Wind driven generator transmission chain rigid-flexible coupling multi-body dynamics analysis method based on dynamic mode decomposition

Abstract

The invention belongs to the technical field of multi-body dynamics analysis, and discloses a wind driven generator transmission chain rigid-flexible coupling multi-body dynamics analysis method based on dynamic mode decomposition, and the method comprises the steps: firstly, enabling multi-degree-of-freedom time series data to be non-linearly embedded into a high-dimensional feature space through an encoder neural network; extracting a dominant mode by utilizing intrinsic orthogonal decomposition (POD), and constructing a low-dimensional feature space; parameterized dynamic mode decomposition and radial basis function regression are adopted, a mapping relation between system parameters and Koopman operators is established, and accurate prediction of dynamic characteristics under variable working conditions is achieved; and finally, reconstructing a physical response through a decoder, and optimizing model parameters in combination with an error driving mechanism. The problems that a traditional method is low in calculation efficiency, poor in nonlinear adaptability and difficult in multi-parameter coupling prediction are effectively solved, the efficiency and precision of transmission chain dynamic analysis are remarkably improved, and reliable technical support is provided for state monitoring and service life prediction of the wind turbine generator.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-body dynamics analysis, and in particular relates to a rigid-flexible coupling multi-body dynamics analysis method for a wind turbine transmission chain based on dynamic mode decomposition. Background Art

[0002] As the core component of a complex electromechanical coupling system, the wind turbine transmission chain has significant engineering value in improving the operational reliability of the entire machine, reducing component fatigue damage, and optimizing control strategies. In recent years, it has received widespread attention in the field of new energy equipment design and operation and maintenance.

[0003] Current research on the dynamic characteristics of transmission chains is mostly based on traditional finite element models or multibody dynamics simulation methods. These methods typically rely on high-resolution modeling and full-order numerical solutions, requiring the construction of a large-scale system of nonlinear differential equations involving components such as gearboxes, main bearings, and couplings. This results in computational complexity that increases exponentially with the number of degrees of freedom. In particular, when considering time-varying fluctuations in wind loads, flexible deformations of transmission chains, and multi-physics coupling effects, conventional numerical solvers face bottlenecks such as repeated assembly of mass / stiffness matrices and high-frequency time-step integration, making it difficult to meet the requirements of real-time analysis and parameter optimization.

[0004] To address computational efficiency challenges, existing research has largely employed model order reduction techniques, achieving dimensionality reduction by truncating low-energy modes. However, traditional POD-Galerkin-type methods rely on a global projection basis for modal truncation. When faced with variations in transmission chain parameters under multiple operating conditions (such as variable gear mesh stiffness and dynamic damping ratio), reconstruction errors accumulate nonlinearly, making it impossible to effectively maintain the predictability of dynamic characteristics. Furthermore, parameterized order reduction models based on offline training typically employ linear interpolation or low-order polynomial regression to construct parameter-modal associations, making it difficult to capture the strongly nonlinear mapping relationships between complex control systems and structural dynamics.

[0005] In recent years, data-driven dynamic analysis methods based on the Koopman operator have shown potential for resolving these issues. By embedding nonlinear systems into an infinite-dimensional linear space, they can achieve global linearization of nonlinear dynamics in a high-dimensional feature space. However, existing Koopman methods often rely on batch processing algorithms such as dynamic mode decomposition (DMD), requiring independent calculation of approximate operators for each parameter configuration. This results in multiple SVD decompositions and matrix inversion operations, resulting in a surge in computational costs in multi-parameter coupled transmission chain analysis scenarios. Furthermore, the traditional DMD framework fails to fully integrate the self-feature extraction capabilities of deep learning, making it difficult to effectively handle the complex motion mode aliasing phenomena of flexible multi-body systems, limiting its practical application in industrial equipment. Summary of the Invention

[0006] The purpose of the present invention is to provide a wind turbine transmission chain rigid-flexible coupling multi-body dynamics analysis method based on dynamic mode decomposition to solve the technical problems of multiple model degrees of freedom, low time domain solution efficiency, and difficulty in extracting multi-parameter coupling features in the rigid-flexible coupling dynamics analysis of wind turbine transmission chains.

[0007] To solve the above technical problems, the specific technical solutions of the wind turbine transmission chain rigid-flexible coupling multi-body dynamics analysis method based on dynamic mode decomposition of the present invention are as follows: A method for analyzing rigid-flexible coupled multi-body dynamics of a wind turbine transmission chain based on dynamic mode decomposition includes the following steps: S1: Generate and extract original multi-degree-of-freedom time series data based on measured data or existing high-fidelity dynamic models; S2: Define two neural network structures including encoder and decoder, and initialize neural network parameters; S3: Map the time series data to a high-dimensional feature space through an encoder neural network to form a high-dimensional system state dataset; S4: Perform intrinsic orthogonal decomposition on the high-dimensional system state dataset to extract the dominant modes and generate a low-dimensional feature space; S5: Pre-train the regression model on low-dimensional feature data under different parameter settings and optimize the parameters of the Koopman operator predictor; S6: For different control parameters of the dynamic system, the Koopman operator of the trained regressor output response is used to perform time-step iterative prediction of the current low-dimensional feature state; S7: Perform the inverse transformation of the intrinsic orthogonal decomposition on the converged low-dimensional feature prediction results to reconstruct them into high-dimensional feature space system state data; S8: Decoder neural network is used to map the high-dimensional feature space system state data back to the physical space to obtain the dynamic response; S9: Determine the convergence of the prediction results based on the preset error threshold. If not, update the network model parameters and repeat S3 to S8. S10: Based on the multi-degree-of-freedom time series data output by S8, the dynamic characteristics of key components of the transmission chain under different system parameters are calculated.

[0008] Furthermore, the S2 includes the following steps: S2.1: Define the encoder-decoder network architecture. Based on the spatiotemporal coupling characteristics of the state variables in the multi-degree-of-freedom physical space of the drive train system, construct a deep fully connected neural network to map the system motion state data in the physical space into a high-dimensional feature space. S2.2: Network parameter initialization and regularization strategy, using the Xavier initialization method to give the network initial weights, in the network training loss function, impose L2 Regularization constraint term, the neural network bias term is initialized to a zero vector.

[0009] Furthermore, the S4 includes the following steps: S4.1: Construct a time-domain snapshot matrix, arranging the system states in the high-order feature space output from the encoder into a snapshot matrix in time series; S4.2: Compute the covariance matrix and perform singular value decomposition; S4.3: Determine the modal cutoff number and calculate the cumulative energy fraction to determine the cutoff order; S4.4: Generate a low-dimensional feature space projection and project the high-dimensional feature data into the dominant modal basis vector space.

[0010] Furthermore, the S5 includes the following steps: S5.1: Construction of parameterized dynamic modal data set. For each system parameter configuration, obtain the system state time series data corresponding to the system in the feature space through step S4, construct the shift data matrix, and finally form the parameter-modal data set; S5.2: Dynamic Mode Decomposition solves the approximate Koopman operator and executes the DMD core algorithm for the system state timing sequence matrix corresponding to each system parameter configuration; S5.3: Parameterized Koopman operator regression modeling, using the radial basis function regression model to construct the mapping relationship between system parameter configuration and Koopman operator.

[0011] Furthermore, the S5.2 specifically includes: S5.2.1: Perform singular value decomposition on the shifted data matrix; S5.2.2: Construct the Koopman operator matrix.

[0012] Furthermore, the S5.3 specifically includes: S5.3.1: Define the RBF kernel function; S5.3.2: Compute the RBF regression model; S5.3.3: The objective function is set to Frobenius norm loss.

[0013] Furthermore, the S6 includes the following steps: S6.1: Online parameter input and Koopman operator prediction. For real-time input system parameters, the RBF regression model trained in S5.3 is called to output the corresponding Koopman operator estimation value. S6.2: Dynamic iterative prediction in feature space, based on the initial low-dimensional features, performing T-step time step calculation.

[0014] Furthermore, the step S7 includes the following steps: S7.1: Read the basis vectors and mean of the dominant mode, and load the modal matrix and system eigenspace mean vector stored in S4; S7.2: Perform linear back-projection reconstruction to reconstruct the high-dimensional feature space based on the low-dimensional system state feature sequence.

[0015] Furthermore, the S9 includes the following steps: S9.1: Multi-objective loss function construction, designing a composite loss function to simultaneously optimize network encoding-decoding accuracy, dynamic linearity, and state prediction performance; S9.2: Loss convergence threshold judgment, set the convergence judgment threshold and the maximum number of iterations, and calculate the composite loss residual ratio; S9.3: Backpropagation optimization, using the Adam optimizer of the neural network to update the network parameters. The optimization is performed after the network training loss is calculated. The gradient is first calculated, and then the momentum adaptive update strategy is executed.

[0016] The rigid-flexible coupled multi-body dynamics analysis method for a wind turbine transmission chain based on dynamic mode decomposition of the present invention has the following advantages: 1. Significantly improve computing efficiency: By combining a deep encoder-decoder network with POD order reduction technology, high-dimensional nonlinear dynamic systems are mapped into a low-dimensional feature space for linear evolution prediction, significantly reducing computational complexity. Compared to traditional finite element or multibody dynamics full-order simulations, this method significantly reduces computational time and resource consumption while maintaining accuracy, making it suitable for rapid parametric analysis and real-time state prediction.

[0017] 2. Enhanced nonlinear dynamic feature extraction capabilities: The introduction of deep neural networks for nonlinear embedding of feature space effectively overcomes the limitations of traditional linear modal decomposition methods in flexible body motion reconstruction, and can more accurately capture the complex dynamic behavior of the transmission chain system under variable working conditions, such as nonlinear responses caused by changes in gear meshing stiffness and time-varying damping.

[0018] 3. Improve the adaptability of multi-parameter coupling: By combining the parameterized Koopman operator with the RBF regression model, a nonlinear mapping relationship between system parameters and dynamic characteristics is constructed, which enhances the model's adaptability to different operating parameters (such as wind speed, load, etc.) and avoids the mode mismatch problem that occurs in traditional reduced-order models when parameters change.

[0019] 4. Achieve end-to-end efficient prediction and reconstruction: By combining the encoding-decoding structure with POD forward and inverse transforms, a bidirectional high-precision mapping from physical space to feature space and then back to physical space is achieved. After a single training session, the full-cycle dynamic response under multiple parameter combinations can be quickly generated, supporting the rapid design and optimization of the transmission chain system.

[0020] 5. Possess good engineering application prospects: This method provides strong technical support for real-time status monitoring, fatigue life prediction, fault diagnosis and variable operating condition control of wind turbine transmission chains. It is particularly suitable for multi-body dynamics analysis and intelligent operation and maintenance scenarios of large wind turbines, and has good practicality and promotion value.

[0021] In summary, the present invention not only achieves innovation in theoretical methods, but also demonstrates significant advantages in efficiency, accuracy, and robustness in engineering practice, providing a new solution for the dynamic modeling and analysis of complex electromechanical systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of a method according to an embodiment of the present invention; Figure 2a Schematic diagram of multi-body dynamics modeling of a rigid-flexible coupling transmission chain for the implementation case; Figure 2b Schematic diagram of the finite element mesh of the flexible body's principal axis; Figure 3a A comparison of the node strains (absolute values) of the principal axis of the flexible body predicted by the method proposed in the present invention and PDMD; Figure 3b Schematic diagram of the relationship between the method proposed in the present invention and the PDMD prediction error and the prediction time. DETAILED DESCRIPTION

[0023] In order to better understand the purpose, structure and function of the present invention, the following further describes in detail the rigid-flexible coupling multi-body dynamics analysis method of the wind turbine transmission chain based on dynamic mode decomposition in conjunction with the accompanying drawings.

[0024] This paper proposes a novel method for analyzing the rigid-flexible coupled multibody dynamics of wind turbine transmission chains based on deep Koopman feature learning. By constructing a deep encoding network, it achieves nonlinear embedding of multi-degree-of-freedom time series data into a high-dimensional feature space. Proper Orthogonal Decomposition (POD) is then used to extract dominant modes and establish a low-dimensional quasi-steady-state feature space. A parameterized Koopman operator predictor is established, and a pre-trained regression model is used to achieve online mapping and association between different operating parameters and dynamic operators. A reconstruction and back-projection algorithm is used to achieve high-precision full-order restoration of low-dimensional feature data. A deep decoding network is then combined to establish a bidirectional nonlinear mapping channel between feature space and physical space. This method effectively addresses the challenge of efficiently predicting the cross-scale dynamic evolution of rigid-flexible coupled systems through end-to-end feature evolution modeling.

[0025] like Figure 1 As shown, the rigid-flexible coupling multi-body dynamics analysis method of a wind turbine transmission chain based on dynamic mode decomposition of the present invention includes the following steps: S1: Generate and extract original multi-degree-of-freedom time series data based on measured data or existing high-fidelity dynamic models.

[0026] S2: Define two neural network structures including encoder and decoder, and initialize the neural network parameters.

[0027] S2.1: Define the encoder-decoder network architecture based on the spatiotemporal coupling characteristics of the state variables in the multi-degree-of-freedom physical space of the transmission chain system (displacement, velocity, acceleration field in The encoder network structure is defined as follows:

[0028] in, represents the encoder network function, Represents the state vector of the system at any time in physical space, which can be summarized as the system's degree of freedom in multi-body dynamics; is the trainable parameter of the encoder neural network, denoted as ; represents the number of encoder network layers, They represent the encoder network The network weight parameters and bias of the layer; Indicates the The activation function of the layer, L Indicates the number of network layers.

[0029] The decoder network is symmetrical to the encoder and can be expressed as:

[0030] in, represents the decoder network function, Represents the system state vector at any time in the high-order feature space; is the trainable parameter of the decoder neural network, denoted as , represents the number of decoder network layers, Denotes the decoder network The network weight parameters and bias of the layer.

[0031] S2.2: Network parameter initialization and regularization strategy, using the Xavier initialization method to assign initial weights to the network, its mathematical expression is:

[0032] in, represents a uniform random distribution, and Respectively The input and output dimensions of the layer network. In the network training loss function, the L2 Regularization constraint , the expression is:

[0033] in, is a hyperparameter, and in the experiment it is ; is the Frobenius norm.

[0034] Neural network bias and Initialized to the zero vector.

[0035] S3: Map the time series data to a high-dimensional feature space through an encoder neural network to form a high-dimensional system state dataset; S4: Perform proper orthogonal decomposition (POD) on the high-dimensional system state dataset to extract the dominant modes and generate a low-dimensional feature space; S4.1: Construct a time-domain snapshot matrix, arranging the system states in the high-order feature space output from the encoder into a snapshot matrix in time series ,in is the dimension of system state variables in high-dimensional feature space, The column vector of the time sampling point matrix corresponds to the feature space state:

[0036] S4.2: Calculate the covariance matrix and perform singular value decomposition. The covariance matrix can be expressed as:

[0037] in and is an orthogonal matrix, For the diagonal matrix of singular values ​​arranged in descending order, in order to achieve order reduction, it is necessary to extract Order dominant mode matrix , which can be expressed as:

[0038] S4.3: Determine the modal cutoff number and assume an energy threshold , calculate the cumulative energy ratio to determine the truncation order :

[0039] in For the singular values, accumulated through a loop ,until .

[0040] S4.4: Generate low-dimensional feature space projection and transform high-dimensional feature data Projected into the dominant mode basis vector space, it is expressed as , the expression is as follows:

[0041] S5: Pre-train the regression model on low-dimensional feature data under different parameter settings and optimize the parameters of the Koopman operator predictor; S5.1: Parameterized dynamic modal dataset construction, for each system parameter configuration , obtain the system state time series data corresponding to the system in the feature space through step S4 , construct the shift data matrix and , the expression is as follows:

[0042] Final composition parameters-modal dataset ,in is the total number of parameter samples.

[0043] S5.2: Dynamic mode decomposition solves the approximate Koopman operator, for each system parameter configuration The corresponding system state timing sequence matrix executes the DMD core algorithm, including: S5.2.1: Shifting the Data Matrix Perform singular value decomposition (SVD):

[0044] S5.2.2: Construct the Koopman operator matrix: .

[0045] S5.3: Parameterized Koopman operator regression modeling, using radial basis function (RBF) regression model to build system parameter configuration To Koopman operator The mapping relationship includes: S5.3.1: Define the RBF kernel function:

[0046] in is the center point of the uniform distribution in parameter space, is the bandwidth function, is the number of basis functions.

[0047] S5.3.2: The regression model is expressed as:

[0048] in is the coefficient tensor, is the bias matrix.

[0049] S5.3.3: The objective function is set to Frobenius norm loss:

[0050] in is the regularization coefficient.

[0051] S6: For different control parameters of the dynamic system, the Koopman operator of the trained regressor output response is used to perform time-step iterative prediction of the current low-dimensional feature state; S6.1: Online parameter input and Koopman operator prediction, for real-time input system parameters , call the RBF regression model trained in S5.3 and output the corresponding Koopman operator estimate ; S6.2: Dynamic iterative prediction in feature space based on initial low-dimensional features ,implement Time stepping calculation:

[0052] S7: Perform POD inverse transformation on the converged low-dimensional feature prediction results to reconstruct them into high-dimensional feature space system state data; S7.1: Read the basis vectors and mean of the dominant mode, and load the modal matrix stored in S4 and the mean vector of the system feature space , S7.2: Perform linear back-projection reconstruction based on the low-dimensional system state characteristic sequence Reconstructing high-dimensional feature space:

[0053] S8: Decoder neural network is used to map the high-dimensional feature space system state data back to the physical space to obtain the dynamic response; S9: Determine the convergence of the prediction results based on the preset error threshold. If not, update the network model parameters and repeat S3 to S8. S9.1: Multi-objective loss function construction. Design a composite loss function to simultaneously optimize network encoding-decoding accuracy, dynamic linearity, and state prediction performance. The loss function expression is as follows:

[0054] in 、 and They are reconstruction loss, linear dynamic loss and prediction loss, and their expressions are as follows:

[0055] in 、 and is the weight coefficient, is the regularization strength, and Represent the calculation process of the encoder network and the decoder network respectively, represents the mean square error.

[0056] S9.2: Loss convergence threshold judgment, set the convergence judgment threshold and the maximum number of iterations , calculate the compound loss residual ratio:

[0057] when satisfy or When the network training is considered to have converged, in the test, set ,set up .

[0058] S9.3: Backpropagation optimization, using the Adam optimizer of the neural network to update the network parameters, performs optimization after the network training loss is calculated, and first calculates the gradient:

[0059] Then execute the momentum adaptive update strategy:

[0060] in 、 is the exponentially decaying momentum term, The learning rate for network training is set to , To avoid smoothing items where the denominator is close to zero and causes numerical overflow, it is usually set .

[0061] S10: Based on the multi-degree-of-freedom time series data output by S8, the dynamic characteristics of the key components of the transmission chain, such as displacement, speed, stress and strain, are calculated under different system parameters.

[0062] Implementation Cases: In order to verify the effectiveness of the method of the present invention, this example introduces the method proposed in the present invention into the rigid-flexible coupled multi-body dynamic model of the three-stage planetary gearbox transmission chain system of a real in-service wind turbine established by Simpack, and compares it with the parametric dynamic mode decomposition method (DMD for parametric dynamic systems, PDMD) proposed by F Andreuzzi et al. (Andreuzzi F, Demo N, Rozza G. A dynamic mode decomposition extension for the forecasting of parametric dynamical systems[J]. SIAMJournal on Applied Dynamical Systems, 2023, 22(3): 2432-2458.).

[0063] like Figure 2a Figure 2b As shown in the test case, the working condition of the multi-body dynamics simulation is set to stable laminar wind, and the wind speed range is set to Meters per second. The strain data of 4000 nodes b of the flexible body main axis in the multi-body dynamics model under different working conditions within 50 seconds are extracted as the initial time series data. The wind speed is set as the input of the system and model. The model training data selects a total of 19 sets of data under wind speed conditions of 3 to 21 meters per second, and predicts the strain of the flexible body main axis under a wind speed of 11.5 meters per second. The network training uses the Adam optimizer, and the initial learning rate is set to 1×10 -3 .

[0064] The flexible body principal axis node strain results of the multi-body dynamics system solved by Simpack under the steady laminar wind condition with a wind speed of 11.5 meters per second are used as the benchmark solution. The results obtained by the method of the present invention are compared with the flexible body principal axis node strain results solved by the method proposed by F Andreuzzi et al. The results are as follows: Figure 3a 、 Figure 3b shown.

[0065] Under the same initial conditions and the same test case, the method of the present invention significantly improves the accuracy of solving the displacement of the flexible body main axis under different working conditions compared to the parameterized dynamic mode decomposition method proposed by FAndreuzzi et al., which shows that the method of the present invention can greatly improve the efficiency and accuracy of rigid-flexible coupling multi-body dynamics analysis. This is mainly because the method introduces an automatic encoder based on a neural network, which can better capture the dynamic evolution characteristics of complex multi-body dynamics systems in the original physical space. At the same time, the parameterized Koopman operator based on the RBF kernel function regression model introduced can dynamically capture the dynamic modes under different system state parameters, thereby realizing efficient and accurate analysis of the dynamic characteristics of rigid-flexible coupling multi-body dynamics systems.

[0066] It will be understood that the present invention is described by way of some embodiments, and it will be appreciated by those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are intended to be protected by the present invention.

Claims

1. A method for analyzing rigid-flexible coupled multi-body dynamics of a wind turbine transmission chain based on dynamic mode decomposition, characterized in that: The steps include: S1: Generate and extract original multi-degree-of-freedom time series data based on measured data or existing high-fidelity dynamic models; S2: Define two neural network structures including encoder and decoder, and initialize neural network parameters; S3: Map the time series data to a high-dimensional feature space through an encoder neural network to form a high-dimensional system state dataset; S4: Perform intrinsic orthogonal decomposition on the high-dimensional system state dataset to extract the dominant modes and generate a low-dimensional feature space; S5: Pre-train the regression model on low-dimensional feature data under different parameter settings and optimize the parameters of the Koopman operator predictor; S6: For different control parameters of the dynamic system, the Koopman operator of the trained regressor output response is used to perform time-step iterative prediction of the current low-dimensional feature state; S7: Perform the inverse transformation of the intrinsic orthogonal decomposition on the converged low-dimensional feature prediction results to reconstruct them into high-dimensional feature space system state data; S8: Decoder neural network is used to map the high-dimensional feature space system state data back to the physical space to obtain the dynamic response; S9: Determine the convergence of the prediction results based on the preset error threshold. If not, update the network model parameters and repeat S3 to S8. S10: Based on the multi-degree-of-freedom time series data output by S8, the dynamic characteristics of key components of the transmission chain under different system parameters are calculated.

2. The method for analyzing rigid-flexible coupled multi-body dynamics of a wind turbine transmission chain based on dynamic mode decomposition according to claim 1 is characterized in that: The S2 comprises the following steps: S2.1: Define the encoder-decoder network architecture. Based on the spatiotemporal coupling characteristics of the state variables in the multi-degree-of-freedom physical space of the drive train system, construct a deep fully connected neural network to map the system motion state data in the physical space into a high-dimensional feature space. S2.2: Network parameter initialization and regularization strategy, using the Xavier initialization method to give the network initial weights, in the network training loss function, impose L2 Regularization constraint term, the neural network bias term is initialized to a zero vector.

3. The method for analyzing rigid-flexible coupling multi-body dynamics of a wind turbine transmission chain based on dynamic mode decomposition according to claim 1 is characterized in that: The S4 comprises the following steps: S4.1: Construct a time-domain snapshot matrix, arranging the system states in the high-order feature space output from the encoder into a snapshot matrix in time series; S4.2: Compute the covariance matrix and perform singular value decomposition; S4.3: Determine the modal cutoff number and calculate the cumulative energy fraction to determine the cutoff order; S4.4: Generate a low-dimensional feature space projection and project the high-dimensional feature data into the dominant modal basis vector space.

4. The method for analyzing rigid-flexible coupling multi-body dynamics of a wind turbine transmission chain based on dynamic mode decomposition according to claim 1 is characterized in that: The S5 comprises the following steps: S5.1: Construction of parameterized dynamic modal data set. For each system parameter configuration, obtain the system state time series data corresponding to the system in the feature space through step S4, construct the shift data matrix, and finally form the parameter-modal data set; S5.2: Dynamic Mode Decomposition solves the approximate Koopman operator and executes the DMD core algorithm for the system state timing sequence matrix corresponding to each system parameter configuration; S5.3: Parameterized Koopman operator regression modeling, using the radial basis function regression model to construct the mapping relationship between system parameter configuration and Koopman operator.

5. The method for analyzing rigid-flexible coupling multi-body dynamics of a wind turbine transmission chain based on dynamic mode decomposition according to claim 4 is characterized in that: S5.2 specifically includes: S5.2.1: Perform singular value decomposition on the shifted data matrix; S5.2.2: Construct the Koopman operator matrix.

6. The method for analyzing rigid-flexible coupling multi-body dynamics of a wind turbine transmission chain based on dynamic mode decomposition according to claim 4 is characterized in that: S5.3 specifically includes: S5.3.1: Define the RBF kernel function; S5.3.2: Compute the RBF regression model; S5.3.3: The objective function is set to Frobenius norm loss.

7. The method for analyzing rigid-flexible coupling multi-body dynamics of a wind turbine transmission chain based on dynamic mode decomposition according to claim 6 is characterized in that: The S6 comprises the following steps: S6.1: Online parameter input and Koopman operator prediction. For real-time input system parameters, the RBF regression model trained in S5.3 is called to output the corresponding Koopman operator estimation value. S6.2: Dynamic iterative prediction in feature space, based on the initial low-dimensional features, performing T-step time step calculation.

8. The method for analyzing rigid-flexible coupling multi-body dynamics of a wind turbine transmission chain based on dynamic mode decomposition according to claim 1 is characterized in that: The S7 comprises the following steps: S7.1: Read the basis vectors and mean of the dominant mode, and load the modal matrix and system eigenspace mean vector stored in S4; S7.2: Perform linear back-projection reconstruction to reconstruct the high-dimensional feature space based on the low-dimensional system state feature sequence.

9. The method for analyzing rigid-flexible coupled multi-body dynamics of a wind turbine transmission chain based on dynamic mode decomposition according to claim 1 is characterized in that: The S9 comprises the following steps: S9.1: Multi-objective loss function construction, designing a composite loss function to simultaneously optimize network encoding-decoding accuracy, dynamic linearity, and state prediction performance; S9.2: Loss convergence threshold judgment, set the convergence judgment threshold and the maximum number of iterations, and calculate the composite loss residual ratio; S9.3: Backpropagation optimization, using the Adam optimizer of the neural network to update the network parameters. The optimization is performed after the network training loss is calculated. The gradient is first calculated, and then the momentum adaptive update strategy is executed.

Images