Wind turbine generator part topology-parameter collaborative optimization method fused with artificial intelligence

Abstract

According to the wind turbine generator part topology-parameter collaborative optimization method fused with artificial intelligence, design is carried out based on actual working conditions, loads and spatial constraints of wind turbine generator parts, it is ensured that an initial topology configuration conforms to an engineering application scene, a variable density method is adopted to solve a material optimal distribution field, and the optimal distribution field is obtained. Materials are distributed in a key stress area in a concentrated mode, non-key areas are reduced reasonably, engineering geometric reconstruction is carried out on an initial topological configuration, key structure parameters are extracted, a parameterized model is established, and a core performance response is obtained by establishing a mapping relation between the key structure parameters and core performance responses. Performance evaluation is converted into model rapid prediction from high-fidelity simulation time consumption calculation, the single performance evaluation time is shortened to a minute level from tens of hours, an optimal parameter combination is obtained based on the AI agent model and in combination with a multi-objective optimization algorithm, the optimization iteration speed is greatly increased, the reliability of the optimization effect is guaranteed, and the performance evaluation efficiency is improved. And the research and development period is further shortened.

Description

Technical Field

[0001] This invention relates to the field of wind turbine operation technology, and in particular to a method for collaborative optimization of topology and parameters of wind turbine components that integrates artificial intelligence. Background Technology

[0002] With the rapid development of the wind power industry, the trend towards larger wind turbine units has become significant. However, the weight of key components such as towers, nacelles, and hubs has increased dramatically with size, leading to a substantial increase in material consumption, transportation, and installation costs, which restricts the economic viability of the wind power industry.

[0003] Lightweight design is the core means to solve the above problems. Traditional lightweight design relies on high-fidelity finite element analysis technology. However, the components of wind turbines have complex structures, and the simulation model with millions of elements makes a single optimization iteration take tens of hours, which seriously reduces R&D efficiency. Meanwhile, existing domestic research often separates topology optimization from parameter optimization. Topology optimization only determines the macroscopic configuration of the structure, while parameter optimization only adjusts local parameters for a given configuration. The two lack synergy and coordination, making it difficult to achieve the global optimal design and fully explore the potential of lightweight design. Therefore, there is an urgent need for a technical solution that can improve simulation efficiency and achieve topology and parameter co-optimization in order to solve the bottleneck problem of lightweight design of existing wind turbine components. Summary of the Invention

[0004] This invention provides a topology-parameter collaborative optimization method for wind turbine components that integrates artificial intelligence, in order to solve the problems mentioned in the background art.

[0005] A topology-parameter collaborative optimization method for wind turbine components integrating artificial intelligence includes: S1: Based on the operating conditions, loads and spatial constraints of wind turbine components, the optimal distribution field of wind turbine components is solved using a topology optimization algorithm to obtain the initial topology configuration; S2: Perform geometric reconstruction on the initial topological configuration, extract key structural parameters, and establish a parametric finite element model; S3: Construct an AI proxy model between the key structural parameters and the performance response of wind turbine components; S4: Based on the AI ​​agent model, combined with a multi-objective optimization algorithm, the key structural parameters are globally optimized to obtain the optimal parameter combination.

[0006] Preferably, in step S1, based on the operating conditions, loads, and spatial constraints of the wind turbine components, a topology optimization algorithm is used to solve for the optimal distribution field of the wind turbine components to obtain the initial topology configuration, including: The historical actual operating data of wind turbine components within a preset time period is obtained. Typical operating conditions are extracted from the historical actual operating data, and the load probability distribution under each typical operating condition is determined. The spatial constraints are obtained by hierarchical coding with the installation boundary as the hard constraint and the hoisting space, processing space and operation and maintenance space as the soft constraints. Temperature and vibration impact data are obtained from historical operational data, and temperature and vibration field constraints are established. These constraints are then coupled and transformed into equivalent mechanical constraints. Based on the load probability distribution, spatial constraints, and equivalent mechanical constraints, the overall constraint rules are established, and the optimization objectives are established with minimizing material volume as the main objective and minimizing maximum stress and minimizing process complexity as auxiliary objectives. Based on the overall constraint rules and optimization objectives, the first distribution field of wind turbine components is obtained by iteratively solving the variable density method according to a preset number of iterations. The second distribution field is obtained by iteratively solving the first distribution field according to a preset number of iterations using the level set method. The second distribution field is then fed back to the variable density method for bidirectional feedback iteration, and finally the optimal distribution field is obtained. The initial topological configuration is obtained by constructing the optimal distribution field.

[0007] Preferably, the iterative process of using the variable density method and the level set method to analyze the distribution field of wind turbine components specifically involves: Using the distribution field as the coarse grid specification, the first stage of coarse iteration is performed using the variable density method and the level set method to obtain the coarse distribution field. The core stress region and non-core region are then extracted from the coarse distribution field. Using the core stress region as the fine grid and the non-core region as the medium grid, the coarse distribution field is subjected to a second-stage fine iteration using the variable density method and the level set method to obtain the fine distribution field. After obtaining the fine distribution field, iterative verification and adjustment are performed based on the overall constraint rules to finally obtain the optimal distribution field.

[0008] Preferably, in step S2, the initial topological configuration is geometrically reconstructed, key structural parameters are extracted, and a parametric finite element model is established, including: The core stress region, non-core stress region, and transition region of the initial topological configuration are obtained. The core stress region is backfitted to generate a continuous solid structure. The non-core stress region is simplified and reconstructed into a regular geometry. The transition region is smoothly connected between the core stress region and the non-core stress region through gradual wall thickness and rounded corner transition, thus obtaining the geometric reconstruction features. The geometric reconstruction features are filtered according to preset extraction criteria to obtain key structural parameters; Geometric expressions for key structural parameters are established, wind turbine components are modularized to obtain component modules, a mapping relationship between geometric expressions and component modules is established, and key structural parameters are simulated based on the mapping relationship to obtain a parametric finite element model.

[0009] Preferably, in step S3, constructing an AI proxy model between the key structural parameters and the performance response of wind turbine components includes: Generate a combination of sample points within the design space of the key structural parameters; The performance response parameters corresponding to the sample point combination are obtained by simulating the sample point combination using a parametric finite element model. A sample set is constructed based on the combination of sample points and performance response parameters. The neural network model is then trained to obtain the final AI agent model.

[0010] Preferably, generating a combination of sample points within the design space of the key structural parameters includes: The key structural parameters are classified and decoupled to obtain the target structural parameters; The target structural parameters are sampled first using the Latin hypercube sampling method to obtain the first sample. High target parameters with high impact on performance are screened out through sensitivity pre-analysis. Orthogonal arrays are designed for the high target parameters to perform the second sampling to obtain the second sample. Extreme operating conditions in the historical operating data of wind turbines are obtained to generate the third sample under extreme operating conditions. The first, second and third samples are fused to obtain the initial sample. Geometric and mechanical corrections are performed on the initial samples to obtain standard samples; The sensitivity coefficient of the target structural parameter is calculated by global screening method. Local structural parameters corresponding to the sensitivity coefficients that are greater than the preset sensitivity threshold are selected. The local structural parameters are perturbed slightly by local parameter perturbation method to obtain the performance parameter change characteristics. Sensitive structural parameters whose change rate of performance parameter change characteristics is greater than the preset change rate are selected. Using the sensitive structural parameters as coordinate axes, a local response surface model is constructed using a preset method, and the curvature value of the local response surface model is determined. Based on the curvature value, the sensitive structural parameters are divided into high-sensitivity structural parameters, low-sensitivity structural parameters, and transitionally sensitive structural parameters. The highly sensitive structural parameters are sampled at a high sampling density to obtain the first supplementary sample, and the performance response critical value of the highly sensitive structural parameters is obtained. Supplementary samples are generated based on the performance response critical value and added to the first supplementary sample. The sampling density of the transition-sensitive structural parameters is determined by linear gradient adjustment to obtain a second supplementary sample. Clustering is performed on low-sensitivity structural parameters to obtain several clusters. A representative sample from each cluster is retained as a third supplementary sample. The standard sample is supplemented based on the first, second, and third supplementary samples to obtain a combination of sample points.

[0011] Preferably, the key structural parameters are classified and decoupled to obtain the target structural parameters, including: The key structural parameters are classified according to their parameter types, resulting in multiple parameter type groups; By using correlation analysis to screen strongly coupled parameters from multiple parameter groups, and then using principal component analysis to reduce the dimensionality of the coupled parameters, the target structural parameters are obtained.

[0012] Preferably, a sample set is constructed based on the combination of sample points and performance response parameters, and the neural network model is trained to finally obtain the AI ​​agent model, including: The sample point combinations are cleaned and standardized, and weighted based on sensitivity characteristics to obtain weighted sample combinations; The initial surrogate model is obtained by training the neural network model based on weighted sample combinations; The initial agent model was validated and optimized based on the test set to obtain the AI ​​agent model.

[0013] Preferably, in step S4, based on the AI ​​agent model and combined with a multi-objective optimization algorithm, the key structural parameters are globally optimized to obtain the optimal parameter combination, including: Based on the multi-objective optimization objective and combined with parameter constraints, a multi-objective optimization algorithm is constructed. By inputting a multi-objective optimization algorithm into the AI ​​agent model, the algorithm performs global optimization on key structural parameters. The AI ​​agent model then predicts the performance response of each set of parameters in real time, ultimately obtaining the optimal parameter combination.

[0014] Preferably, the initial sample is subjected to geometric and mechanical corrections to obtain a standard sample, including: Substitute the initial sample into the parametric geometric model to generate a three-dimensional model. Perform collision simulation on the three-dimensional model to determine whether there is structural interference. If so, use the gradient adjustment method to correct the sample with interference. Based on a simplified finite element model, the initial sample is quickly stressed. Based on the calculation results, abnormal samples with stress values ​​greater than the stress threshold are selected and locally resampled. The samples after geometric and mechanical corrections to the initial samples are used as standard samples.

[0015] Compared with the prior art, the present invention has achieved the following beneficial effects: Design is based on the actual working conditions, loads, and spatial constraints of wind turbine components to ensure that the initial topology conforms to the engineering application scenario and avoids ineffective designs that are divorced from reality. The variable density method is used to solve for the optimal material distribution field, which can accurately form an efficient force transmission path, allowing the material to be concentrated in key stress areas and reasonably reduced in non-critical areas. This locks in the lightweight potential at the macro level and provides a high-quality basic configuration for subsequent optimization. By performing engineering geometric reconstruction on the initial topology, redundant structures are eliminated and core force transmission characteristics are retained, making the topology feasible for actual processing and optimization. Key structural parameters such as stiffener thickness and chamfer radius are extracted and a parametric model is established, transforming the macroscopic topology into a measurable model. The system utilizes customizable and controllable design variables to break down the traditional barriers between topology and parameter optimization, enabling coordinated optimization of both. By establishing a mapping relationship between key structural parameters and core performance responses such as mass, stress, and natural frequency, performance evaluation is transformed from time-consuming high-fidelity simulation calculations into rapid predictions by AI models. The time for a single performance evaluation is reduced from tens of hours to minutes. Based on the AI ​​proxy model and combined with multi-objective optimization algorithms, the system performs global optimization of key structural parameters to obtain the optimal parameter combination. Relying on the efficient evaluation capabilities of the AI ​​proxy model, the speed of optimization iteration is significantly improved, allowing global optimization to be completed in a short time. This ensures the reliability of the optimization effect and further compresses the R&D cycle.

[0016] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the construction of an AI agent model in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the process of obtaining the optimal parameter combination in an embodiment of the present invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] Example 1: This embodiment of the invention provides a method for collaborative optimization of topology and parameters of wind turbine components that integrates artificial intelligence, such as... Figure 1 As shown, it includes: S1: Based on the operating conditions, loads and spatial constraints of wind turbine components, the optimal distribution field of wind turbine components is solved using a topology optimization algorithm to obtain the initial topology configuration; S2: Perform geometric reconstruction on the initial topological configuration, extract key structural parameters, and establish a parametric finite element model; S3: Construct an AI proxy model between the key structural parameters and the performance response of wind turbine components; S4: Based on the AI ​​agent model, combined with a multi-objective optimization algorithm, the key structural parameters are globally optimized to obtain the optimal parameter combination.

[0021] In this embodiment, the topology optimization algorithm is the variable density method and the level set method.

[0022] In this embodiment, key structural parameters are globally optimized to achieve the goals of weight reduction and structural performance meeting design requirements.

[0023] In this embodiment, the key structural parameters include, for example, the thickness of the stiffener plate, the chamfer radius, and the shape of the doorway.

[0024] In this embodiment, the performance response of wind turbine components includes, for example, performance parameters such as mass, stress, and natural frequency.

[0025] The beneficial effects of the above design scheme are as follows: Based on the actual working conditions, loads, and spatial constraints of wind turbine components, the design ensures that the initial topology conforms to the engineering application scenario, avoiding ineffective designs that are divorced from reality. The use of the variable density method to solve for the optimal material distribution field accurately forms an efficient force transmission path, allowing materials to be concentrated in key stress areas and reasonably reduced in non-critical areas, thus locking in lightweight potential at a macroscopic level and providing a high-quality basic configuration for subsequent optimization. Through engineering geometric reconstruction of the initial topology, redundant structures are eliminated while core force transmission characteristics are retained, making the topology feasible for actual processing and optimization. Key structural parameters such as stiffener thickness and chamfer radius are extracted and a parametric model is established, transforming the macroscopic topology... The configuration is transformed into a quantifiable and controllable design variable, breaking down the traditional barriers between topology and parameter optimization and enabling their coordinated optimization. By establishing a mapping relationship between key structural parameters and core performance responses such as mass, stress, and natural frequency, performance evaluation is transformed from time-consuming calculations in high-fidelity simulations to rapid predictions by AI models. The time for a single performance evaluation is shortened from tens of hours to minutes. Based on the AI ​​proxy model and combined with multi-objective optimization algorithms, the key structural parameters are globally optimized to obtain the optimal parameter combination. Relying on the efficient evaluation capabilities of the AI ​​proxy model, the speed of optimization iteration is greatly improved, allowing global optimization to be completed in a short time. This ensures the reliability of the optimization effect and further compresses the R&D cycle.

[0026] Example 2: Based on Example 1, this embodiment of the invention provides a topology-parameter collaborative optimization method for wind turbine components that integrates artificial intelligence. In step S1, based on the operating conditions, loads, and spatial constraints of the wind turbine components, a topology optimization algorithm is used to solve for the optimal distribution field of the wind turbine components to obtain the initial topology configuration, including: The historical actual operating data of wind turbine components within a preset time period is obtained. Typical operating conditions are extracted from the historical actual operating data, and the load probability distribution under each typical operating condition is determined. The spatial constraints are obtained by hierarchical coding with the installation boundary as the hard constraint and the hoisting space, processing space and operation and maintenance space as the soft constraints. Temperature and vibration impact data are obtained from historical operational data, and temperature and vibration field constraints are established. These constraints are then coupled and transformed into equivalent mechanical constraints. Based on the load probability distribution, spatial constraints, and equivalent mechanical constraints, the overall constraint rules are established, and the optimization objectives are established with minimizing material volume as the main objective and minimizing maximum stress and minimizing process complexity as auxiliary objectives. Based on the overall constraint rules and optimization objectives, the first distribution field of wind turbine components is obtained by iteratively solving the variable density method according to a preset number of iterations. The second distribution field is obtained by iteratively solving the first distribution field according to a preset number of iterations using the level set method. The second distribution field is then fed back to the variable density method for bidirectional feedback iteration, and finally the optimal distribution field is obtained. The initial topological configuration is obtained by constructing the optimal distribution field.

[0027] In this embodiment, typical operating conditions include rated wind speed, extreme wind speed, start-up, shutdown, and maintenance.

[0028] In this embodiment, determining the load probability distribution under each typical working condition specifically involves generating the load probability distribution under each working condition using Monte Carlo simulation, transforming a single load value into a load range and weighting coefficient, and allowing topology optimization to adapt to real dynamic stress scenarios.

[0029] In this embodiment, the installation boundary is such as the connection interface between the hub and the main shaft and the blade; the hoisting space is such that when the tower is segmented, hoisting point space needs to be reserved to avoid the inability to hoist after optimization; the processing space is such that welding operation space and processing tool avoidance space need to be reserved; the maintenance space is such that when the engine room is internal, maintenance passage needs to be reserved.

[0030] In this embodiment, the temperature impact data is, for example, the temperature impact of the gearbox heat dissipation in the nacelle on the surrounding structure, and the vibration impact data is, for example, the resonance between the topology configuration and the blade rotation frequency.

[0031] In this embodiment, the preset number of iterations is, for example, 5.

[0032] In this embodiment, the variable density method quickly solves the approximate region of the optimal material distribution, and the level set method smooths the distribution boundary, solving the problem of jagged and difficult-to-process boundary in the variable density method.

[0033] The beneficial effects of the above design scheme are as follows: By extracting typical operating conditions and load probability distributions from historical operating data, optimization is no longer based on ideal operating conditions, but on probabilistic operating conditions under actual service environments. By using installation boundaries as hard constraints and hoisting space, processing space, and maintenance space as soft constraints, layered coding is performed to obtain spatial constraints, solving the feasibility problem caused by a one-size-fits-all approach to constraints. Core constraints are prioritized, while secondary constraints are handled flexibly, reducing iteration conflicts. Temperature and vibration impact data are obtained from historical actual operating data to establish temperature and vibration field constraints. These constraints are then coupled and transformed into equivalent mechanical constraints, considering the risks of thermal expansion and contraction deformation and resonance failure in actual operation. This avoids the failure problems of traditional designs under extreme temperature or vibration conditions, enabling the design to simultaneously meet strength, thermal stability, and vibration resistance, adapting to the complex service environment of wind turbine units. Overall constraint rules are established based on load probability distribution, spatial constraints, and equivalent mechanical constraints, with the primary objective of minimizing material volume, minimizing maximum stress, and minimizing process complexity. Minimization is set as an auxiliary objective, and an optimization objective is established to ensure that the strength and stiffness are not reduced while improving manufacturability. Based on the overall constraint rules and optimization objective, the first distribution field of wind turbine components is obtained by iteratively solving the variable density method for a preset number of iterations. The second distribution field is obtained by iteratively solving the first distribution field for a preset number of iterations using the level set method. The second distribution field is fed back to the variable density method for bidirectional feedback iteration, and finally the optimal distribution field is obtained. The initial topology configuration is constructed based on the optimal distribution field. The variable density method solves the problem of the difficulty in determining the initial interface of the level set method. The level set method improves the boundary roughness problem of the variable density method. The generated topology configuration has smooth and continuous boundaries, which is more in line with the processing requirements. The bidirectional feedback mechanism enables the topology configuration to simultaneously optimize material distribution and boundary morphology in each iteration, forming a virtuous cycle from material distribution to boundary optimization to performance improvement and then to optimization again, which provides structural stiffness and makes the stress distribution more uniform. Finally, through complete constraint modeling, multi-objective optimization and bidirectional feedback, it is ensured that the global optimum rather than the local optimum solution is obtained, so that the components perform well under all working conditions.

[0034] Example 3: Based on Example 2, this embodiment of the invention provides a topology-parameter collaborative optimization method for wind turbine components that integrates artificial intelligence. Specifically, it iterates the distribution field of wind turbine components using the variable density method and the level set method: Using the distribution field as the coarse grid specification, the first stage of coarse iteration is performed using the variable density method and the level set method to obtain the coarse distribution field. The core stress region and non-core region are then extracted from the coarse distribution field. Using the core stress region as the fine grid and the non-core region as the medium grid, the coarse distribution field is subjected to a second-stage fine iteration using the variable density method and the level set method to obtain the fine distribution field. After obtaining the fine distribution field, iterative verification and adjustment are performed based on the overall constraint rules to finally obtain the optimal distribution field.

[0035] In this embodiment, iterative verification and adjustment are performed based on the overall constraint rules. For example, spatial conflict judgment is performed. If a spatial conflict occurs, such as the optimized structure encroaching on the operation and maintenance channel, the penalty factor of the soft constraint is automatically adjusted and the iteration is restarted. Performance judgment is performed. If the performance does not meet the standard, such as the strain energy exceeding the threshold, the strength target weight is automatically increased and the material distribution is optimized in reverse.

[0036] The beneficial effects of the above design scheme are as follows: By using the distribution field as the coarse grid specification and employing the variable density method and level set method for the first stage of coarse iteration, a coarse distribution field is obtained, from which the core stress region and non-core region are extracted; using the core stress region as the fine grid and the non-core region as the medium grid, the coarse distribution field is subjected to the second stage of fine iteration using the variable density method and level set method to obtain the fine distribution field; after obtaining the fine distribution field, iterative verification and adjustment are performed based on the overall constraint rules to finally obtain the optimal distribution field; the coarse grid coarse iteration quickly locks the core and non-core regions, significantly reducing the initial computational load and shortening the overall iteration cycle; by differentiating the fine grid in the core region and the medium grid in the non-core region, the calculation accuracy of key stress parts is ensured while avoiding computational waste, balancing accuracy and efficiency; and during the iteration, the overall constraints are combined for real-time verification and adjustment, timely correction of the distribution field that deviates from the constraints, ensuring that the optimal distribution field meets both mechanical requirements and engineering constraints, thus improving the reliability of the results.

[0037] Example 4: Based on Example 1, this embodiment of the invention provides a topology-parameter collaborative optimization method for wind turbine components that integrates artificial intelligence. In step S2, the initial topology configuration is geometrically reconstructed, key structural parameters are extracted, and a parameterized finite element model is established, including: The core stress region, non-core stress region, and transition region of the initial topological configuration are obtained. The core stress region is backfitted to generate a continuous solid structure. The non-core stress region is simplified and reconstructed into a regular geometry. The transition region is smoothly connected between the core stress region and the non-core stress region through gradual wall thickness and rounded corner transition, thus obtaining the geometric reconstruction features. The geometric reconstruction features are filtered according to preset extraction criteria to obtain key structural parameters; Geometric expressions for key structural parameters are established, wind turbine components are modularized to obtain component modules, a mapping relationship between geometric expressions and component modules is established, and key structural parameters are simulated based on the mapping relationship to obtain a parametric finite element model.

[0038] In this embodiment, the core stress-bearing region is inversely fitted to generate a continuous solid structure. Specifically, NURBS surface inverse fitting is used, with topological density contour lines as boundaries, to generate a smooth and continuous solid structure, preserving the curvature characteristics of the stress-bearing surface and avoiding stress concentration.

[0039] In this embodiment, the transition region is smoothly connected to the core stress area and the non-core stress area by gradually changing the wall thickness and rounding the corners, eliminating the risk of mechanical failure caused by abrupt interface changes.

[0040] In this embodiment, the preset extraction criteria are to extract parameters that have a significant impact on performance, are processable and controllable, and are strongly related to lightweighting.

[0041] In this embodiment, key structural parameters include, for example, dimensional parameters such as stiffener thickness, flange thickness, hole diameter, chamfer radius, and honeycomb hole spacing; shape parameters such as the radius of the door opening arc, stiffener inclination angle, and mounting base taper; and distribution parameters such as hole array density and truss rod diameter gradient.

[0042] In this embodiment, the geometric expression is, for example, chamfer radius = stiffener thickness × 0.5.

[0043] In this embodiment, the component module, for example, the cabin includes a main body, stiffening plates, and interface modules.

[0044] The beneficial effects of the above design scheme are: by performing engineering geometric reconstruction on the initial topology, redundant structures are eliminated and core force transmission characteristics are retained, making the topology result feasible for actual processing and optimization. Key structural parameters such as stiffener thickness and chamfer radius are extracted and a parametric model is established, transforming the macroscopic topology into quantifiable and controllable design variables, breaking down the traditional barrier between topology and parameter optimization, and achieving the linkage optimization of the two.

[0045] Example 5: Based on Example 1, this embodiment of the invention provides a wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence, such as... Figure 2 As shown, in step S3, constructing an AI proxy model between the key structural parameters and the performance response of wind turbine components includes: Generate a combination of sample points within the design space of the key structural parameters; The performance response parameters corresponding to the sample point combination are obtained by simulating the sample point combination using a parametric finite element model. A sample set is constructed based on the combination of sample points and performance response parameters. The neural network model is then trained to obtain the final AI agent model.

[0046] In this embodiment, the neural network model is a deep neural network regression model.

[0047] The beneficial effects of the above design scheme are as follows: generating sample point combinations within the design space ensures comprehensive sample coverage and reasonable distribution, providing an unbiased data foundation for model training and guaranteeing the generalization ability of the surrogate model; using parametric finite element model simulation to obtain performance response ensures that the sample set data accurately matches the actual engineering situation, providing high-quality labeled data for model training; using deep neural network regression model training accurately fits the nonlinear mapping relationship between parameters and performance, ensuring the prediction accuracy of the surrogate model; at the same time, the AI ​​surrogate model can replace time-consuming simulation, significantly improving the efficiency of subsequent performance evaluation.

[0048] Example 6: Based on Example 5, a wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence is used to generate sample point combinations within the design space of the key structural parameters, including: The key structural parameters are classified and decoupled to obtain the target structural parameters; The target structural parameters are sampled first using the Latin hypercube sampling method to obtain the first sample. High target parameters with high impact on performance are screened out through sensitivity pre-analysis. Orthogonal arrays are designed for the high target parameters to perform the second sampling to obtain the second sample. Extreme operating conditions in the historical operating data of wind turbines are obtained to generate the third sample under extreme operating conditions. The first, second and third samples are fused to obtain the initial sample. Geometric and mechanical corrections are performed on the initial samples to obtain standard samples; The sensitivity coefficient of the target structural parameter is calculated by global screening method. Local structural parameters corresponding to the sensitivity coefficients that are greater than the preset sensitivity threshold are selected. The local structural parameters are perturbed slightly by local parameter perturbation method to obtain the performance parameter change characteristics. Sensitive structural parameters whose change rate of performance parameter change characteristics is greater than the preset change rate are selected. Using the sensitive structural parameters as coordinate axes, a local response surface model is constructed using a preset method, and the curvature value of the local response surface model is determined. Based on the curvature value, the sensitive structural parameters are divided into high-sensitivity structural parameters, low-sensitivity structural parameters, and transitionally sensitive structural parameters. The highly sensitive structural parameters are sampled at a high sampling density to obtain the first supplementary sample, and the performance response critical value of the highly sensitive structural parameters is obtained. Supplementary samples are generated based on the performance response critical value and added to the first supplementary sample. The sampling density of the transition-sensitive structural parameters is determined by linear gradient adjustment to obtain a second supplementary sample. Clustering is performed on low-sensitivity structural parameters to obtain several clusters. A representative sample from each cluster is retained as a third supplementary sample. The standard sample is supplemented based on the first, second, and third supplementary samples to obtain a combination of sample points.

[0049] In this embodiment, representative samples are, for example, samples whose performance response is at the median value within the cluster.

[0050] In this embodiment, the local response surface model is constructed using a preset method, specifically by using a quadratic polynomial and Kriging interpolation.

[0051] In this embodiment, the sampling density determination method using linear gradient adjustment is adopted for the transitionally sensitive structural parameters. Specifically, the sampling density is high in regions with high sensitivity structural parameters and low in regions with low sensitivity structural parameters.

[0052] In this embodiment, sensitive structural parameters are classified into high-sensitivity, low-sensitivity, and moderately sensitive structural parameters based on curvature values. Specifically, curvature ≥ 0.8: small changes in parameters cause drastic performance fluctuations, such as stress soaring from 300MPa to 340MPa in the rib thickness range of 12-13mm, which is defined as a high-sensitivity structural parameter; 0.5 ≤ curvature < 0.8: moderate performance fluctuations, defined as a moderately sensitive structural parameter; curvature < 0.5: gradual performance changes, defined as a low-sensitivity structural parameter.

[0053] In this embodiment, the global filtering method is the global Morris filtering method.

[0054] In this embodiment, the K-means clustering algorithm is used to cluster the low-sensitivity structural parameters.

[0055] The beneficial effects of the above design scheme are as follows: By classifying and decoupling key structural parameters, logical contradictions in parameter combinations are eliminated, ensuring the rationality of the sampling samples. The target structural parameters are sampled using a Latin hypercube sampling method to obtain the first sample. High-impact parameters with significant performance impact are selected through sensitivity pre-analysis. An orthogonal array is designed for these high-impact parameters for the second sampling, yielding the second sample. Extreme operating conditions from historical wind turbine data are obtained to generate a third sample under these extreme conditions. The first, second, and third samples are then fused to obtain the initial sample. The first sample ensures uniform global coverage of the target structural parameters, avoiding omissions in local areas. The second sample specifically covers all level combinations of high-impact parameters, ensuring the model fully learns the effects of key parameters. The third sample completes the parameter combinations under extreme scenarios, preventing model failure in extreme condition predictions. The initial sample formed by the fusion of these three samples takes into account conventional operating conditions, key parameters, and extreme scenarios, with a comprehensive coverage far exceeding that of a single sampling method. Geometric and mechanical corrections are applied to the initial sample to obtain standard samples, laying a high-quality foundation for subsequent sensitive area densification and sample optimization. A global screening method is used to refine the sample. This method accurately identifies highly sensitive parameters, pinpointing those that drive performance response. The local parameter perturbation method verifies the rate of performance change through small perturbations, further filtering out sensitive parameters where fine-tuning triggers significant performance fluctuations. This avoids misjudgment of sensitive areas, precisely identifying sensitive parameters and providing clear targets for subsequent differentiated sampling. It avoids sample redundancy caused by blind encryption. By classifying sensitivity levels according to curvature values, the sampling priority of different parameters is clarified, making the sampling strategy more targeted. High-density sampling of highly sensitive structural parameters and supplementing performance response critical value samples strengthen the support of the sample set for the core optimization objective, providing reliable data for subsequent multi-objective optimization. Linear gradient sampling of transitionally sensitive structural parameters ensures that the mapping pattern in the transition region is fully learned without wasting computational resources in non-critical areas, balancing model generalization ability and sampling efficiency. Cluster analysis removes redundant samples in low-sensitivity areas, retaining representative samples to ensure that the mapping pattern in low-sensitivity areas is not lost, while significantly saving computational costs for subsequent simulation and model training. Finally, a combination of supplementary sample points is obtained, satisfying the learning needs of deep neural network regression models for nonlinear mappings while controlling the total number of samples, balancing model accuracy and optimization efficiency.

[0056] Example 7: Based on Example 6, this embodiment of the invention provides a wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence. It classifies and decouples key structural parameters to obtain target structural parameters, including: The key structural parameters are classified according to their parameter types, resulting in multiple parameter type groups; By using correlation analysis to screen strongly coupled parameters from multiple parameter groups, and then using principal component analysis to reduce the dimensionality of the coupled parameters, the target structural parameters are obtained.

[0057] In this embodiment, the correlation analysis method is, for example, based on the Pearson coefficient.

[0058] In this embodiment, the dimensionality reduction of the coupling parameters using principal component analysis specifically involves extracting principal components as virtual sampling variables, and then restoring them to the original parameter combination through inverse transformation after sampling.

[0059] The beneficial effects of the above design scheme are as follows: by classifying key structural parameters according to parameter type, multiple parameter groups of different types are obtained; by using correlation analysis to screen strongly coupled parameters in multiple parameter groups of different types, and by using principal component analysis to reduce the dimensionality of the coupled parameters, the target structural parameters are obtained. This not only preserves the parameter correlation characteristics but also avoids logical contradictions during sampling, thus providing a basis for the extraction of sample groups.

[0060] Example 8: Based on Example 5, this embodiment of the invention provides a topology-parameter collaborative optimization method for wind turbine components that integrates artificial intelligence. A sample set is constructed based on sample point combinations and performance response parameters, and a neural network model is trained to ultimately obtain an AI proxy model, including: The sample point combinations are cleaned and standardized, and weighted based on sensitivity characteristics to obtain weighted sample combinations; The initial surrogate model is obtained by training the neural network model based on weighted sample combinations; The initial agent model was validated and optimized based on the test set to obtain the AI ​​agent model.

[0061] In this embodiment, the test set is extracted based on combinations of sample points.

[0062] In this embodiment, the initial agent model is validated and optimized based on the test set by using Dropout regularization to avoid overfitting, adjusting parameters such as the learning rate and the number of hidden layer neurons to optimize the model, and finally obtaining a high-precision, fast-responding AI agent model.

[0063] The beneficial effects of the above design scheme are: using a deep neural network regression model for training, accurately fitting the nonlinear mapping relationship between parameters and performance, and ensuring the prediction accuracy of the surrogate model; at the same time, the AI ​​surrogate model can replace time-consuming simulation, greatly improving the efficiency of subsequent performance evaluation.

[0064] Example 9: Based on Example 1, this embodiment of the invention provides a wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence, such as... Figure 3As shown, in step S4, based on the AI ​​agent model and combined with a multi-objective optimization algorithm, the key structural parameters are globally optimized to obtain the optimal parameter combination, including: Based on the multi-objective optimization objective and combined with parameter constraints, a multi-objective optimization algorithm is constructed. By inputting a multi-objective optimization algorithm into the AI ​​agent model, the algorithm performs global optimization on key structural parameters. The AI ​​agent model then predicts the performance response of each set of parameters in real time, ultimately obtaining the optimal parameter combination.

[0065] The beneficial effects of the above design scheme are: by using the AI ​​agent model and combining it with a multi-objective optimization algorithm, the key structural parameters are globally optimized to obtain the optimal parameter combination. Relying on the efficient evaluation capability of the AI ​​agent model, the optimization iteration speed is greatly improved, allowing the global optimization to be completed in a short time, which not only ensures the reliability of the optimization effect, but also further compresses the R&D cycle.

[0066] Example 10: Based on Example 6, this embodiment of the invention provides a topology-parameter collaborative optimization method for wind turbine components that integrates artificial intelligence. It performs geometric and mechanical corrections on initial samples to obtain standard samples, including: Substitute the initial sample into the parametric geometric model to generate a three-dimensional model. Perform collision simulation on the three-dimensional model to determine whether there is structural interference. If so, use the gradient adjustment method to correct the sample with interference. Based on a simplified finite element model, the initial sample is quickly stressed. Based on the calculation results, abnormal samples with stress values ​​greater than the stress threshold are selected and locally resampled. The samples after geometric and mechanical corrections to the initial samples are used as standard samples.

[0067] In this embodiment, abnormal samples are locally resampled, for example, five new samples are generated within the neighborhood of the sample point and then re-verified.

[0068] In this embodiment, the 3D model is generated automatically using the UG / SolidWorks API interface.

[0069] The beneficial effects of the above design scheme are: by performing geometric and mechanical corrections on the initial samples, standard samples are obtained, laying a high-quality foundation for subsequent sensitive area densification and sample optimization.

[0070] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this application and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for collaborative optimization of topology and parameters of wind turbine components integrating artificial intelligence, characterized in that, include: S1: Based on the operating conditions, loads and spatial constraints of wind turbine components, the optimal distribution field of wind turbine components is solved using a topology optimization algorithm to obtain the initial topology configuration; S2: Perform geometric reconstruction on the initial topological configuration, extract key structural parameters, and establish a parametric finite element model; S3: Construct an AI proxy model between the key structural parameters and the performance response of wind turbine components; S4: Based on the AI ​​agent model, combined with a multi-objective optimization algorithm, the key structural parameters are globally optimized to obtain the optimal parameter combination.

2. The wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence as described in claim 1, characterized in that, In step S1, based on the operating conditions, loads, and spatial constraints of the wind turbine components, a topology optimization algorithm is used to solve for the optimal distribution field of the wind turbine components, obtaining the initial topology configuration, including: The historical actual operating data of wind turbine components within a preset time period is obtained. Typical operating conditions are extracted from the historical actual operating data, and the load probability distribution under each typical operating condition is determined. The spatial constraints are obtained by hierarchical coding with the installation boundary as the hard constraint and the hoisting space, processing space and operation and maintenance space as the soft constraints. Temperature and vibration impact data are obtained from historical operational data, and temperature and vibration field constraints are established. These constraints are then coupled and transformed into equivalent mechanical constraints. Based on the load probability distribution, spatial constraints, and equivalent mechanical constraints, the overall constraint rules are established, and the optimization objectives are established with minimizing material volume as the main objective and minimizing maximum stress and minimizing process complexity as auxiliary objectives. Based on the overall constraint rules and optimization objectives, the first distribution field of wind turbine components is obtained by iteratively solving the variable density method according to a preset number of iterations. The second distribution field is obtained by iteratively solving the first distribution field according to a preset number of iterations using the level set method. The second distribution field is then fed back to the variable density method for bidirectional feedback iteration, and finally the optimal distribution field is obtained. The initial topological configuration is obtained by constructing the optimal distribution field.

3. The wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence according to claim 2, characterized in that, The iterative process of analyzing the distribution field of wind turbine components using the variable density method and the level set method is as follows: Using the distribution field as the coarse grid specification, the first stage of coarse iteration is performed using the variable density method and the level set method to obtain the coarse distribution field. The core stress region and non-core region are then extracted from the coarse distribution field. Using the core stress region as the fine grid and the non-core region as the medium grid, the coarse distribution field is subjected to a second-stage fine iteration using the variable density method and the level set method to obtain the fine distribution field. After obtaining the fine distribution field, iterative verification and adjustment are performed based on the overall constraint rules to finally obtain the optimal distribution field.

4. The wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence as described in claim 1, characterized in that, In step S2, the initial topological configuration is geometrically reconstructed, key structural parameters are extracted, and a parametric finite element model is established, including: The core stress region, non-core stress region, and transition region of the initial topological configuration are obtained. The core stress region is backfitted to generate a continuous solid structure. The non-core stress region is simplified and reconstructed into a regular geometry. The transition region is smoothly connected between the core stress region and the non-core stress region through gradual wall thickness and rounded corner transition, thus obtaining the geometric reconstruction features. The geometric reconstruction features are filtered according to preset extraction criteria to obtain key structural parameters; Geometric expressions for key structural parameters are established, wind turbine components are modularized to obtain component modules, a mapping relationship between geometric expressions and component modules is established, and key structural parameters are simulated based on the mapping relationship to obtain a parametric finite element model.

5. The wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence according to claim 1, characterized in that, In step S3, an AI proxy model is constructed between the key structural parameters and the performance response of wind turbine components, including: Generate a combination of sample points within the design space of the key structural parameters; The performance response parameters corresponding to the sample point combination are obtained by simulating the sample point combination using a parametric finite element model. A sample set is constructed based on the combination of sample points and performance response parameters. The neural network model is then trained to obtain the final AI agent model.

6. The wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence according to claim 5, characterized in that, Generate sample point combinations within the design space of the key structural parameters, including: The key structural parameters are classified and decoupled to obtain the target structural parameters; The target structural parameters are sampled first using the Latin hypercube sampling method to obtain the first sample. High target parameters with high impact on performance are screened out through sensitivity pre-analysis. Orthogonal arrays are designed for the high target parameters to perform the second sampling to obtain the second sample. Extreme operating conditions in the historical operating data of wind turbines are obtained to generate the third sample under extreme operating conditions. The first, second and third samples are fused to obtain the initial sample. Geometric and mechanical corrections are performed on the initial samples to obtain standard samples; The sensitivity coefficient of the target structural parameter is calculated by global screening method. Local structural parameters corresponding to the sensitivity coefficients that are greater than the preset sensitivity threshold are selected. The local structural parameters are perturbed slightly by local parameter perturbation method to obtain the performance parameter change characteristics. Sensitive structural parameters whose change rate of performance parameter change characteristics is greater than the preset change rate are selected. Using the sensitive structural parameters as coordinate axes, a local response surface model is constructed using a preset method, and the curvature value of the local response surface model is determined. Based on the curvature value, the sensitive structural parameters are divided into high-sensitivity structural parameters, low-sensitivity structural parameters, and transitionally sensitive structural parameters. The highly sensitive structural parameters are sampled at a high sampling density to obtain the first supplementary sample, and the performance response critical value of the highly sensitive structural parameters is obtained. Supplementary samples are generated based on the performance response critical value and added to the first supplementary sample. The sampling density of the transition-sensitive structural parameters is determined by linear gradient adjustment to obtain a second supplementary sample. Clustering is performed on low-sensitivity structural parameters to obtain several clusters. A representative sample from each cluster is retained as a third supplementary sample. The standard sample is supplemented based on the first, second, and third supplementary samples to obtain a combination of sample points.

7. The wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence according to claim 6, characterized in that, The key structural parameters are classified and decoupled to obtain the target structural parameters, including: The key structural parameters are classified according to their parameter types, resulting in multiple parameter type groups; By using correlation analysis to screen strongly coupled parameters from multiple parameter groups, and then using principal component analysis to reduce the dimensionality of the coupled parameters, the target structural parameters are obtained.

8. The wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence according to claim 5, characterized in that, A sample set is constructed based on the combination of sample points and performance response parameters. The neural network model is then trained to obtain the final AI agent model, which includes: The sample point combinations are cleaned and standardized, and weighted based on sensitivity characteristics to obtain weighted sample combinations; The initial surrogate model is obtained by training the neural network model based on weighted sample combinations; The initial agent model was validated and optimized based on the test set to obtain the AI ​​agent model.

9. The wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence according to claim 1, characterized in that, In step S4, based on the AI ​​agent model and combined with a multi-objective optimization algorithm, the key structural parameters are globally optimized to obtain the optimal parameter combination, including: Based on the multi-objective optimization objective and combined with parameter constraints, a multi-objective optimization algorithm is constructed. By inputting a multi-objective optimization algorithm into the AI ​​agent model, the algorithm performs global optimization on key structural parameters. The AI ​​agent model then predicts the performance response of each set of parameters in real time, ultimately obtaining the optimal parameter combination.

10. The wind turbine component topology-parameter collaborative optimization method integrating artificial intelligence according to claim 6, characterized in that, Geometric and mechanical corrections are performed on the initial samples to obtain standard samples, including: Substitute the initial sample into the parametric geometric model to generate a three-dimensional model. Perform collision simulation on the three-dimensional model to determine whether there is structural interference. If so, use the gradient adjustment method to correct the sample with interference. Based on a simplified finite element model, the initial sample is quickly stressed. Based on the calculation results, abnormal samples with stress values ​​greater than the stress threshold are selected and locally resampled. The samples after geometric and mechanical corrections to the initial samples are used as standard samples.

Images