A wind turbine digital twin construction method

Abstract

The application provides a wind turbine digital twin construction method, belonging to the technical field of wind power generation. The method comprises: constructing a refined mechanism model, inputting a historical meteorological data sequence into the refined mechanism model after converting the sequence into a physical field parameter sequence, and outputting a simulation output sequence; obtaining the simulation output sequence and the actual operation output sequence of the wind turbine, and calculating the derivative dynamic time warping distance of the two sequences; performing sparse processing on the parameter vector of the refined mechanism model, and screening out a key parameter subset affecting the derivative dynamic time warping distance; taking the minimization of the derivative dynamic time warping distance as an objective function, and using a moss optimization algorithm to optimize and solve the key parameter subset to obtain an optimal parameter combination; and backfilling the optimal parameter combination into the corresponding parameters of the constructed refined mechanism model to obtain a wind turbine digital twin model. The model precision is improved.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation technology, and in particular relates to a method for constructing a digital twin of a wind turbine. Background Technology

[0002] In recent years, with the continuous expansion of wind power installed capacity and the increasing complexity of grid-connected control structures, the dynamic behavior of wind power systems has exhibited strong nonlinearity, multi-temporal and spatial coupling, and high-dimensional characteristics. Especially under the background of high-proportion power electronics, the interaction between factors such as wind speed disturbances in wind farms and changes in grid strength makes the system operation exhibit typical time-varying characteristics and multi-source randomness.

[0003] Traditional simulation methods still face core challenges such as insufficient model accuracy and difficulties in virtual-real interaction in high-dimensional parameter spaces. On the one hand, existing methods do not adequately characterize the aerodynamic performance of wind turbines, tower shading effects, and structural dynamic responses; on the other hand, they lack effective evaluation indicators to quantify differences in dynamic responses, making it difficult to accurately reflect the consistency between digital virtual entities and physical entities in terms of dynamic characteristics, thus limiting the effectiveness of parameter calibration. Summary of the Invention

[0004] This invention provides a method for constructing a digital twin of a wind turbine, which at least solves the problem of insufficient model accuracy in the prior art.

[0005] This application provides a method for constructing a digital twin of a wind turbine, the method comprising:

[0006] Step S1: Construct a refined mechanism model, which includes an aerodynamic-mechanical simulation model of the wind turbine and a fourth-order electromechanical transient model of the doubly-fed generator. The aerodynamic-mechanical simulation model includes higher-order blade element momentum theory, tower shadow effect model and load model.

[0007] Step S2: Obtain historical meteorological data sequences and convert them into physical field parameter sequences. Input the physical field parameter sequences into the refined mechanism model, and the refined mechanism model outputs simulation output sequences.

[0008] Step S3: Obtain the simulation output sequence and the actual operating output sequence of the wind turbine, and use the dynamic time warping algorithm to calculate the derivative dynamic time warping distance between the actual operating output sequence and the simulation output sequence;

[0009] Step S4: Using a random gating-based grouped sparse feature selection algorithm, the parameter vector of the refined mechanism model is sparsified to filter out the subset of key parameters that affect the dynamic time warping distance of the derivative.

[0010] Step S5: Using the minimization of the derivative dynamic time warping distance as the objective function, the moss optimization algorithm is used to optimize and solve the subset of key parameters to obtain the optimal parameter combination;

[0011] Step S6: Backfill the optimal parameter combination into the corresponding parameters of the constructed refined mechanism model, complete the parameter calibration, and obtain the digital twin model of the wind turbine.

[0012] Furthermore, in step S1, the aerodynamic-mechanical simulation model includes: a blade element momentum theory model, a tower shadow wake model, and a load model;

[0013] Calculation of thrust and torque generated by a single leaf element based on leaf element momentum theory model;

[0014] The local wind speed on which the leaf element momentum theory model depends is corrected based on the tower shadow wake model;

[0015] The load model is used to receive the local wind speed after correction by the tower shadow wake model, and to solve the aerodynamic torque of the wind turbine based on the thrust generated by a single blade element micro-element calculated by the blade element momentum theory model.

[0016] Further, in step S11, the thrust and torque generated by a single leaf element are calculated based on the leaf element momentum theory model, and their expressions are as follows:

[0017]

[0018]

[0019] in, For thrust; For torque; air density; The local radius of a given leaf element location; It is the axial induction factor; Tangential induction factor; This refers to the wind turbine rotation speed; Wind speed at infinity; This is the Prandtl correction factor.

[0020] Furthermore, the expression for the Prandtl correction factor is:

[0021]

[0022]

[0023] in, This is the maximum radius of the wind turbine; The relative angle of attack is the angle of inflow. With leaf element twist angle sum; This refers to the number of leaves;

[0024] The expression for the axial induction factor is:

[0025]

[0026] in, It is the critical inducing factor. ; Correct the empirical coefficients for Glauert; The degree of leaf pigmentation; Normal force coefficient;

[0027] The expression for Glauert's corrected empirical coefficient is:

[0028]

[0029] The expression for the normal force coefficient is:

[0030]

[0031] in, The lift coefficient; This is the drag coefficient.

[0032] Further, in step S11, the local wind speed upon which the leaf element momentum theory model depends is corrected based on the tower shadow wake model, and its expression is:

[0033]

[0034]

[0035] In the formula, This is the corrected local wind speed for the leaf element unit. This represents the local wind speed deficit caused by the tower shadow effect. The dimensionless radial distance is expressed as follows:

[0036]

[0037] In the formula, for axis, for axis.

[0038] Furthermore, the load model is used to receive the local wind speed corrected by the tower shadow wake model, and to solve for the aerodynamic torque of the wind turbine based on the thrust generated by a single blade element calculated by the blade element momentum theory model, specifically including:

[0039] Based on the expression for the thrust generated by a single blade element, the increment of the aerodynamic torque of a single blade element is derived, and its expression is as follows:

[0040]

[0041] In the formula, This represents the increment of the aerodynamic torque; The chord length of the leaf element unit; The corrected local wind speed for the leaf element unit; This is the distance from the leaf element unit to the leaf root, i.e., the lever arm length; The length of a leaf element unit; The angle of entry;

[0042] Summing the increments of the aerodynamic torque of all blade elements along the blade span yields the aerodynamic torque acting on the entire rotor, expressed as:

[0043]

[0044] In the formula, This refers to the pneumatic torque.

[0045] Furthermore, a fourth-order electromechanical transient model is constructed using stator and rotor flux linkages as state variables:

[0046]

[0047] In the formula, The fundamental angular frequency; The state matrix; It is a voltage vector; It is a resistance matrix; For inductance matrix; To include synchronous angular velocity With slip angular velocity The rotation transformation matrix;

[0048] The expression for the state matrix is:

[0049]

[0050] The expression for the voltage vector is:

[0051]

[0052] The expression for the stator and rotor flux linkage is:

[0053]

[0054] In the formula, This is the stator flux linkage vector; is the rotor flux linkage vector.

[0055] Further, in step S3, the actual operating output sequence and the simulation output sequence of the wind turbine are obtained, and the derivative dynamic time warping distance between the actual operating output sequence and the simulation output sequence is calculated using a dynamic time warping algorithm, specifically including:

[0056] Let the actual output sequence be The simulation output sequence is ;

[0057] For the actual output sequence is and simulation output sequence are Each of them calculates its derivative approximation, and its expression is:

[0058]

[0059] In the formula, Indicates the actual output sequence In the Approximate values ​​of the first derivative at each time point. Represents the simulation output sequence In the Approximate values ​​of the first derivative at each time point;

[0060] Define the actual output sequence in the derivative space. The Points and simulation output sequence The The local distance between points is expressed as:

[0061]

[0062] In the formula, Indicates the actual output sequence The Points and simulation output sequence The Local distance between points;

[0063] Using dynamic programming recursion with dynamic time warping, we can derive the sequence starting from the beginning. The minimum cumulative distance to the current point is expressed as:

[0064]

[0065] In the formula, Indicates starting from the beginning of the sequence Minimum cumulative distance to the current point;

[0066] By path length For minimum cumulative distance After normalization, the normalized derivative dynamic time warped distance is obtained, and its expression is:

[0067]

[0068] In the formula, For the actual output sequence With simulation output sequence The dynamic time-normalized distance between the derivatives.

[0069] Further, in step S4, a random-gated grouped sparse feature selection algorithm is used to sparsify the structure and control parameter vectors of the wind turbine model, and to filter out a subset of key parameters that affect the dynamic time warping distance of the derivative, specifically including:

[0070] Suppose that the fan structure and control parameters, after normalization, form an N-dimensional parameter vector:

[0071]

[0072] In the formula, Represents the original parameter vector; Indicates parameters; Represents the set of real numbers; Indicates the total number of original parameters;

[0073] Before the input layer of the deep neural network, set a learnable random gating vector for each dimension parameter and its physical grouping:

[0074]

[0075] In the formula, Represents a random gate vector; Indicates the gate value;

[0076] Obtain effective inputs from actual participants in the modeling process:

[0077]

[0078] In the formula, This represents element-wise multiplication. This represents a subset of key parameters.

[0079] Further, in step S5, with minimizing the derivative dynamic time warping distance as the objective function, the moss optimization algorithm is used to optimize and solve the subset of key parameters to obtain the optimal parameter combination, specifically including:

[0080] Obtain a subset of key parameters, denoted by dimension . , ;

[0081] The search space is determined based on the engineering constraints of each parameter in the key parameter subset, and the feasible region of the parameter is defined. ;

[0082] Initialize the control parameters of the moss optimization algorithm, including population size. Maximum number of evaluations Current number of assessments ;

[0083] In the feasible region Internal random generation The initial solution for the combination of parameters is denoted as... ;

[0084] For each initial solution The RT-OpenFAST simulation framework is invoked to run the simulation and calculate the fitness based on the derivative dynamic time warping algorithm. ,in, To use the initial solution The simulation output, This refers to measured data of physical entities;

[0085] Record the current optimal solution and the current optimal solution fitness ;

[0086] The number of assessments has been updated to ;

[0087] When the number of evaluations Less than the maximum number of evaluations hour:

[0088] For each initial solution in the population Update the position and generate a new initial solution. ; Calculate the new initial solution fitness ;

[0089] If the new initial solution fitness Less than the current optimal solution fitness Then update the optimal solution. , ;

[0090] like and Then use replace ;

[0091] like or Then keep constant;

[0092] If retained Keeping the current solution unchanged, calculate the current optimal solution. With the initial solution The direction vector is used to explore and generate new individuals. ;

[0093] Generate random numbers Rand. If Rand ≤ 0.8, then use historical best solutions. For the initial solution Perform a small perturbation to generate new individuals. And assess its fitness; if it is better, update the individual;

[0094] If the current number of evaluations Reaching the maximum number of evaluations Or the fitness of the optimal solution improves by less than a preset threshold over several consecutive generations. hour:

[0095] Return the current optimal solution As the optimal combination of parameters.

[0096] As can be seen from the above technical solutions, the present invention has the following advantages:

[0097] The method for constructing a digital twin of a wind turbine provided in this application can more accurately characterize the aerodynamic performance of the wind turbine, the tower shading effect, and the structural dynamic response by constructing a refined mechanism model that includes higher-order blade element momentum theory, tower shadow effect model, and load model, thereby significantly improving the accuracy of the digital twin in aerodynamic-mechanical coupling dynamic simulation.

[0098] By employing a dynamic time warping algorithm to calculate the derivative dynamic time warping distance between the actual running output sequence and the simulated output sequence, the consistency between digital virtual entities and physical entities in dynamic response is effectively quantified, providing an accurate objective function for subsequent parameter optimization and overcoming the shortcomings of traditional methods in measuring temporal similarity.

[0099] A random gating-based grouped sparse feature selection algorithm is used to sparsify the parameter vector of the refined mechanism model, and a subset of key parameters affecting the dynamic time warping distance of the derivative is selected. This significantly reduces the redundancy of optimization in the high-dimensional parameter space and improves the efficiency and interpretability of parameter calibration.

[0100] Using the minimization of the derivative dynamic time warping distance as the objective function, and employing the moss optimization algorithm to optimize the subset of key parameters, the optimal parameter combination can be obtained automatically and efficiently, making the dynamic response of the digital twin approximate the physical entity. This solves the problems of traditional methods relying on human experience and having slow convergence speed when tuning parameters.

[0101] The optimal parameter combination is backfilled into the corresponding parameters of the constructed refined mechanism model to complete parameter calibration, and finally a high-fidelity and highly consistent digital twin model of the wind turbine is obtained, providing a reliable digital foundation for subsequent applications such as load dynamic evaluation, operation status optimization and ultra-short-term power generation prediction of wind turbines. Attached Figure Description

[0102] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0103] Figure 1 This is a flowchart of the method for constructing a digital twin of a wind turbine according to the present invention.

[0104] Figure 2 This is a diagram showing the digital twin fusion prediction results of the wind turbine digital twin construction method described in this invention.

[0105] Figure 3 This is a diagram of the aerodynamic-mechanical simulation module architecture of the wind turbine digital twin construction method described in this invention.

[0106] Figure 4 This is a schematic diagram of the interface between the aerodynamic-mechanical simulation module and the fourth-order electromechanical transient model in the wind turbine digital twin construction method of the present invention.

[0107] Figure 5 This is a schematic diagram of the digital twin framework of the wind turbine digital twin construction method described in this invention.

[0108] Figure 6 This is a schematic diagram illustrating the operational optimization of the wind turbine digital twin construction method described in this invention. Detailed Implementation

[0109] To make the purpose, features, and advantages of this application more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solution protected by this application. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0110] This application provides a method for constructing a digital twin of a wind turbine, which solves the current urgent technical problem of insufficient model accuracy in high-dimensional parameter space.

[0111] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0112] Figure 1 A flowchart illustrating a method for constructing a digital twin of a wind turbine generator as provided in an embodiment of this application. Figure 1 As shown in the figure, the method for constructing a digital twin of a wind turbine provided in this application embodiment specifically includes the following steps:

[0113] Step S1: Construct a refined mechanism model, which includes an aero-mechanical simulation model of the wind turbine and a fourth-order electromechanical transient model of the doubly-fed generator. The aero-mechanical simulation model includes higher-order blade element momentum theory, tower shadow effect model and load model.

[0114] Step S2: Obtain historical meteorological data sequences and convert them into physical field parameter sequences. Input the physical field parameter sequences into the refined mechanism model, and the refined mechanism model outputs simulation output sequences.

[0115] Step S3: Obtain the simulation output sequence and the actual operating output sequence of the wind turbine, and use the dynamic time warping algorithm to calculate the derivative dynamic time warping distance between the actual operating output sequence and the simulation output sequence, so as to quantify the consistency of the dynamic response between the digital twin and the physical entity.

[0116] Step S4: Using a random gating-based grouped sparse feature selection algorithm, the parameter vector of the refined mechanism model is sparsified to select a subset of key parameters that affect the dynamic time warping distance of the derivative.

[0117] Step S5: Using the minimization of the derivative dynamic time warping distance as the objective function, the moss optimization algorithm is used to optimize and solve the subset of key parameters to obtain the optimal parameter combination, so that the dynamic response of the digital twin approximates the physical entity.

[0118] Step S6: Backfill the optimal parameter combination into the corresponding parameters of the constructed refined mechanism model, complete the parameter calibration, and obtain the digital twin model of the wind turbine.

[0119] As an example, in step S1, the aerodynamic-mechanical simulation model includes: blade element momentum theory model, tower shadow wake model, and load model;

[0120] Calculation of thrust and torque generated by a single leaf element based on leaf element momentum theory model;

[0121] The local wind speed on which the leaf element momentum theory model depends is corrected based on the tower shadow wake model in order to quantify the tower's blocking effect on the incoming wind field.

[0122] The load model is used to receive the local wind speed after correction by the tower shadow wake model, and to solve the aerodynamic torque of the wind turbine based on the thrust generated by a single blade element micro-element calculated by the blade element momentum theory model, thus completing the transfer from aerodynamic load to structural load.

[0123] It should be noted that in step S11, the thrust and torque generated by a single leaf element are calculated based on the leaf element momentum theory model, and the expression is as follows:

[0124]

[0125]

[0126] in, For thrust; For torque; air density; The local radius of a given leaf element location; It is the axial induction factor; Tangential induction factor; This refers to the wind turbine rotation speed; Wind speed at infinity; This is the Prandtl correction factor.

[0127] In this embodiment, the expression for the Prandtl correction factor is:

[0128]

[0129]

[0130] in, This is the maximum radius of the wind turbine; The relative angle of attack is the angle of inflow. With leaf element twist angle sum; This refers to the number of leaves;

[0131] The expression for the axial induction factor is:

[0132]

[0133] in, It is the critical inducing factor. ; Correct the empirical coefficients for Glauert; The degree of leaf pigmentation; Normal force coefficient;

[0134] The expression for Glauert's corrected empirical coefficient is:

[0135]

[0136] The expression for the normal force coefficient is:

[0137]

[0138] in, The lift coefficient; This is the drag coefficient.

[0139] Blade Element Momentum Theory (BEM) is commonly used to analyze the aerodynamic performance of wind turbine blades, playing a crucial role in wind turbine load analysis. It is used to calculate the aerodynamic forces acting on the blades under different wind speeds and rotational speeds. BEM theory simplifies the wind turbine model into a single-element flow tube, and then divides the flow tube into... Each height is The annular unit is used, with streamlines as the outer boundary. Finally, an iterative process is established to determine the aerodynamics and the induced velocity near the rotor.

[0140] The leaf element theory assumes that the blade can be divided into small leaf elements that are independent of the effects of other surrounding leaf elements. These small leaf elements are then used as aerodynamic components in a two-dimensional airfoil calculated based on local flow conditions for aerodynamic analysis.

[0141] It should be noted that the construction process of the leaf element momentum theory model in this embodiment includes:

[0142] Step S11: Determine the initial leaf element momentum theoretical model, the expression of which is:

[0143]

[0144]

[0145] in, For infinitesimal thrust, For thrust; For infinitesimal torque, For torque; air density; This refers to the number of leaves; The inflow velocity of the blades; This is the chord length of the leaf blade; Normal force coefficient; This is the tangential force coefficient; The local radius of the leaf element; The angle of attack is the relative angle of entry. With leaf element twist angle sum.

[0146] The expression for the inflow velocity of the blades is:

[0147]

[0148] in, for Wind speed of the axis, for Wind speed on the axis.

[0149] The expression for the normal force coefficient is:

[0150]

[0151] in, The lift coefficient; This is the drag coefficient.

[0152] The expression for the tangential force coefficient is:

[0153] .

[0154] Step S12: Considering the influence of wake rotation on the axial and tangential induced velocities, an intermediate process blade element momentum theoretical model is constructed based on the initial blade element momentum theoretical model. Its expression is:

[0155]

[0156]

[0157] In the formula, It is the axial induction factor; Tangential induction factor; This refers to the wind turbine rotation speed; The wind speed at infinity.

[0158] When the blade angle of attack changes rapidly over time, delayed separation occurs in the airfoil flow, leading to significant differences in lift and drag experienced by the actual blade compared to steady-state conditions. Therefore, the Prandtl tip loss correction and Glauber's correction are used in the model to mitigate these effects. The Prandtl correction includes a Prandtl correction factor, expressed as follows:

[0159]

[0160]

[0161] In the formula, This is the Prandtl correction factor. The maximum radius of the wind turbine. The local radius is the location of a certain leaf element.

[0162] Step S13: Correct the momentum equation in the leaf element momentum theory using Prandtl correction factor to obtain the final leaf element momentum theory model, whose expression is:

[0163]

[0164]

[0165] when At that time, the error of the original leaf element momentum theory model will increase with... The critical induction factor increases with the increase of the blade tip speed ratio; in order to cope with the situation when the unit is operating at a high blade tip speed ratio, the Glauert correction calculation method is introduced to set the critical induction factor. :

[0166] when hour,

[0167]

[0168] when hour,

[0169]

[0170] In the formula, the expression for the Glauert corrected empirical coefficient is:

[0171]

[0172] In the formula, This indicates the Glaubert-corrected empirical coefficient.

[0173] According to an embodiment of this application, in step S11, the local wind speed on which the leaf element momentum theory model depends is corrected based on the tower shadow wake model, and its expression is:

[0174]

[0175]

[0176] In the formula, This is the corrected local wind speed for the leaf element unit. This represents the local wind speed deficit caused by the tower shadow effect. The dimensionless radial distance is expressed as follows:

[0177]

[0178] In the formula, for axis, for axis.

[0179] The tower shadow wake model simplifies the physical process of flow around the tower, enabling an effective description of wind speed attenuation in the region behind the tower with relatively low computational cost. Therefore, in real-time simulations, multi-unit wind farm solutions, and conventional load conditions, it can provide sufficient tower shadow correction accuracy at a controllable computational cost.

[0180] In the digital twin system of this invention, the tower shadow model, as an important component of the wind field module, calculates the local wind speed loss and influence range behind the tower by using tower structural parameters and incoming wind speed.

[0181] According to an embodiment of this application, the load model is used to receive the local wind speed corrected by the tower shadow wake model, and to solve for the aerodynamic torque of the wind turbine based on the thrust generated by a single blade element calculated by the blade element momentum theory model. Specifically, this includes:

[0182] Based on the expression for the thrust generated by a single blade element, the increment of the aerodynamic torque of a single blade element is derived. Specifically, based on the thrust generated by a single blade element, the force related to the blade element is extracted, and then the force related to the blade element is decomposed into tangential components according to the pitch angle. Multiplying these components by the lever arm length of the blade element element yields the increment of the aerodynamic torque of a single blade element, expressed as follows:

[0183]

[0184] In the formula, This represents the increment of the aerodynamic torque; The chord length of the leaf element unit; The corrected local wind speed for the leaf element unit; This is the distance from the leaf element unit to the leaf root, i.e., the lever arm length; The length of a leaf element unit; The angle of entry;

[0185] Summing the increments of the aerodynamic torque of all blade elements along the blade span yields the aerodynamic torque acting on the entire rotor, expressed as:

[0186]

[0187] In the formula, For aerodynamic torque;

[0188] A torsional vibration dynamic equation is constructed to simulate the torsional vibration behavior of the transmission chain between the wind turbine and the generator. Its expression is as follows:

[0189]

[0190] In the formula, The moment of inertia of the wind turbine (kg) m²), The moment of inertia of the generator (kg) m²); The angular velocity of the wind turbine (rad / s) The generator's angular velocity (rad / s); The equivalent stiffness of the transmission chain (N) m / rad), For the equivalent damping of the transmission chain (N) m (s / rad); The angle of twist is rad. This refers to the gearbox transmission ratio; pneumatic torque (N) m), electromagnetic torque (N) m), whose expression is:

[0191]

[0192] In the formula, For electromagnetic torque, For magnetizing inductance; For the self-sensing of the stator; For rotor self-inductance; The stator flux linkage cross-axis component; This refers to the direct-axis component of the rotor flux linkage; For the direct-axis component of the stator flux linkage; The cross-axis component of the rotor flux linkage; Let be the leakage flux coefficient, and its expression is:

[0193] .

[0194] The torsional dynamics equation describes the aerodynamic torque How to achieve this through the transmission chain (stiffness) and damping The speed variation of the driving wind turbine and generator ( and (and consider electromagnetic torque) This addresses the dynamic error problem caused by neglecting the flexibility of the transmission chain in traditional models.

[0195] By using torsional vibration dynamics equations, digital twins can simulate torsional vibration effects in transient processes, avoiding mechanical resonance and improving load prediction accuracy.

[0196] The tower is subjected to axial thrust and vibration loads, and its load characteristics are strongly correlated with aerodynamic effects and structural flexibility. The moment equation at the base of the tower is constructed as follows:

[0197]

[0198] In the formula, For the tower base moment, For height wind shear force, It is the first The lower end height coordinates of each beam element. It is the first The upper height coordinates of each beam element This represents the total number of discrete beam elements in the tower.

[0199] Load models are the core support for state simulation in digital twins of wind farms. Their goal is to accurately characterize the mechanical responses of key components such as blades, drive shafts, and towers under multi-field coupling. Wind turbine load models include the aerodynamic torque of the blades and the tower base moment.

[0200] Load transfer follows the path of "aerodynamic load → blade → hub → drive train → doubly-fed generator → tower", and is finally transferred to the foundation. Among them, the aerodynamic load captured by the blade is converted into torque and thrust through the hub, which acts on the top of the tower; the lateral vibration of the tower caused by the torsional vibration of the drive train will react on the hub, forming a "drive train-tower" coupled vibration, which makes the load transfer and coupling exhibit dynamic characteristics.

[0201] The drive shaft is a key component connecting the wind turbine and the generator. Its load is mainly torsional vibration. In OpenFAST (an open-source wind turbine simulation tool), the calculation of the drive shaft is implemented by the structural dynamics module (ElastoDyn). Based on the structural parameters of the drive shaft, this module first defines its torsional stiffness, bending stiffness and moment of inertia. Then, combined with the aerodynamic load of the blades and the inertial load of the transmission chain transmitted by the aerodynamic module (AeroDyn), it outputs key load data such as the torque and bending moment of the main shaft in real time.

[0202] First, the aerodynamic module (AeroDyn) of OpenFAST (an open-source wind turbine simulation tool) calculates the aerodynamic loads and transmits them to the blades in the structural dynamics module (ElastoDyn), driving the blade structure to vibrate. The blade vibration then changes its angle of attack. At this point, the aerodynamic module (AeroDyn) and the structural dynamics module (ElastoDyn) achieve bidirectional coupling between the aerodynamics and the blades through real-time data interaction.

[0203] Secondly, after the torque and thrust of the blades are transmitted to the hub of the ElastoDyn structural power module, they output torque to the ElastoDyn drive train on the one hand, and output thrust to the ElastoDyn tower on the other hand, triggering the dynamic response of the tower beam unit. At the same time, the tower vibration will change the hub attitude, and the torsional vibration of the drive train will excite the lateral vibration of the tower.

[0204] Finally, the dynamic load of the drive train is transmitted to the servo control module (ServoDyn). This module outputs the generator electromagnetic torque command according to the control strategy and acts in reverse on the drive train, forming a control dynamics closed loop of "drive train-generator".

[0205] Combination Figure 3 Using data from the InflowWind module of OpenFAST (a simplified electrical model), the AeroDyn module calculates aerodynamic loads and transmits them to the blades in the ElastoDyn structural dynamics module, driving blade structural vibration. This blade vibration, in turn, changes its angle of attack. Real-time data interaction between AeroDyn and ElastoDyn enables bidirectional aerodynamic-blade coupling. Next, the blade torque and thrust are transmitted to the ElastoDyn hub, outputting torque to the ElastoDyn drivetrain and thrust to the ElastoDyn tower, triggering the tower beam unit's dynamic response. Simultaneously, tower vibration alters the hub attitude, and torsional vibration of the drivetrain excites lateral tower vibration. Finally, the dynamic load of the drivetrain is transmitted to the ServoDyn servo control module. This module, based on its control strategy, outputs generator electromagnetic torque commands that act in reverse on the drivetrain, forming a closed-loop control dynamics system between the drivetrain and the generator.

[0206] According to another embodiment of the present invention, on the electrical side, in order to accurately capture the dynamic response under power grid disturbances, the present invention directly uses the stator and rotor flux linkages as state variables to construct a fourth-order electromechanical transient model:

[0207]

[0208] In the formula, The fundamental angular frequency; The state matrix; It is a voltage vector; It is a resistance matrix; For inductance matrix; To include synchronous angular velocity With slip angular velocity The rotation transformation matrix;

[0209] The expression for the state matrix is:

[0210]

[0211] The expression for the voltage vector is:

[0212]

[0213] The expression for the stator and rotor flux linkage is:

[0214]

[0215] In the formula, This is the stator flux linkage vector; is the rotor flux linkage vector.

[0216] To achieve real-time digital twin solution of the model, the trapezoidal integral method with A-stable properties is used to discretize the above continuous model. The state variables of the stator and rotor flux linkages in the step Represented as:

[0217]

[0218] This discretization strategy supports a large simulation step size of 1-5ms while ensuring the numerical accuracy of the stator-rotor coupling and the dq-axis cross-coupling relationship, thus meeting the requirements of real-time simulation.

[0219] To achieve coupled simulation of the entire "wind-turbine-electric" process, a fourth-order electromechanical transient model is integrated with the aerodynamic-mechanical simulation module (RT-OpenFAST). This invention utilizes the Bladed dynamic link library (DLL) interface provided by the aerodynamic-mechanical simulation module (RT-OpenFAST) to encapsulate the fourth-order electromechanical transient model of the doubly-fed induction generator (DFIG) and its controller into an independent module. For example... Figure 4 As shown, through this interface, the pneumatic-mechanical simulation module (RT-OpenFAST) and the fourth-order electromechanical transient model exchange high-frequency data within each simulation step, forming a closed-loop coupled simulation.

[0220] The data exchanged between the two parties revolves around two core aspects: mechanical-electrical energy transfer and control command closed loop. Within each coupling step, the aero-mechanical simulation module (RT-OpenFAST) transmits the wind turbine angular velocity to the fourth-order electromechanical transient model. The fourth-order electromechanical transient model feeds back the electromagnetic torque to the aero-mechanical simulation module (RT-OpenFAST). This forms a closed loop.

[0221] Timing synchronization and data consistency are crucial for coupled simulations. Due to the different time constants of the dynamic characteristics of aero-mechanical and electrical systems, the aero-mechanical simulation module (RT-OpenFAST) and the fourth-order electromechanical transient model often employ different simulation step sizes. To ensure strict physical synchronization and avoid information loss, this invention designs a master-slave hierarchical step size management and data interpolation mechanism for real-time synchronization.

[0222] (1) Master-slave step size management: The slower mechanical side step size is used as the master step size, and bidirectional data exchange is performed at the starting point of each master step size.

[0223] (2) High-precision interpolation: For variables obtained from the mechanical side, linear interpolation or quadratic hold method is used within the sub-step of the electrical side to generate continuous inputs to ensure the accuracy of electrical model calculation; conversely, when the high-frequency torque command output by the electrical side is transmitted to the mechanical side, the average value within the step is taken to match its dynamic response frequency.

[0224] (3) Real-time synchronization: By using shared memory and real-time semaphores, the computation cycles of both parties are strictly aligned to ensure the determinism of data exchange and eliminate the cumulative error caused by asynchronous simulation.

[0225] Integrating the fourth-order electromechanical transient model with the aero-mechanical simulation module (RT-OpenFAST) effectively solves the timing challenges in multi-rate coupled simulation, providing a high-fidelity and real-time simulation foundation for wind turbine controller hardware-in-the-loop testing, fault ride-through research, and full-field electromechanical coupling dynamic analysis.

[0226] To achieve accurate mapping of digital virtual entities to physical entities within this system, this invention designs, as follows: Figure 5 The digital twin framework shown enables real-time data-driven model updates. This framework includes an object layer—the wind turbine, a virtual fusion layer, and a management layer.

[0227] The physical wind turbine side deploys dedicated equipment such as blade fiber Bragg grating sensors, grid-connected PMUs, and wind-measuring radars. The sensing and measurement system collects load, electrical quantity, and micro-wind speed data. After preprocessing by the edge gateway, a wind turbine twin model is obtained and a verification benchmark is provided. The virtual fusion layer, based on a fully coupled "air-bearing-mechanical-electrical" mechanism model, combines Derivative Dynamic Time Warping (DTW) and Moss Optimization (MGO) algorithms for intelligent optimization, achieving online dynamic calibration of structural and control parameters within engineering parameter constraints. The management layer can perform fault perception and twin data analysis based on the physical and virtual fusion layers, generating corresponding operation and maintenance strategies. Ultimately, this forms a virtual-physical fusion architecture adapted to the strong randomness and multi-field coupling characteristics of wind power, providing high-precision, interpretable, unified model support for core wind power scenarios such as ultra-short-term power prediction.

[0228] In an exemplary embodiment, to construct a fitness evaluation system suitable for metaheuristic algorithms, an objective function needs to be designed to quantify the similarity between the operating parameters of the physical wind turbine and the twin wind turbine. This objective function provides a clear optimization direction for the metaheuristic algorithm by comparing the output sequences of the physical wind turbine and the twin wind turbine, driving the control parameters of the twin wind turbine to approximate those of the physical wind turbine.

[0229] This invention employs a derivative dynamic time warping algorithm. During the registration process, it does not directly compare the original amplitudes of the samples, but rather compares the first derivatives of the sequences, thus focusing more on the consistency of signal shape rather than absolute value differences.

[0230] In step S3, the actual operating output sequence and the simulation output sequence of the wind turbine are obtained, and the derivative dynamic time warping distance between the actual operating output sequence and the simulation output sequence is calculated using a dynamic time warping algorithm. Specifically, this includes:

[0231] Let the actual output sequence be The simulation output sequence is ;

[0232] For the actual output sequence is and simulation output sequence are Each of them calculates its derivative approximation, and its expression is:

[0233]

[0234] In the formula, Indicates the actual output sequence In the Approximate values ​​of the first derivative at each time point. Represents the simulation output sequence In the Approximate values ​​of the first derivative at each time point;

[0235] Define the actual output sequence in the derivative space. The Points and simulation output sequence The The local distance between points is expressed as:

[0236]

[0237] In the formula, Indicates the actual output sequence The Points and simulation output sequence The Local distance between points;

[0238] Using dynamic programming recursion with dynamic time warping, we can derive the sequence starting from the beginning. The minimum cumulative distance to the current point is expressed as:

[0239]

[0240] In the formula, Indicates starting from the beginning of the sequence Minimum cumulative distance to the current point;

[0241] By path length For minimum cumulative distance After normalization, the normalized derivative dynamic time warped distance is obtained, and its expression is:

[0242]

[0243] In the formula, For the actual output sequence With simulation output sequence The dynamic time-normalized distance between the derivatives.

[0244] The derivative dynamic time warping algorithm reflects the actual running output sequence. With simulation output sequence The shape similarity makes digital twins less susceptible to environmental interference, thus enabling higher precision simulation.

[0245] In step S4, a random-gated grouped sparse feature selection algorithm is used to sparsify the structure and control parameter vectors of the wind turbine model, and to select a subset of key parameters that affect the dynamic time warping distance of the derivative, specifically including:

[0246] Suppose that the fan structure and control parameters, after normalization, form an N-dimensional parameter vector:

[0247]

[0248] In the formula, Represents the original parameter vector; Indicates parameters; Represents the set of real numbers; Indicates the total number of original parameters;

[0249] Before the input layer of the deep neural network, a learnable random gating vector is set for each dimension parameter and its physical grouping (pneumatic, mechanical, electrical, etc.):

[0250]

[0251] In the formula, Represents a random gate vector; Indicates the gate value;

[0252] Obtain effective inputs from actual participants in the modeling process:

[0253]

[0254] In the formula, This represents element-wise multiplication. This represents a subset of key parameters, which is the effective parameter vector that is ultimately input into the optimization algorithm.

[0255] By applying an approximate sparse regularization to the gating coefficients, the random-gated grouped sparse feature selection algorithm automatically suppresses redundant features while fitting the nonlinear mapping relationship between parameters and responses, thus achieving embedded feature selection. Specifically, with As network input, a virtual-real difference index based on DDTW is used as a supervision signal. During training, the network learns the nonlinear mapping between wind turbine parameter perturbations and output response, and also learns through gating vectors. Sparsification is achieved by retaining only a small subset of parameters that significantly contribute to the error and are frequently selected during multiple training iterations. The randomized gated grouped sparse feature selection algorithm can significantly compress the parameter search space while maintaining model prediction accuracy, thereby improving the real-time calibration efficiency and interpretability of digital twin models under stochastic wind power conditions.

[0256] In step S5, with minimizing the dynamic time warping distance of the derivative as the objective function, the moss optimization algorithm is used to optimize and solve the subset of key parameters to obtain the optimal parameter combination, specifically including:

[0257] Obtain a subset of key parameters, denoted by dimension . , ;

[0258] The search space is determined based on the engineering constraints of each parameter in the key parameter subset, and the feasible region of the parameter is defined. ;

[0259] Initialize the control parameters of the moss optimization algorithm, including population size. Maximum number of evaluations Current number of assessments ;

[0260] In the feasible region Internal random generation The initial solution for the combination of parameters is denoted as... ;

[0261] For each initial solution The RT-OpenFAST simulation framework is invoked to run the simulation and calculate the fitness based on the derivative dynamic time warping algorithm. ,in, To use the initial solution The simulation output, This refers to measured data of physical entities; The smaller the value, the better the initial solution;

[0262] Record the current optimal solution and the current optimal solution fitness ;

[0263] The number of assessments has been updated to ;

[0264] When the number of evaluations Less than the maximum number of evaluations hour:

[0265] Enter the iteration loop and process each initial solution in the population. Update the position and generate a new initial solution. ; Calculate the new initial solution fitness ;

[0266] If the new initial solution fitness Less than the current optimal solution fitness Then update the optimal solution. , ;

[0267] like and Then use replace ;

[0268] like or Then keep constant;

[0269] If retained Keeping the current solution unchanged, calculate the current optimal solution. With the initial solution The direction vector is used to explore and generate new individuals. ;

[0270] Generate random numbers Rand. If Rand ≤ 0.8, then use historical best solutions. For the initial solution Perform a small perturbation to generate new individuals. And assess its fitness; if it is better, update the individual;

[0271] If the current number of evaluations Reaching the maximum number of evaluations Or the fitness of the optimal solution improves by less than a preset threshold over several consecutive generations. hour:

[0272] The algorithm terminates and returns the current optimal solution. This optimal parameter combination minimizes the DDTW distance between the digital twin simulation output and the measured data of the physical entity, thus achieving the highest consistency between the virtual and real data.

[0273] The optimal parameter combination is then backfilled into the corresponding parameters of the constructed refined mechanism model to complete the parameter calibration of the digital twin, thereby obtaining a high-fidelity, highly consistent personalized digital twin model for subsequent state extrapolation and prediction.

[0274] For example, after obtaining the digital twin model of the wind turbine, the sequence of physical field parameters is input into the digital twin model of the wind turbine, and the digital twin model of the wind turbine outputs the future state sequence of the wind turbine to realize ultra-short-term power generation prediction.

[0275] In this embodiment, to achieve dynamic load assessment, a damage equivalent load assessment method based on rainflow counting is adopted, in the constructed... Figure 6 On the digital twin platform shown, the non-stationary torque time series is decomposed into several closed load cycles through a cyclic statistical process. Based on Miller's linear damage accumulation theorem, the damage caused by each cycle is superimposed to calculate the equivalent damage-equivalent load value.

[0276] In rainflow-count-based damage equivalent load assessment, a fatigue torque cycle refers to the complete fluctuation of fatigue torque between two local extrema. A cycle typically consists of a peak and a trough. The characteristic of each cycle is determined by its amplitude, i.e., the peak-to-trough difference. Within a given time interval, the fatigue torque signal usually contains multiple fatigue torque cycles with varying amplitudes.

[0277] To obtain continuous fatigue torque signals These cycles are extracted, and the time-domain signal of fatigue torque is decomposed based on the rainflow counting method. Unlike the simple peak-valley counting method that only considers the pairing of adjacent extrema, the rainflow counting method also considers the pairing of non-adjacent extrema that can form a closed fluctuation path. For example, if there are several smaller peaks and valleys embedded between a large peak and a large valley, the rainflow counting method will prioritize pairing the large peaks and valleys and process the intermediate extrema separately to form additional small cycles.

[0278] Through this process, the fatigue torque signal Decomposed into fatigue torque cycle ,in Indicates the first The amplitude of each fatigue torque cycle, This indicates the cumulative number of iterations in the loop.

[0279] Based on the fatigue characteristic parameters of the material, the first Fatigue damage corresponding to each cycle It can be represented as:

[0280]

[0281] In the formula, For the material at fatigue torque amplitude The fatigue life under the following conditions, that is, the fatigue life experienced After the second cycle, material failure phenomena such as fracture occur; The fatigue index is the material's coefficient.

[0282] Total fatigue damage when considering all cycles Based on the Palmgren–Miner linear damage accumulation criterion, it can be expressed as:

[0283]

[0284] Damage equivalent load Defined as the total number of loops The same fatigue damage occurs. The constant amplitude load. The relationship between the two can be expressed as:

[0285]

[0286] Therefore, the damage equivalent load The calculation formula is:

[0287] (1)

[0288] The damage equivalent load model constructed according to equation (1) can calculate fatigue damage under different working conditions, thereby realizing the dynamic evaluation of load.

[0289] Based on a digital twin, a load optimization framework is constructed. Using the damage equivalent load model constructed according to equation (1), dynamic adjustments to the operating state are achieved through an active power optimization allocation strategy. Taking into full account individual differences in the units and power constraints, a two-layer architecture of "edge sensitivity estimation-centralized optimization" and a GPU-accelerated algorithm are adopted, which can significantly reduce fatigue damage to key components and extend the service life of the units. The specific process is as follows: Figure 6 As shown.

[0290] The fatigue torque model of the nonlinear wind turbine is linearized based on Taylor expansion to obtain the fatigue torque. The formula for calculating linear sensitivity is as follows:

[0291] (2)

[0292] In the formula, Indicates power reference increment For fatigue torque increment Sensitivity factor; Indicates the current working point The initial value.

[0293] For any unit in the framework, local state variables are first measured, and then the corresponding sensitivity factor is calculated based on equation (2). In addition, the required available power needs to be estimated locally at the edge. Finally, these calculation results are transmitted to the central control center for centralized optimization.

[0294] Figure 2 The figure shows the prediction results of the wind turbine digital twin model of the present invention using data from a Huaneng wind farm. As can be seen from the figure, the wind turbine digital twin model of the present invention can effectively reflect the fluctuations and trends in wind power generation.

[0295] To evaluate the performance of the wind turbine digital twin model, mainstream ultra-short-term prediction models such as Long Short-Term Memory (LSTM) and Recurrent Neural Network (RNN) were selected and compared with the wind turbine digital twin model proposed in this invention. Three commonly used indicators, namely root mean squared error (RMSE), mean absolute error (MAE), and coefficient of determination (R²), were used for comprehensive performance analysis.

[0296] The three methods were used to predict ultra-short-term power generation using data from a Huaneng wind farm. The performance of each method is analyzed, as shown in Table 1. As can be seen from the table, the prediction performance of the wind turbine digital twin model proposed in this invention is superior to other commonly used prediction models.

[0297] Table 1 Comparison of Model Prediction Evaluation Indicators

[0298]

[0299] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0300] Any changes, modifications, substitutions, and variations made to the embodiments without departing from the principles and spirit of the present invention still fall within the protection scope of the present invention.

Claims

1. A method for constructing a digital twin of a wind turbine generator, characterized in that, The method includes: Step S1: Construct a refined mechanism model, which includes an aerodynamic-mechanical simulation model of the wind turbine and a fourth-order electromechanical transient model of the doubly-fed generator. The aerodynamic-mechanical simulation model includes higher-order blade element momentum theory, tower shadow effect model and load model. Step S2: Obtain historical meteorological data sequences and convert them into physical field parameter sequences. Input the physical field parameter sequences into the refined mechanism model, and the refined mechanism model outputs simulation output sequences. Step S3: Obtain the simulation output sequence and the actual operating output sequence of the wind turbine, and use the first derivative dynamic time warping algorithm of the comparison sequence to calculate the derivative dynamic time warping distance between the actual operating output sequence and the simulation output sequence; Step S4: Using a random gating-based grouped sparse feature selection algorithm, the parameter vector of the refined mechanism model is sparsified to filter out the subset of key parameters that affect the dynamic time warping distance of the derivative. Step S5: Using the minimization of the derivative dynamic time warping distance as the objective function, the moss optimization algorithm is used to optimize and solve the subset of key parameters to obtain the optimal parameter combination; Step S6: Backfill the optimal parameter combination into the corresponding parameters of the constructed refined mechanism model, complete the parameter calibration, and obtain the digital twin model of the wind turbine. In step S4, a random gating-based grouped sparse feature selection algorithm is used to sparsify the structure and control parameter vectors of the wind turbine model, and to filter out a subset of key parameters that affect the dynamic time warping distance of the derivative, specifically including: Suppose that the fan structure and control parameters, after normalization, form an N-dimensional parameter vector: In the formula, Represents the original parameter vector; Indicates parameters; Represents the set of real numbers; Indicates the total number of original parameters; Before the input layer of the deep neural network, set a learnable random gating vector for each dimension parameter and its physical grouping: In the formula, Represents a random gate vector; Indicates the gate value; Obtain effective inputs from actual participants in the modeling process: In the formula, This represents element-wise multiplication. This represents a subset of key parameters.

2. The method according to claim 1, characterized in that, In step S1, the aerodynamic-mechanical simulation model includes: blade element momentum theory model, tower shadow wake model, and load model; Calculation of thrust and torque generated by a single leaf element based on leaf element momentum theory model; The local wind speed on which the leaf element momentum theory model depends is corrected based on the tower shadow wake model; The load model is used to receive the local wind speed after correction by the tower shadow wake model, and to solve the aerodynamic torque of the wind turbine based on the thrust generated by a single blade element micro-element calculated by the blade element momentum theory model.

3. The method according to claim 2, characterized in that, The thrust and torque generated by a single leaf element are calculated based on the leaf element momentum theory model, and their expressions are as follows: in, For thrust; For torque; air density; The local radius of a given leaf element location; It is the axial induction factor; Tangential induction factor; This refers to the wind turbine rotation speed; Wind speed at infinity; This is the Prandtl correction factor.

4. The method according to claim 3, characterized in that, The expression for Prandtl's correction factor is: in, This is the maximum radius of the wind turbine; The relative angle of attack is the angle of inflow. With leaf element twist angle sum; This refers to the number of leaves; The expression for the axial induction factor is: in, It is the critical inducing factor. ; Correct the empirical coefficients for Glauert; The degree of leaf pigmentation; Normal force coefficient; The expression for Glauert's corrected empirical coefficient is: The expression for the normal force coefficient is: in, The lift coefficient; This is the drag coefficient.

5. The method according to claim 4, characterized in that, The local wind speed upon which the leaf element momentum theory model is based, modified by the tower shadow wake model, is expressed as follows: In the formula, This is the corrected local wind speed for the leaf element unit. This represents the local wind speed deficit caused by the tower shadow effect. The dimensionless radial distance is expressed as follows: In the formula, for axis, for axis.

6. The method according to claim 5, characterized in that, The load model is used to receive the local wind speed corrected by the tower shadow wake model, and to solve for the aerodynamic torque of the wind turbine based on the thrust generated by a single blade element calculated by the blade element momentum theory model. Specifically, it includes: Based on the expression for the thrust generated by a single blade element, the increment of the aerodynamic torque of a single blade element is derived, and its expression is as follows: In the formula, This represents the increment of the aerodynamic torque; The chord length of the leaf element unit; The corrected local wind speed for the leaf element unit; This is the distance from the leaf element unit to the leaf root, i.e., the lever arm length; The length of a leaf element unit; The angle of entry; Summing the increments of the aerodynamic torque of all blade elements along the blade span yields the aerodynamic torque acting on the entire rotor, expressed as: In the formula, This refers to the pneumatic torque.

7. The method according to claim 1, characterized in that, A fourth-order electromechanical transient model is constructed using stator and rotor flux linkages as state variables: In the formula, The fundamental angular frequency; The state matrix; It is a voltage vector; It is a resistance matrix; For inductance matrix; To include synchronous angular velocity With slip angular velocity The rotation transformation matrix; The expression for the state matrix is: The expression for the voltage vector is: The expression for the stator and rotor flux linkage is: In the formula, This is the stator flux linkage vector; is the rotor flux linkage vector.

8. The method according to claim 1, characterized in that, In step S3, the actual operating output sequence and the simulation output sequence of the wind turbine are obtained, and the derivative dynamic time warping distance between the actual operating output sequence and the simulation output sequence is calculated using the first derivative dynamic time warping algorithm of the comparison sequence. Specifically, this includes: Let the actual output sequence be The simulation output sequence is ; For the actual output sequence is and simulation output sequence are Each of them calculates its derivative approximation, and its expression is: In the formula, Indicates the actual output sequence In the Approximate values ​​of the first derivative at each time point. Represents the simulation output sequence In the Approximate values ​​of the first derivative at each time point; Define the actual output sequence in the derivative space. The Points and simulation output sequence The The local distance between points is expressed as: In the formula, Indicates the actual output sequence The Points and simulation output sequence The Local distance between points; Using dynamic programming recursion with dynamic time warping, we can derive the sequence starting from the beginning. The minimum cumulative distance to the current point is expressed as: In the formula, Indicates starting from the beginning of the sequence Minimum cumulative distance to the current point; By path length For minimum cumulative distance After normalization, the normalized derivative dynamic time warped distance is obtained, and its expression is: In the formula, For the actual output sequence With simulation output sequence The dynamic time-normalized distance between the derivatives.

9. The method according to any one of claims 1 to 8, characterized in that, In step S5, with minimizing the dynamic time warping distance of the derivative as the objective function, the moss optimization algorithm is used to optimize the subset of key parameters to obtain the optimal parameter combination, specifically including: Obtain a subset of key parameters, whose dimension is denoted as . , ;in, This represents the total number of original parameters; The search space is determined based on the engineering constraints of each parameter in the key parameter subset, and the feasible region of the parameter is defined. ; Initialize the control parameters of the moss optimization algorithm, including population size. Maximum number of evaluations Current number of assessments ; In the feasible region Internal random generation The initial solution for the combination of parameters is denoted as... ; For each initial solution The RT-OpenFAST simulation framework is invoked to run the simulation and calculate the fitness based on the derivative dynamic time warping algorithm. ,in, To use the initial solution The simulation output, This refers to measured data of physical entities; Record the current optimal solution and the current optimal solution fitness ; The number of assessments has been updated to ; When the number of evaluations Less than the maximum number of evaluations hour: For each initial solution in the population Update the position and generate a new initial solution. ; Calculate the new initial solution fitness ; If the new initial solution fitness Less than the current optimal solution fitness Then update the optimal solution. , ; like and Then use replace ; like or Then keep constant; If retained Keeping the current solution unchanged, calculate the current optimal solution. With the initial solution The direction vector is used to explore and generate new individuals. ; Generate random numbers Rand. If Rand ≤ 0.8, then use historical best solutions. For the initial solution Perform a small perturbation to generate new individuals. And assess its fitness; if it is better, update the individual; If the current number of evaluations Reaching the maximum number of evaluations Or the fitness of the optimal solution improves by less than a preset threshold over several consecutive generations. hour: Return the current optimal solution As the optimal combination of parameters.

Images