mixing matrix, respectively, based on each simulation output quantity and the benchmark transient output quantity corresponding to each simulation output quantity for each subspace. The parameter filtering module 505 is used to determine the full-order sensitivity index corresponding to each parameter to be filtered based on each function value and the total variance corresponding to each function value, and to determine the parameters to be filtered with a full-order sensitivity index greater than a preset threshold as the dominant parameters; wherein, the dominant parameters are used for parameter identification of wind turbine units.

[0105] Each module in the aforementioned wind turbine master parameter screening device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0106] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores the parameter data to be filtered. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for filtering the dominant parameters of a wind turbine.

[0107] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0108] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method steps.

[0109] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above method steps.

[0110] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the above-described method steps.

[0111] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0112] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0113] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for selecting the dominant parameters of a wind turbine generator, characterized in that, The method includes: Based on the decoupling characteristics and control principle of the unit topology, multiple parameters to be screened for the wind turbine are divided into the machine-side electrical and control subspace, the grid-side conventional dynamic subspace, and the fault ride-through subspace, and the optimal excitation and disturbance conditions are matched for each subspace. For each subspace, two independent and highly uniform parameter sample matrices are generated based on the perturbation sequence, and a hybrid matrix is ​​constructed based on the two parameter sample matrices; wherein, each parameter sample matrix includes the physical values ​​of the corresponding parameter to be screened under different simulation states in the subspace; Each parameter sample matrix and the hybrid matrix are input into the simulation model of the wind turbine, and the simulation output datasets corresponding to each parameter sample matrix and the hybrid matrix are output; wherein, the simulation output datasets include the simulation output quantities of the active power, reactive power, DC bus voltage and rotor speed of the wind turbine. For each subspace, based on each simulation output and the corresponding baseline transient output, the parameter sample matrix and the mixture matrix are respectively determined to correspond to the trajectory error objective function of the subspace. Based on each function value and the total variance corresponding to each function value, the full-order sensitivity index corresponding to each parameter to be screened is determined, and the parameters to be screened whose full-order sensitivity index is greater than a preset threshold are determined as dominant parameters; wherein, the dominant parameters are used for parameter identification of the wind turbine.

2. The method according to claim 1, characterized in that, The step of inputting each of the parameter sample matrices and the mixture matrix into the simulation model of the wind turbine, and outputting the simulation output datasets corresponding to each of the parameter sample matrices and the mixture matrix, includes: For each of the subspaces, determine the simulation conditions for that subspace; According to the simulation conditions, the parameters in each parameter sample matrix and the corresponding hybrid matrix are respectively input into the simulation model of the wind turbine, and the simulation output datasets corresponding to each parameter sample matrix and the simulation output datasets corresponding to the hybrid matrix are output.

3. The method according to claim 1, characterized in that, For each subspace, determining the function values ​​of the trajectory error objective function corresponding to the parameter sample matrix and the mixing matrix for each subspace based on the simulation output and the corresponding baseline transient output includes: For each subspace, the baseline transient output of the simulation model under the baseline operating condition is obtained; the baseline operating condition is the same as the simulation operating condition of the simulation output. Based on the baseline transient output and each of the simulation outputs, the function values ​​of the trajectory error objective function corresponding to the hybrid matrix and each of the parameter sample matrices in the subspace are determined.

4. The method according to claim 1, characterized in that, Each of the function values ​​includes a first function value corresponding to the mixing matrix; and a second function value corresponding to each of the parameter sample matrices. The step of determining the full-order sensitivity index corresponding to each of the aforementioned function values ​​and the total variance corresponding to each of the aforementioned function values ​​includes: Based on the first function value, each of the second function values, and the total variance, a first-order sensitivity index is determined for each of the parameters to be screened; wherein, the first-order sensitivity index is used to measure the degree of contribution of a single parameter to the output variance when the parameter changes independently; For each of the parameters to be screened, based on the first function value, the second function value, and the total variance, the full-order sensitivity index corresponding to each parameter to be screened is determined.

5. The method according to claim 1, characterized in that, The method further includes: The parameters to be screened whose full-order sensitivity index is less than or equal to a preset threshold are determined as fixed empirical values; Based on the fixed empirical values ​​and the dominant parameters, parameter identification is performed on the wind turbine to obtain the parameter identification results of the wind turbine.

6. A device for screening the dominant parameters of a wind turbine generator, characterized in that, The device includes: The parameter partitioning module is used to partition multiple parameters to be screened in the wind turbine into the turbine-side electrical and control subspace, the grid-side conventional dynamic subspace, and the fault ride-through subspace based on the decoupling characteristics and control principles of the turbine topology, and to match the optimal excitation and disturbance conditions for each subspace. The parameter processing module is used to generate two independent and highly uniform parameter sample matrices based on the perturbation sequence for each subspace, and to construct a hybrid matrix based on the two parameter sample matrices; wherein each parameter sample matrix includes the physical values ​​of the corresponding parameter to be screened under different simulation states in the subspace; The simulation module is used to input the parameter sample matrices and the hybrid matrix into the simulation model of the wind turbine, and output the simulation output datasets corresponding to each parameter sample matrix and the hybrid matrix; wherein, the simulation output datasets include the simulation output quantities of the active power, reactive power, DC bus voltage and rotor speed of the wind turbine. The function value determination module is used to determine, for each of the subspaces, the function values ​​of the trajectory error objective function corresponding to the parameter sample matrix and the mixture matrix respectively, based on each of the simulation output quantities and the reference transient output quantity corresponding to each of the simulation output quantities; The parameter filtering module is used to determine the full-order sensitivity index corresponding to each of the function values ​​and the total variance corresponding to each function value, and to determine the parameters to be filtered that have a full-order sensitivity index greater than a preset threshold as the dominant parameters; wherein, the dominant parameters are used for parameter identification of the wind turbine.

Images