Fan twin parameter online updating method and system based on neural deformation and moss optimization algorithm

Abstract

This invention belongs to the field of digital twin modeling and intelligent optimization technology for new energy power generation systems, and relates to a method and system for online updating of wind turbine twin parameters based on neural deformability and moss optimization algorithms. The method proposes an online updating method for digital twin model parameters coupled with neural deformability networks and moss optimization algorithms. By constructing a learnable deep elastic time series similarity metric, it solves the evaluation problem of multivariable temporal phase drift and local deformation. Combined with the efficient global search characteristics of the moss optimization algorithm, it achieves high-precision, adaptive, and online updating of key parameters of the wind turbine digital twin model, ultimately improving the consistency of dynamic behavior between the virtual and real systems, and laying a model foundation for the engineering application of digital twins in the wind power field.

Description

Technical Field

[0001] This invention belongs to the field of digital twin modeling and intelligent optimization technology of new energy power generation systems, specifically involving a method and system for online updating of wind turbine twin parameters based on neural deformation and moss optimization algorithms. Background Technology

[0002] The digital twin model of wind turbine is the core carrier for condition monitoring and fault early warning. However, in actual operation, due to the coupling effect of multiple factors such as wind speed fluctuations, component aging, electrical parameter temperature drift and operating condition switching, the equivalent electrical parameters and mechanical parameters of the unit continue to drift, resulting in "twin deviation" between the digital twin model and the physical wind turbine. This leads to the degradation of the consistency between the virtual and real dynamics, which reduces the model's prediction accuracy and engineering reliability.

[0003] Among existing methods for updating parameters in digital twins of wind turbines, estimation methods based on principles such as least squares and Kalman filtering have poor adaptability to nonlinear strongly coupled systems and rely on accurate mathematical models. Traditional data-driven metaheuristic algorithms are limited by the design of the objective function and are insufficient in handling phase drift, local disturbances, and noise superposition problems in multivariate time series data. Furthermore, they lack efficiency in searching parameters and global optimization capabilities in high-dimensional parameter spaces. Conventional time series similarity measurement methods are difficult to characterize the dynamic behavior differences between virtual and real system outputs, which can easily lead to distortion in similarity evaluation and cause the parameter optimization results to deviate from the optimal solution.

[0004] In view of this, it is very necessary to provide a method and system for online updating of wind turbine twin parameters based on neural deformation and moss optimization algorithms to solve the above-mentioned defects in the prior art. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing wind turbine digital twin parameter update technologies, such as reliance on precise mathematical models, limitations in similarity measurement methods that prevent accurate characterization of the dynamic behavior differences of wind turbines over time, and insufficient search efficiency and global optimization capabilities of traditional metaheuristic algorithms in high-dimensional parameter spaces. This invention provides a method and system for online updating wind turbine twin parameters based on neural deformation and moss optimization algorithms to solve the aforementioned technical problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for online updating of wind turbine twin parameters based on neural morphology and moss optimization algorithms includes the following steps: Step S1: Based on the structural topology, air-elastic-mechanical-electric coupling mechanism and control logic of the physical wind turbine, construct a digital twin model of the wind turbine unit; determine the key parameter vector to be optimized in the digital twin model, input the key parameter vector to be optimized into the digital twin model, and obtain the simulation output time series; Step S2: Collect multivariate time-series operating data of the physical fan during actual operation; perform engineering preprocessing on the collected raw multivariate time-series operating data to obtain the output time series of the physical fan; Step S3: Construct a neural morphing network similarity calculation model based on "encoder-morphing module-aggregation module"; output continuous similarity scores based on the neural morphing network similarity calculation model and the simulation output time series. Step S4: Using the similarity score output by the neural morphology network as the optimization objective, construct the fitness function of the moss optimization algorithm, transforming the parameter update problem of the digital twin model into a fitness function minimization problem; using the global search characteristics of the moss optimization algorithm, search for the optimal parameter vector within the parameter feasible region to achieve high-precision parameter optimization. Step S5: Inverse normalize the optimal parameter vector; send it online to the digital twin model of the wind turbine, replace the key parameter vector to be optimized, and complete the parameter calibration and update of the digital twin model through the engineering optimization process of "parameter sending-quiet period-parameter stability confirmation-sampling window-similarity evaluation", so that the output of the digital twin model maintains a high degree of dynamic consistency with the operating data of the physical wind turbine. Step S6 involves continuously collecting multivariate time-series operating data of the physical wind turbine and repeating steps S2-S5 at fixed time windows to dynamically update the parameters of the digital twin model, suppress the "twin deviation" phenomenon in real time, and ensure the dynamic consistency and engineering credibility of the digital twin model throughout the entire life cycle of the wind turbine.

[0007] Preferably, step S1 specifically includes: Step S11: Based on the structural topology, aero-elastic-mechanical-electric coupling mechanism and control logic of the physical wind turbine, construct a high-fidelity digital twin model of the wind turbine. The digital twin model includes: wind turbine aerodynamic model, synchronous generator model, grid-connected converter model, transmission chain mechanical model, etc., to fully reproduce the dynamic operating characteristics of the physical wind turbine. Step S12, determine the key parameter vector θ to be optimized in the digital twin model; the key parameter vector θ to be optimized consists of the core electrical and mechanical parameters that affect the dynamic behavior of the wind turbine. , in, For stator leakage, For rotor leakage inductance, For stator resistance, For rotor resistance, It is the inertial constant. For magnetizing inductance, For the resistance of branch RL, For the RL branch inductance, For parallel capacitors; the initial values ​​of each parameter are based on the factory calibration values ​​of the physical fan or the benchmark measured values, and the search range is limited to a reasonable range that is physically permissible; Step S13: Input the vector of key parameters to be optimized into the digital twin model to obtain the simulation output time series of the digital twin model. .

[0008] This step can achieve the following technical effects: By constructing a high-fidelity digital twin model of the wind turbine and clearly defining the core electrical and mechanical parameters to be optimized, a foundation was laid for subsequent parameter identification. By inputting the parameter vectors to be optimized into the digital twin model to obtain the simulation output time series, a complete correspondence between the physical wind turbine and the digital twin model at the input and output levels was achieved, ensuring that the two are comparable and providing unified data and support for subsequent parameter optimization based on deep elastic similarity.

[0009] Preferably, step S2 specifically includes: Step S21: Collect multivariate time-series operating data of the physical wind turbine during actual operation. The multivariate time-series operating data are core electromechanical coupling variables that correspond one-to-one with the output of the digital twin model, including: generator stator active power, stator current RMS value, stator voltage RMS value, DC bus voltage, electromagnetic torque, rotor speed, reactive power, etc., which constitute the output time series of the physical wind turbine. Step S22 involves performing engineering preprocessing on the raw multivariate time-series operating data in the physical wind turbine output time series. This includes outlier removal, missing value completion, time alignment, and data normalization to eliminate measurement noise and interference from the wind turbine's transient processes, ensuring the validity and consistency of the data, and obtaining the preprocessed physical wind turbine output time series. Where T is the time series length and C is the number of variable channels.

[0010] This step can achieve the following technical effects: By collecting core electromechanical coupling variables, a complete operating state characterization of the physical wind turbine was constructed. Through engineering preprocessing operations, the interference of measurement noise and transient processes of the wind turbine was eliminated, ensuring the validity and consistency of the benchmark reference data and improving the accuracy of subsequent similarity evaluation and parameter optimization.

[0011] Preferably, step S3 specifically includes: Step S31: Construct a neural morphing network similarity calculation model. The neural morphing network similarity calculation model includes: an encoder, a morphing module, and an aggregation module. The encoder is used for multi-scale feature extraction, the morphing module is used for soft alignment weight learning, and the aggregation module is used for global similarity scoring. Step S32: Output time series of physical fan The simulation output time series obtained in step S1 Input encoder Perform multi-scale feature extraction: extract the original time series data and Mapping to a high-dimensional feature space yields a sequence of physical features. With simulation feature sequence This enables dimensionality reduction and feature enhancement of data; the encoder is constructed using CNN (Deep Convolutional Network) or LSTM (Long Short-Term Memory Network) and has the ability to extract local and global features from time-series data; Step S33: Learn the soft alignment weight matrix in the time dimension through the deformation module. The soft alignment weight matrix is ​​used to characterize the nonlinear correspondence between the physical feature sequence and the simulation feature sequence in the time dimension: , in, This is a similarity metric function that can calculate the physical feature sequence at time i. Simulated feature sequence at time j Local similarity; It is a normalization function along the column dimension, which makes the weight matrix satisfy the probability distribution characteristics, thereby achieving the focus alignment of key time series segments and the weakening of non-key segments; The element in the i-th row and j-th column of the soft alignment weight matrix represents the i-th time step of the physical feature sequence. With the simulated feature sequence at time j Alignment weights between them; Step S34, in the aggregation module, adjust the soft alignment weight matrix. The global continuous similarity score is obtained by weighted summation with local similarity and then mapped using the sigmoid activation function. This enables a comprehensive evaluation of the dynamic trends and amplitude consistency of the virtual and real systems. , in, For the Sigmoid activation function, For the local similarity function in the feature space, This serves as a normalization factor to ensure consistency in the scoring scale.

[0012] This step can achieve the following technical effects: This step constructs a three-stage neural deformable network similarity calculation model consisting of an encoder, a deformable module, and an aggregation module. Multi-scale feature extraction and dimensionality reduction enhancement are achieved through the encoder using either a CNN or LSTM. The deformable module learns a time-dimensional soft-alignment weight matrix, avoiding the forced point-by-point alignment defects of DTW (Dynamic Time Warping) and enabling adaptive handling of phase drift and local deformation. The aggregation module outputs a continuous similarity score after Sigmoid activation. This model can simultaneously characterize the dual consistency of dynamic trends and amplitudes in both virtual and real systems, improving the accuracy and robustness of similarity evaluation.

[0013] Preferably, step S4 specifically includes: Step S41, define the fitness function of the moss optimization algorithm as follows: , in, The fitness value is represented by the complement of the similarity score. Therefore, the smaller the fitness value, the higher the degree of consistency between the dynamic behavior of the physical wind turbine and the digital twin model. This represents the continuous similarity score output by the neural morphology network similarity calculation model. Step S42 involves designing four core mechanisms: wind direction guidance, spore dispersal, dual reproduction, and cryptobiosis. A moss optimization algorithm is then used to dynamically adjust global exploration and local exploitation in a high-dimensional parameter space to obtain the optimal parameter vector. .

[0014] This step can achieve the following technical effects: By using the similarity score output by the neural morphology network as the optimization objective, a moss optimization algorithm with the similarity score complement as the fitness function is constructed, transforming the parameter update problem into a fitness function minimization problem. By designing four core mechanisms—wind guidance, spore dispersal, dual reproduction, and cryptobiosis—a dynamic balance between global exploration and local exploitation in a high-dimensional parameter space is achieved. Compared with traditional metaheuristic algorithms, this algorithm has the advantages of faster convergence speed, higher search accuracy, and stronger resistance to local optima.

[0015] Preferably, the specific search process of the moss optimization algorithm in step S42 is as follows: Step S421, Population Initialization: Assign each individual algorithmic unit of the moss optimization algorithm to a parameter vector to be optimized. An initial population is randomly generated within the physical feasible region of each parameter, and each parameter is normalized to a uniform interval to eliminate the interference of dimensional differences on the search process. Step S422: Select each parameter vector to be optimized in the population as a candidate parameter vector. Substitute the digital twin model into the simulation to obtain the simulation output time series. Similarity scores are calculated using a neural morphology network similarity calculation model. And obtain the fitness value according to the fitness function. ; Step S423, Wind-guided evolution: Obtain the algorithm individual with the smallest fitness value in the current population and mark it as the current best individual. For each individual algorithm in the population, update its position vector based on the difference vector between the best individual and its own position; by using the best individual in the current population as the "wind direction", make the individual algorithms in the population move closer to the current best region, constrain the direction of parameter search, and reduce invalid random search in high-dimensional parameter space. Step S424: Simulate the random diffusion behavior of moss spores in the natural environment. Perform Gaussian perturbation on each individual algorithm in the population to expand the search range of the parameter space and prevent the algorithm from getting trapped in local optima. The diffusion model update expression is: , in, These are random numbers distributed according to a standard normal distribution. For the updated algorithm individual; This refers to the current algorithm instance, specifically the i-th candidate parameter vector in the t-th iteration. The diffusion step size is determined based on the turbulence intensity under the current operating conditions of the wind turbine. Dynamic adjustment: When turbulence intensity When the wind condition is stable, a large step size is used. To broaden the scope of the global search; When turbulence intensity When this occurs, it indicates that the current operating condition is turbulent wind condition, and a small step size is adopted. To improve the accuracy of local searches; When turbulence intensity At that time, linear interpolation is used to determine the diffusion step size; Step S425: To improve the algorithm's local search accuracy in the neighborhood of the optimal solution, perform the following double breeding operation on each individual algorithm in the current population: (1) Crossover reproduction: The current algorithm individual is weighted and merged with the current global best individual to generate crossover offspring; (2) Mutation reproduction: Based on crossbreeding, with a preset probability Randomly select a parameter dimension j and apply a small random perturbation to the parameter value of that dimension; (3) Greedy selection: Compare the fitness values ​​of the original individuals, crossover offspring and mutated offspring, and retain the best individuals to enter the next generation of the population; Step S426: During the iterative process of the moss optimization algorithm, record and retain the historical optimal parameter vector. The fitness value of the individual is determined by the algorithm. When the algorithm is detected to be trapped in a local optimum, a hidden backtracking mechanism is triggered. The hidden backtracking mechanism includes restoring the current population to its historical optimal state and restarting the search to escape the local optimum and ensure the algorithm's global optimization ability. The condition for being trapped in a local optimum is that the fitness value of the best individual in the population has not improved during L consecutive iterations. After each iteration of the moss optimization algorithm, it is determined whether any of the following termination conditions are met: (1) The current iteration number t reaches the preset maximum iteration number G; (2) The convergence change of the fitness value over M consecutive generations is lower than the preset accuracy threshold; If any condition is met, the iteration terminates, and the optimal parameter vector in the current population is output. Otherwise, continue to the next iteration and return to step S422.

[0016] This step can achieve the following technical effects: In the moss optimization algorithm, dimensional differences are eliminated through population initialization; wind direction guidance leads the population to converge toward the optimal region, reducing ineffective random searches; spore diffusion is designed to adaptively adjust the step size based on turbulence intensity, achieving global exploration adapted to wind conditions; dual reproduction is designed to achieve refined search of the optimal neighborhood through cross-fusion and single-dimensional fine-tuning; a cryptic mechanism is designed to record historical optimal states, triggering backtracking when stagnant to escape local optima; this improved moss optimization algorithm process balances global exploration efficiency and local development accuracy, ensuring global optimization capabilities in high-dimensional parameter space.

[0017] Preferably, the engineered optimization process of step S5, which involves "parameter distribution - silent period - parameter stability confirmation - sampling window - similarity evaluation," specifically includes: Step S51: Generate the optimal parameter vector using the moss optimization algorithm. The parameter interface of the digital twin model is sent through the OPC UA communication protocol or Modbus TCP communication protocol to replace the current key parameter vector to be optimized in the digital twin model. Step S52: After the key parameter vector to be optimized is distributed, set the quiet period duration. Entering a silent period, during which no simulation output data is collected, and waiting for the digital twin model to complete transient adjustments and enter a steady state; Step S53: After the quiet period ends, continuously monitor the sliding window standard deviation of the simulation output time series variables of the digital twin model; when the standard deviation of three consecutive windows is lower than the preset threshold, it is determined that the digital twin model has entered a stable state; otherwise, continue to wait. Step S54: After the digital twin model is stably confirmed, open the length of... The sampling window simultaneously collects measured data from the physical fan. Simulation output data of digital twin models ; Step S55: Collect the measured data within the sampling window. With simulation output data The similarity calculation model of the neural deformable network is input, and the similarity score and fitness value are calculated and fed back to the moss optimization algorithm.

[0018] This step can achieve the following technical effects: In order to ensure the compatibility of parameter search with actual industrial wind power systems, an engineering optimization process of "parameter issuance - quiet period - parameter stability confirmation - sampling window - similarity evaluation" was designed in this step. This ensures the stable output of the digital twin model corresponding to the fitness evaluation of each candidate parameter, reduces evaluation noise caused by system transient non-convergence, and improves the accuracy of parameter optimization.

[0019] Furthermore, this invention also provides an online update system for wind turbine twin parameters based on neural morphology and moss optimization algorithms, comprising: The physical fan data acquisition unit contains: Based on the structural topology, air-elastic-mechanical-electric coupling mechanism and control logic of the physical wind turbine, a digital twin model of the wind turbine is constructed; the key parameter vector to be optimized in the digital twin model is determined, and the key parameter vector to be optimized is input into the digital twin model to obtain the simulation output time series; Digital twin simulation modeling unit, in which: Collect multivariate time-series operating data of physical fans during actual operation; perform engineering preprocessing on the collected raw multivariate time-series operating data to obtain the output time series of physical fans; The neural morphology similarity calculation unit contains: A neural morphing network similarity calculation model based on "encoder-morphing module-aggregation module" is constructed; based on the neural morphing network similarity calculation model and the simulation output time series, a continuous similarity score is output; The moss optimization parameter search unit contains: Using the similarity score output by the neural morphology network as the optimization objective, a fitness function for the moss optimization algorithm is constructed, transforming the parameter update problem of the digital twin model into a fitness function minimization problem. By utilizing the global search characteristic of the moss optimization algorithm, the optimal parameter vector is searched within the parameter feasible region to achieve high-precision parameter optimization. The online calibration unit for model parameters includes: The optimal parameter vector is denormalized; it is then sent online to the digital twin model of the wind turbine to replace the key parameter vector to be optimized. The parameter calibration and update of the digital twin model are completed through an engineering optimization process of "parameter sending - quiet period - parameter stability confirmation - sampling window - similarity evaluation", so that the output of the digital twin model and the operating data of the physical wind turbine maintain a high degree of dynamic consistency. Iterative optimization and control unit, in which: The system continuously collects multivariate time-series operational data of the physical wind turbine and repeatedly executes the contents of the digital twin simulation modeling unit, neural deformation similarity calculation unit, moss optimization parameter search unit, and model parameter online calibration unit within a fixed time window period. This dynamically updates the parameters of the digital twin model, suppresses the "twin deviation" phenomenon in real time, and ensures the dynamic consistency and engineering credibility of the digital twin model throughout the entire life cycle of the wind turbine.

[0020] The beneficial effects of this invention are as follows: It enables deep, elastic similarity evaluation, accurately characterizing the differences in dynamic behavior between virtual and real systems. A neural deformation network consisting of an encoder, deformation module, and aggregation module is constructed. Adaptive time alignment is achieved through a soft-aligned weight matrix, simultaneously characterizing the dual consistency of dynamic trends and amplitudes, thus solving the technical problem of distortion in similarity evaluation. It also enables efficient global optimization, overcoming the bottleneck of high-dimensional search. Addressing the problem of low search efficiency and susceptibility to local optima in high-dimensional parameter spaces by traditional metaheuristic algorithms, a moss optimization algorithm is introduced. Through four mechanisms—wind direction guidance, spore diffusion, dual reproduction, and cryptic growth—a dynamic balance between global exploration and local development is achieved, improving convergence speed and optimization accuracy. Furthermore, through an engineering-based closed-loop design, industrial applicability is ensured. Addressing the problem of existing methods' strong dependence on precise mathematical models and poor engineering practicality, an engineering process of "parameter distribution—quiet period—parameter stability confirmation—sampling window—similarity evaluation" is constructed. This suppresses transient noise interference, achieves online parameter updates compatible with existing SCADA systems in wind farms, and ensures the dynamic consistency of the digital twin model throughout its entire lifecycle.

[0021] Therefore, it is evident that the present invention has outstanding substantive features and significant progress compared with the prior art, and the beneficial effects of its implementation are also obvious. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating the framework of an online update method for wind turbine twin parameters based on neural deformation and moss optimization algorithms provided by this invention.

[0024] Figure 2 This is a schematic diagram of the principle of an online update system for wind turbine twin parameters based on neural deformation and moss optimization algorithms provided by the present invention.

[0025] The system comprises: 1-Physical fan data acquisition unit; 2-Digital twin simulation modeling unit; 3-Neural deformity similarity calculation unit; 4-Moss optimization parameter search unit; 5-Model parameter online calibration unit; and 6-Iterative optimization and control unit. Detailed Implementation

[0026] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The following embodiments are explanations of the present invention, but the present invention is not limited to the following implementation methods.

[0027] Example 1: like Figure 1 As shown, an online update method for wind turbine twin parameters based on neural morphology and moss optimization algorithms includes the following steps: Step S1: Based on the structural topology, air-elastic-mechanical-electric coupling mechanism and control logic of the physical wind turbine, construct a digital twin model of the wind turbine unit; determine the key parameter vector to be optimized in the digital twin model, input the key parameter vector to be optimized into the digital twin model, and obtain the simulation output time series; Step S1 specifically includes: Step S11: Based on the structural topology, aero-elastic-mechanical-electric coupling mechanism and control logic of the physical wind turbine, construct a high-fidelity digital twin model of the wind turbine. The digital twin model includes: wind turbine aerodynamic model, synchronous generator model, grid-connected converter model, transmission chain mechanical model, etc., to fully reproduce the dynamic operating characteristics of the physical wind turbine. Step S12, determine the key parameter vector θ to be optimized in the digital twin model; the key parameter vector θ to be optimized consists of the core electrical and mechanical parameters that affect the dynamic behavior of the wind turbine. , in, For stator leakage, For rotor leakage inductance, For stator resistance, For rotor resistance, It is the inertial constant. For magnetizing inductance, For the resistance of branch RL, For the RL branch inductance, For parallel capacitors; the initial values ​​of each parameter are based on the factory calibration values ​​of the physical fan or the benchmark measured values, and the search range is limited to a reasonable range that is physically permissible; Step S13: Input the vector of key parameters to be optimized into the digital twin model to obtain the simulation output time series of the digital twin model. .

[0028] This step can achieve the following technical effects: By constructing a high-fidelity digital twin model of the wind turbine and clearly defining the core electrical and mechanical parameters to be optimized, a foundation was laid for subsequent parameter identification. By inputting the parameter vectors to be optimized into the digital twin model to obtain the simulation output time series, a complete correspondence between the physical wind turbine and the digital twin model at the input and output levels was achieved, ensuring that the two are comparable and providing unified data and support for subsequent parameter optimization based on deep elastic similarity.

[0029] Step S2: Collect multivariate time-series operating data of the physical fan during actual operation; perform engineering preprocessing on the collected raw multivariate time-series operating data to obtain the output time series of the physical fan; Step S2 specifically includes: Step S21: Collect multivariate time-series operating data of the physical wind turbine during actual operation. The multivariate time-series operating data are core electromechanical coupling variables that correspond one-to-one with the output of the digital twin model, including: generator stator active power, stator current RMS value, stator voltage RMS value, DC bus voltage, electromagnetic torque, rotor speed, reactive power, etc., which constitute the output time series of the physical wind turbine. Step S22 involves performing engineering preprocessing on the raw multivariate time-series operating data in the physical wind turbine output time series. This includes outlier removal, missing value completion, time alignment, and data normalization to eliminate measurement noise and interference from the wind turbine's transient processes, ensuring the validity and consistency of the data, and obtaining the preprocessed physical wind turbine output time series. Where T is the time series length and C is the number of variable channels.

[0030] This step can achieve the following technical effects: By collecting core electromechanical coupling variables, a complete operating state characterization of the physical wind turbine was constructed. Through engineering preprocessing operations, the interference of measurement noise and transient processes of the wind turbine was eliminated, ensuring the validity and consistency of the benchmark reference data and improving the accuracy of subsequent similarity evaluation and parameter optimization.

[0031] Step S3: Construct a neural morphing network similarity calculation model based on "encoder-morphing module-aggregation module"; output continuous similarity scores based on the neural morphing network similarity calculation model and the simulation output time series. Step S3 specifically includes: Step S31: Construct a neural morphing network similarity calculation model. The neural morphing network similarity calculation model includes: an encoder, a morphing module, and an aggregation module. The encoder is used for multi-scale feature extraction, the morphing module is used for soft alignment weight learning, and the aggregation module is used for global similarity scoring. Step S32: Output time series of physical fan The simulation output time series obtained in step S1 Input encoder Perform multi-scale feature extraction: extract the original time series data and Mapping to a high-dimensional feature space yields a sequence of physical features. With simulation feature sequence To achieve dimensionality reduction and feature enhancement of data: , in, , D is the feature dimension. The encoder is constructed using CNN or LSTM and has the ability to extract local and global features from time series data. Step S33: Learn the soft alignment weight matrix in the time dimension through the deformation module. The soft alignment weight matrix is ​​used to characterize the nonlinear correspondence between the physical feature sequence and the simulation feature sequence in the time dimension: , in, This is a similarity metric function that can calculate the physical feature sequence at time i. Simulated feature sequence at time j Local similarity; It is a normalization function along the column dimension, which makes the weight matrix satisfy the probability distribution characteristics, thereby achieving the focus alignment of key time series segments and the weakening of non-key segments; The element in the i-th row and j-th column of the soft alignment weight matrix represents the i-th time step of the physical feature sequence. With the simulated feature sequence at time j Alignment weights between them; Step S34, in the aggregation module, adjust the soft alignment weight matrix. The global continuous similarity score is obtained by weighted summation with local similarity and then mapped using the sigmoid activation function. This enables a comprehensive evaluation of the dynamic trends and amplitude consistency of the virtual and real systems. , in, For the Sigmoid activation function, For the local similarity function in the feature space, This serves as a normalization factor to ensure consistency in the scoring scale.

[0032] This step can achieve the following technical effects: This step constructs a three-segment neural deformable network similarity calculation model consisting of an encoder, a deformable module, and an aggregation module. Multi-scale feature extraction and dimensionality reduction enhancement are achieved through the encoder using either a CNN or LSTM. The deformable module learns a time-dimensional soft-alignment weight matrix, avoiding the shortcomings of forced point-by-point alignment in DTW and enabling adaptive handling of phase drift and local deformation. The aggregation module outputs a continuous similarity score after Sigmoid activation. This model can simultaneously characterize the dual consistency of dynamic trends and amplitudes in both virtual and real systems, improving the accuracy and robustness of similarity evaluation.

[0033] Step S4: Using the similarity score output by the neural morphology network as the optimization objective, construct the fitness function of the moss optimization algorithm, transforming the parameter update problem of the digital twin model into a fitness function minimization problem; using the global search characteristics of the moss optimization algorithm, search for the optimal parameter vector within the parameter feasible region to achieve high-precision parameter optimization. Step S4 specifically includes: Step S41, define the fitness function of the moss optimization algorithm as follows: , in, The fitness value is represented by the complement of the similarity score. Therefore, the smaller the fitness value, the higher the degree of consistency between the dynamic behavior of the physical wind turbine and the digital twin model. This represents the continuous similarity score output by the neural morphology network similarity calculation model. Step S42 involves designing four core mechanisms: wind direction guidance, spore dispersal, dual reproduction, and cryptobiosis. A moss optimization algorithm is then employed to achieve a dynamic balance between global exploration and local exploitation in a high-dimensional parameter space, yielding the optimal parameter vector. .

[0034] This step can achieve the following technical effects: By using the similarity score output by the neural morphology network as the optimization objective, a moss optimization algorithm with the similarity score complement as the fitness function is constructed, transforming the parameter update problem into a fitness function minimization problem. By designing four core mechanisms—wind guidance, spore dispersal, dual reproduction, and cryptobiosis—a dynamic balance between global exploration and local exploitation in a high-dimensional parameter space is achieved. Compared with traditional metaheuristic algorithms, this algorithm has the advantages of faster convergence speed, higher search accuracy, and stronger resistance to local optima.

[0035] The specific search process of the moss optimization algorithm in step S42 is as follows: Step S421, Population Initialization: Assign each individual algorithmic unit of the moss optimization algorithm to a parameter vector to be optimized. An initial population is randomly generated within the physical feasible region of each parameter, and each parameter is normalized to a uniform interval to eliminate the interference of dimensional differences on the search process. Step S422: Select each parameter vector to be optimized in the population as a candidate parameter vector. Substitute the digital twin model into the simulation to obtain the simulation output time series. Similarity scores are calculated using a neural morphology network similarity calculation model. And obtain the fitness value according to the fitness function. ; Step S423, Wind-guided evolution: Obtain the algorithm individual with the smallest fitness value in the current population and mark it as the current best individual. For each individual algorithm in the population, update its position vector based on the difference vector between the best individual and its own position: , in, This is the wind direction guidance coefficient. A random number within the interval [0,1]. For the updated algorithm individual, Let i be the i-th algorithm individual; by taking the best individual in the current population as the "wind direction", the algorithm individuals in the population are made to move closer to the current best region, which constrains the parameter search direction and reduces invalid random search in the high-dimensional parameter space; Step S424: Simulate the random diffusion behavior of moss spores in the natural environment. Perform Gaussian perturbation on each individual algorithm in the population to expand the search range of the parameter space and prevent the algorithm from getting trapped in local optima. The diffusion model update expression is: , in, These are random numbers distributed according to a standard normal distribution. For the updated algorithm individual; This refers to the current algorithm instance, specifically the i-th candidate parameter vector in the t-th iteration. The diffusion step size is determined based on the turbulence intensity under the current operating conditions of the wind turbine. Dynamic adjustment: When turbulence intensity When the current operating condition is stable with low turbulence intensity, a large step size is used. To expand the global search scope, among which The lower limit threshold of turbulence intensity for stable wind conditions. This is a preset value for a large step size; When turbulence intensity When the current operating condition is turbulent with high turbulence intensity, a small step size is used. To improve local search accuracy, among which This represents the upper limit threshold for turbulence intensity in turbulent wind conditions. Preset value for small step size; When turbulence intensity At that time, linear interpolation is used to determine the diffusion step size; Step S425: To improve the algorithm's local search accuracy in the neighborhood of the optimal solution, perform the following double breeding operation on each individual algorithm in the current population: (1) Crossover: The current algorithm individual is weighted and merged with the current global best individual to generate crossover offspring. , in, The fusion coefficient, with a value range of [0, 1], is used to control the degree to which the current individual moves closer to the optimal individual; Offspring individuals generated through crossbreeding; It is the currently globally optimal individual; For the current algorithm individual; (2) Mutation reproduction: Based on crossbreeding, with a preset probability Randomly select a parameter dimension j, and apply a small random perturbation to the parameter value of that dimension: , in, To fine-tune the step size, the value is set to 1% to 5% of the search range for this parameter; A uniformly random number within the interval [-1, 1]; The uniformly random numbers, together with the fine-tuning step size, constitute the random perturbation term. ; Offspring individuals generated through mutation and reproduction; To crossbreed offspring; (3) Greedy selection: Compare the fitness values ​​of the original individuals, crossover offspring and mutated offspring, and retain the best individuals to enter the next generation of the population; Step S426: During the iterative process of the moss optimization algorithm, record and retain the historical optimal parameter vector. The fitness value of the individual is determined by the algorithm. When the algorithm is detected to be trapped in a local optimum, a hidden backtracking mechanism is triggered. The hidden backtracking mechanism includes restoring the current population to its historical optimal state and restarting the search to escape the local optimum and ensure the algorithm's global optimization ability. The condition for being trapped in a local optimum is that the fitness value of the best individual in the population has not improved during L consecutive iterations. After each iteration of the moss optimization algorithm, it is determined whether any of the following termination conditions are met: (1) The current iteration number t reaches the preset maximum iteration number G; (2) The convergence change of the fitness value over M consecutive generations is lower than the preset accuracy threshold; If any condition is met, the iteration terminates, and the optimal parameter vector in the current population is output. Otherwise, continue to the next iteration and return to step S422.

[0036] This step can achieve the following technical effects: In the moss optimization algorithm, dimensional differences are eliminated through population initialization; wind direction guidance leads the population to converge toward the optimal region, reducing ineffective random searches; spore diffusion is designed to adaptively adjust the step size based on turbulence intensity, achieving global exploration adapted to wind conditions; dual reproduction is designed to achieve refined search of the optimal neighborhood through cross-fusion and single-dimensional fine-tuning; a cryptic mechanism is designed to record historical optimal states, triggering backtracking when stagnant to escape local optima; this improved moss optimization algorithm process balances global exploration efficiency and local development accuracy, ensuring global optimization capabilities in high-dimensional parameter space.

[0037] Step S5: Inverse normalize the optimal parameter vector; send it online to the digital twin model of the wind turbine, replace the key parameter vector to be optimized, and complete the parameter calibration and update of the digital twin model through the engineering optimization process of "parameter sending-quiet period-parameter stability confirmation-sampling window-similarity evaluation", so that the output of the digital twin model maintains a high degree of dynamic consistency with the operating data of the physical wind turbine. The engineered optimization process of "parameter issuance - silent period - parameter stability confirmation - sampling window - similarity evaluation" in step S5 specifically includes: Step S51: Generate the optimal parameter vector using the moss optimization algorithm. The parameter interface of the digital twin model is sent through the OPC UA communication protocol or Modbus TCP communication protocol to replace the current key parameter vector to be optimized in the digital twin model. Step S52: After the key parameter vector to be optimized is distributed, set the quiet period duration. The time is 5-10 seconds; then a silent period begins, during which no simulation output data is collected, and the digital twin model is allowed to complete its transient adjustment and enter a steady state. Step S53: After the quiet period ends, continuously monitor the sliding window standard deviation of the simulation output time series variables of the digital twin model; when the standard deviation of three consecutive windows is lower than the preset threshold, it is determined that the digital twin model has entered a stable state; otherwise, continue to wait. Step S54: After the digital twin model is stably confirmed, open the length of... A sampling window of seconds is used to simultaneously collect measured data from the physical fan. Simulation output data of digital twin models ; Step S55: Collect the measured data within the sampling window. With simulation output data The similarity calculation model of the neural deformable network is input, and the similarity score and fitness value are calculated and fed back to the moss optimization algorithm.

[0038] This step can achieve the following technical effects: In order to ensure the compatibility of parameter search with actual industrial wind power systems, an engineering optimization process of "parameter issuance - quiet period - parameter stability confirmation - sampling window - similarity evaluation" was designed in this step. This ensures the stable output of the digital twin model corresponding to the fitness evaluation of each candidate parameter, reduces evaluation noise caused by system transient non-convergence, and improves the accuracy of parameter optimization.

[0039] Step S6 involves continuously collecting multivariate time-series operating data of the physical wind turbine and repeating steps S2-S5 at fixed time windows to dynamically update the parameters of the digital twin model, suppress the "twin deviation" phenomenon in real time, and ensure the dynamic consistency and engineering credibility of the digital twin model throughout the entire life cycle of the wind turbine.

[0040] Example 2: like Figure 2 As shown, an online update system for wind turbine twin parameters based on neural morphology and moss optimization algorithms includes: Physical fan data acquisition unit 1, in which: Based on the structural topology, air-elastic-mechanical-electric coupling mechanism and control logic of the physical wind turbine, a digital twin model of the wind turbine is constructed; the key parameter vector to be optimized in the digital twin model is determined, and the key parameter vector to be optimized is input into the digital twin model to obtain the simulation output time series; Digital twin simulation modeling unit 2, in which: Collect multivariate time-series operating data of physical fans during actual operation; perform engineering preprocessing on the collected raw multivariate time-series operating data to obtain the output time series of physical fans; Neural Deformation Similarity Calculation Unit 3, in which: A neural morphing network similarity calculation model based on "encoder-morphing module-aggregation module" is constructed; based on the neural morphing network similarity calculation model and the simulation output time series, a continuous similarity score is output; Moss optimization parameter search unit 4, in which: Using the similarity score output by the neural morphology network as the optimization objective, a fitness function for the moss optimization algorithm is constructed, transforming the parameter update problem of the digital twin model into a fitness function minimization problem. By utilizing the global search characteristic of the moss optimization algorithm, the optimal parameter vector is searched within the parameter feasible region to achieve high-precision parameter optimization. Online calibration unit 5 for model parameters, in which: The optimal parameter vector is denormalized; it is then sent online to the digital twin model of the wind turbine to replace the key parameter vector to be optimized. The parameter calibration and update of the digital twin model are completed through an engineering optimization process of "parameter sending - quiet period - parameter stability confirmation - sampling window - similarity evaluation", so that the output of the digital twin model and the operating data of the physical wind turbine maintain a high degree of dynamic consistency. Iterative optimization and control unit 6, in which: The system continuously collects multivariate time-series operational data of the physical wind turbine and repeatedly executes the contents of the digital twin simulation modeling unit 2, neural deformation similarity calculation unit 3, moss optimization parameter search unit 4, and model parameter online calibration unit 5 within a fixed time window period. This dynamically updates the parameters of the digital twin model, suppresses the "twin deviation" phenomenon in real time, and ensures the dynamic consistency and engineering credibility of the digital twin model throughout the entire life cycle of the wind turbine.

[0041] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the systems disclosed in the embodiments; relevant details can be found in the method section.

[0042] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0043] In the embodiments provided by this invention, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0044] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0045] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit.

[0046] Similarly, in the various embodiments of the present invention, each processing unit can be integrated into a functional module, or each processing unit can exist physically, or two or more processing units can be integrated into a functional module.

[0047] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0048] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0049] The above-disclosed embodiments are merely preferred embodiments of the present invention, but the present invention is not limited thereto. Any non-creative variations that can be conceived by those skilled in the art, as well as any improvements and modifications made without departing from the principles of the present invention, should fall within the protection scope of the present invention.

Claims

1. A method for online updating of wind turbine twin parameters based on neural morphology and moss optimization algorithms, characterized in that, Includes the following steps: Step S1: Construct a digital twin model of the wind turbine; determine the key parameter vector to be optimized in the digital twin model, input the key parameter vector to be optimized into the digital twin model, and obtain the simulation output time series; Step S2: Collect multivariate time-series operating data of the physical wind turbine during actual operation; The collected raw multivariate time-series operating data are preprocessed in an engineering manner to obtain the physical wind turbine output time series; Step S3: Construct a neural morphing network similarity calculation model based on "encoder-morphing module-aggregation module"; output continuous similarity scores based on the neural morphing network similarity calculation model and the simulation output time series. Step S4: Using the similarity score output by the neural morphology network as the optimization objective, construct the fitness function of the moss optimization algorithm; use the moss optimization algorithm to search for the optimal parameter vector within the parameter feasible region; Step S5: Inverse normalize the optimal parameter vector; The parameters of the digital twin model of the wind turbine are sent offline to replace the key parameter vectors to be optimized, and the parameter calibration and update of the digital twin model are completed through an engineering optimization process. Step S6: Continuously collect multivariate time-series operating data of the physical wind turbine, and repeat steps S2-S5 at fixed time windows to dynamically update the parameters of the digital twin model.

2. The online update method for wind turbine twin parameters based on neural morphology and moss optimization algorithm according to claim 1, characterized in that, Step S1 specifically includes: Step S11: Based on the physical wind turbine's structural topology, air-elastic-mechanical-electric coupling mechanism, and control logic, construct a high-fidelity digital twin model of the wind turbine. Step S12: Determine the key parameter vector θ to be optimized in the digital twin model; Step S13: Input the vector of key parameters to be optimized into the digital twin model to obtain the simulation output time series of the digital twin model. .

3. The online update method for wind turbine twin parameters based on neural morphology and moss optimization algorithm according to claim 2, characterized in that, The key parameter vector θ to be optimized consists of the core electrical and mechanical parameters that affect the dynamic behavior of the wind turbine. ， in, For stator leakage, For rotor leakage inductance, For stator resistance, For rotor resistance, It is the inertial constant. For magnetizing inductance, For the resistance of branch RL, For the RL branch inductance, For parallel capacitors; the initial values ​​of each parameter are based on the factory calibration values ​​of the physical fan or the benchmark measured values, and the search range is limited to a reasonable range that is physically permissible.

4. The online update method for wind turbine twin parameters based on neural morphology and moss optimization algorithm according to claim 1, characterized in that, Step S2 specifically includes: Step S21: Collect multivariate time-series operation data of the physical wind turbine during actual operation. The multivariate time-series operation data are core electromechanical coupling variables that correspond one-to-one with the output of the digital twin model, including: generator stator active power, stator current RMS value, stator voltage RMS value, DC bus voltage, electromagnetic torque, rotor speed, and reactive power, which constitute the output time series of the physical wind turbine. Step S22 involves performing engineering preprocessing on the raw multivariate time-series operating data in the output time series of the physical wind turbines, including outlier removal, missing value completion, time alignment, and data normalization, to obtain the preprocessed output time series of the physical wind turbines. Where T is the time series length and C is the number of variable channels.

5. The online update method for wind turbine twin parameters based on neural morphology and moss optimization algorithm according to claim 1, characterized in that, Step S3 specifically includes: Step S31: Construct a neural morphology network similarity calculation model. The neural morphology network similarity calculation model includes: encoder, morphology module, and aggregation module. Step S32: Output time series of physical fan The simulation output time series obtained in step S1 Encoder of input neural morphology network similarity calculation model Perform multi-scale feature extraction: extract the original time series data and Mapping to a high-dimensional feature space yields a sequence of physical features. With simulation feature sequence The encoder is constructed using CNN or LSTM and has the ability to extract local and global features from time-series data. Step S33: Learn the soft alignment weight matrix in the time dimension through the deformable module of the neural deformable network similarity calculation model. The soft alignment weight matrix represents the nonlinear correspondence between the physical feature sequence and the simulation feature sequence in the time dimension: ， in, Let be the similarity metric function, and calculate the physical feature sequence at time i. Simulated feature sequence at time j Local similarity; For the normalization function along the column dimension; The element in the i-th row and j-th column of the soft alignment weight matrix represents the i-th time step of the physical feature sequence. Compared with the simulation feature sequence at time j Alignment weights between them; Step S34: In the aggregation module of the neural deformable network similarity calculation model, the soft alignment weight matrix is... The global continuous similarity score is obtained by weighted summation with local similarity and then mapped using the sigmoid activation function. : ， in, The Sigmoid activation function is used. For the local similarity function in the feature space, This is the normalization factor.

6. The online update method for wind turbine twin parameters based on neural morphology and moss optimization algorithm according to claim 1, characterized in that, Step S4 specifically includes: Step S41, define the fitness function of the moss optimization algorithm; Step S42 involves designing four core mechanisms: wind direction guidance, spore dispersal, dual reproduction, and cryptobiosis. A moss optimization algorithm is then used to dynamically adjust global exploration and local exploitation in a high-dimensional parameter space to obtain the optimal parameter vector. .

7. The online update method for wind turbine twin parameters based on neural morphology and moss optimization algorithm according to claim 6, characterized in that, The fitness function of the moss optimization algorithm is: ， in, The fitness value is the value of fitness. The smaller the fitness value, the higher the consistency between the dynamic behavior of the physical wind turbine and the digital twin model. This represents the continuous similarity score output by the neural deformable network similarity calculation model.

8. The online update method for wind turbine twin parameters based on neural morphology and moss optimization algorithm according to claim 6, characterized in that, The specific search process of the moss optimization algorithm in step S42 is as follows: Step S421, Population Initialization: Assign each individual algorithmic unit of the moss optimization algorithm to a parameter vector to be optimized. An initial population is randomly generated within the physical feasible region of each parameter, and each parameter is normalized to a unified interval. Step S422: Select each parameter vector to be optimized in the population as a candidate parameter vector. Substitute the digital twin model into the simulation to obtain the simulation output time series. The similarity score is calculated using a neural deformable network similarity calculation model, and the fitness value is obtained based on the fitness function. Step S423, Wind-guided evolution: Obtain the algorithm individual with the smallest fitness value in the current population and mark it as the current best individual. For each individual algorithm in the population, update its position vector based on the difference vector between the best individual and its own position. Step S424: Simulate the random diffusion behavior of moss spores in the natural environment. Apply a Gaussian perturbation to each individual algorithm in the population. The diffusion model update expression is: ， in, These are random numbers distributed according to a standard normal distribution. For the updated algorithm individual; This refers to the current algorithm instance, specifically the i-th candidate parameter vector in the t-th iteration. The diffusion step size is determined based on the turbulence intensity under the current operating conditions of the wind turbine. Dynamic adjustment: When turbulence intensity At that time, a large step size was adopted. ,in The lower limit threshold of turbulence intensity for stable wind conditions. This is a preset value for a large step size; When turbulence intensity At that time, a small step size is adopted. ,in This represents the upper limit threshold for turbulence intensity in turbulent wind conditions. Preset value for small step size; When turbulence intensity At that time, linear interpolation is used to determine the diffusion step size; Step S425, perform the following double reproduction operation on each algorithm individual in the current population: (1) Crossover reproduction: The current algorithm individual is weighted and merged with the current global best individual to generate crossover offspring; (2) Mutation reproduction: Based on crossbreeding, with a preset probability Randomly select a parameter dimension j and apply a small random perturbation to the parameter value of that dimension; (3) Greedy selection: Compare the fitness values ​​of the original individuals, crossover offspring and mutated offspring, and retain the best individuals to enter the next generation of the population; Step S426: During the iterative process of the moss optimization algorithm, record and retain the historical optimal parameter vector. The algorithm detects the fitness value of the individual and its fitness value. When the algorithm is detected to be trapped in a local optimum, a hidden backtracking mechanism is triggered. The hidden backtracking mechanism includes restoring the current population to its historical optimal state and restarting the search. The condition for being trapped in a local optimum is that the fitness value of the best individual in the population has not improved during L consecutive iterations. After each iteration of the moss optimization algorithm, it is determined whether any of the following termination conditions are met: (1) The current iteration number t reaches the preset maximum iteration number G; (2) The convergence change of the fitness value over M consecutive generations is lower than the preset accuracy threshold; If any condition is met, the iteration terminates, and the optimal parameter vector in the current population is output. Otherwise, continue to the next iteration and return to step S422.

9. The online update method for wind turbine twin parameters based on neural morphology and moss optimization algorithm according to claim 1, characterized in that, The engineering optimization process in step S5 specifically includes: Step S51: Generate the optimal parameter vector using the moss optimization algorithm. The parameter interface is sent to the digital twin model via the communication protocol to replace the current key parameter vector to be optimized in the digital twin model. Step S52: After the key parameter vector to be optimized is distributed, set the quiet period duration. Entering a silent period, during which no simulation output data is collected, and waiting for the digital twin model to complete transient adjustments and enter a steady state; Step S53: After the quiet period ends, continuously monitor the sliding window standard deviation of the simulation output time series variables of the digital twin model; when the standard deviation of three consecutive windows is lower than the preset threshold, it is determined that the digital twin model has entered a stable state; otherwise, continue to wait. Step S54: After the digital twin model is stably confirmed, open the length of... The sampling window simultaneously collects measured data from the physical fan. Simulation output data of digital twin models ; Step S55: Collect the measured data within the sampling window. With simulation output data The similarity calculation model of the neural deformable network is input, and the similarity score and fitness value are calculated and fed back to the moss optimization algorithm.

10. A wind turbine twin parameter online update system based on neural morphology and moss optimization algorithms, characterized in that, include: The physical fan data acquisition unit contains: Constructing a digital twin model of a wind turbine; Determine the key parameter vector to be optimized in the digital twin model, input the key parameter vector to be optimized into the digital twin model, and obtain the simulation output time series; Digital twin simulation modeling unit, in which: Collect multivariable time-series operational data of physical wind turbines during actual operation; The collected raw multivariate time-series operating data are preprocessed in an engineering manner to obtain the physical wind turbine output time series; The neural morphology similarity calculation unit contains: A neural morphing network similarity calculation model based on "encoder-morphing module-aggregation module" is constructed; based on the neural morphing network similarity calculation model and the simulation output time series, a continuous similarity score is output; The moss optimization parameter search unit contains: The similarity score output by the neural morphology network is used as the optimization objective to construct the fitness function of the moss optimization algorithm; the moss optimization algorithm is then used to search for the optimal parameter vector within the parameter feasible region. The online calibration unit for model parameters includes: Inverse normalize the optimal parameter vector; The parameters of the digital twin model of the wind turbine are sent offline to replace the key parameter vectors to be optimized, and the parameter calibration and update of the digital twin model are completed through an engineering optimization process. Iterative optimization and control unit, in which: The system continuously collects multivariate time-series operational data of the physical wind turbine and repeatedly executes the contents of the digital twin simulation modeling unit, neural deformation similarity calculation unit, moss optimization parameter search unit, and model parameter online calibration unit within a fixed time window period, dynamically updating the parameters of the digital twin model.

Images