Multi-parameter collaborative optimization method and system for savonius wind turbine

Abstract

The present application relates to the technical field of wind power generation equipment optimization design, and discloses a Savonius fan multi-parameter collaborative optimization method and system, the method comprising: based on SolidWorks, a three-dimensional parametric model is established with the number of blades, the twist angle and the overlap ratio as design variables; through the MATLAB and Fluent joint platform, multi-working-condition CFD flow field numerical simulation is carried out; a multi-objective optimization model is constructed with the maximum wind energy utilization coefficient as the target and the structural strength and the starting wind speed as the constraints; a hybrid strategy of adaptive inertia weight particle swarm and genetic algorithm is adopted for solving; the optimal parameters are returned to the modeling and simulation link for closed-loop verification until convergence, and after optimization, the power coefficient of the three-blade helical twist angle wind wheel is improved by about 12% to 18%, and the starting wind speed is reduced to below 2.8 m / s.

Description

Technical Field

[0001] This invention relates to the field of wind power generation equipment optimization design technology, and in particular to the Savonius wind turbine multi-parameter collaborative optimization method and system. Background Technology

[0002] Vertical axis wind turbines have gained widespread attention in distributed wind energy utilization scenarios due to their advantages such as omnidirectional airflow acceptance, simple structure, and low noise. Savonius wind turbines, as a typical representative of drag-type vertical axis wind turbines, rely on the drag difference between the concave and convex surfaces of the blades to generate rotational torque, possessing the inherent advantage of self-starting under low wind speed conditions. Therefore, they are particularly suitable for locations with relatively scarce wind resources, such as urban building rooftops, remote islands, and agricultural and pastoral areas. However, the wind energy utilization coefficient of traditional Savonius wind turbines is typically only between 0.15 and 0.25, far lower than that of horizontal axis wind turbines and Darrieus-type vertical axis wind turbines. How to significantly improve their aerodynamic efficiency while maintaining self-starting characteristics has always been a core research topic in this field.

[0003] In recent years, researchers have conducted extensive work on the geometric parameters of Savonius wind turbines. For example, Chinese invention application CN114169088A discloses a wind turbine blade optimization design method and system based on the Wilson model and genetic algorithm. This scheme takes horizontal axis wind turbine blades as the object, calculates the rated wind speed and rotor diameter according to the wind field characteristic model, uses the Wilson model to iteratively solve the axial and tangential induction coefficients, and then uses a genetic algorithm to search and optimize the chord length of each blade element, thus achieving joint optimization of airfoil parameters. Although this scheme effectively improves the aerodynamic performance of horizontal axis wind turbine blades, its optimization process is essentially still a serial optimization using the output of the Wilson model as the input of the genetic algorithm, and does not achieve true multivariate collaborative search between different parameters. In addition, this scheme is aimed at the two-dimensional aerodynamic parameters of the airfoil section and cannot be directly transferred to optimization problems like Savonius wind turbines, which use three-dimensional overall structural parameters as the core design variables.

[0004] In the field of Savonius wind turbine parameter optimization, most existing technologies employ a single-factor controlled variable approach, i.e., fixing other parameters and adjusting only one design variable to examine its impact on the wind energy utilization coefficient. For example, some studies have examined the change in wind energy utilization coefficient when the torsion angle changes from 0° to 180° under the condition of a fixed number of blades (2) and an overlap ratio of 0, finding that the power coefficient can reach about 0.223 at a torsion angle of 45°. Other studies have analyzed the performance changes in the overlap ratio range from 0 to 0.20 under the condition of fixed torsion angle and number of blades, indicating that an overlap ratio of 12.76% can achieve a maximum power coefficient of about 0.29. However, there is a strong coupling relationship among the number of blades, torsion angle, and overlap ratio. Single-factor optimization cannot reveal the interaction effects between parameters, and the conclusions obtained are often limited to specific parameter combinations, making it difficult to find the globally optimal solution. Although some studies have introduced response surface methodology or surrogate models to assist in optimization, their optimization algorithms mostly use a single genetic algorithm or particle swarm optimization algorithm, which is prone to getting trapped in local extrema in high-dimensional multimodal objective function space, and lacks a closed-loop feedback verification mechanism to evaluate the deviation between surrogate model predictions and actual CFD simulations.

[0005] In summary, existing technologies for optimizing Savonius wind turbine parameters have the following shortcomings: First, they lack a framework for synchronous and collaborative optimization that uses the number of blades, torsion angle, and overlap ratio as joint decision variables; second, single optimization algorithms are insufficient in global optimization capabilities when facing a non-convex search space with multiple coupled parameters; and third, the optimization results lack closed-loop CFD verification, and the accuracy of model prediction cannot be guaranteed. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a Savonius multi-parameter collaborative optimization method for wind turbines. This method first establishes a three-dimensional parametric wind turbine model based on SolidWorks software, with the number of blades, blade torsion angle, and blade overlap ratio as adjustable design variables. Then, the generated wind turbine geometry is imported into a MATLAB and Fluent co-simulation platform. Unsteady-state CFD flow field numerical simulations are performed under multiple preset wind speed conditions to obtain the torque coefficient and wind energy utilization coefficient corresponding to each parameter combination. Based on this, a multi-objective collaborative optimization model is constructed with the objective function of maximizing the weighted comprehensive value of the wind energy utilization coefficient, and constraints that the equivalent stress of the blades does not exceed the allowable stress of the material and the starting wind speed does not exceed a preset threshold. At the algorithm level, a hybrid optimization strategy combining particle swarm optimization and genetic algorithms is adopted. By introducing an adaptive inertia weight that decreases nonlinearly with the number of iterations and a crossover probability that is adaptively adjusted with the population fitness variance, a dynamic balance between global exploration and local fine-grained search is achieved. Finally, the optimal combination of design variables obtained through optimization is fed back to the 3D modeling and CFD simulation stage for closed-loop verification. If the deviation between the predicted value and the verified value exceeds the convergence threshold, the optimization model is updated and re-optimized until convergence.

[0007] This method deeply couples five stages—3D parametric modeling, multi-condition CFD simulation, multi-objective optimization construction, adaptive hybrid optimization, and closed-loop feedback verification—to form a closed-loop collaborative architecture. This allows the three design variables—blade number, torsion angle, and overlap ratio—to be searched synchronously within a unified optimization framework, fully capturing the interaction effects and synergistic gains between parameters. This avoids the drawback of individual factor tuning, which may miss the global optimum. Experimental results show that the optimized three-bladed helical torsion angle wind turbine achieves a power coefficient increase of approximately 12% to 18% at rated wind speed compared to the traditional design, and reduces the start-up wind speed to below 2.8 m / s, significantly expanding its application range in low-wind-speed environments.

[0008] This invention also provides a Savonius wind turbine multi-parameter collaborative optimization system, including a 3D parametric modeling module, a multi-condition CFD simulation module, a multi-objective optimization construction module, an adaptive hybrid optimization module, and a closed-loop feedback verification module. The 3D parametric modeling module is configured to build a 3D parametric model based on SolidWorks with the number of blades, torsion angle, and overlap ratio as design variables, and generate the wind turbine geometry. The multi-condition CFD simulation module is configured to import the wind turbine geometry into the MATLAB and Fluent co-simulation platform for multi-condition unsteady-state numerical simulation to extract aerodynamic performance indicators. The multi-objective optimization construction module is configured to build a multi-objective model with the goal of maximizing wind energy utilization coefficient and with structural strength and start-up wind speed as constraints. The adaptive hybrid optimization module is configured to use a PSO-GA hybrid strategy and introduce adaptive inertia weights and cross-probability for solving the problem. The closed-loop feedback verification module is configured to feed back the optimal parameters to the modeling and simulation stages for verification, and then determine whether to update the model and iterate again based on the deviation. The data flow of each module is bidirectional, forming a closed loop. Attached Figure Description

[0009] Figure 1 This is a flowchart of the Savonius wind turbine multi-parameter collaborative optimization method provided in the embodiments of the present invention.

[0010] Figure 2 This is an architecture diagram of the Savonius wind turbine multi-parameter collaborative optimization system provided in an embodiment of the present invention. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0012] Reference Figure 1 As shown in the figure, this invention provides a multi-parameter collaborative optimization method for Savonius wind turbines. This method uses three key geometric parameters—blade number, blade torsion angle, and blade overlap ratio—as joint decision variables. Through deep coupling and closed-loop collaboration of five core steps—three-dimensional parametric modeling, multi-condition CFD flow field simulation, multi-objective optimization model construction, adaptive hybrid PSO-GA intelligent optimization, and closed-loop feedback verification—it achieves the globally optimal design of the Savonius wind turbine's aerodynamic performance. The steps are described in detail below.

[0013] Step S1: 3D parametric wind turbine modeling. The core objective of this step is to establish a complete Savonius 3D parametric geometric model of the wind turbine, so that any combination of design variables in the subsequent optimization process can automatically drive the generation of the corresponding wind turbine geometric entity, thereby providing geometric input for batch CFD simulation.

[0014] In one embodiment of the present invention, SolidWorks software is used as the 3D modeling platform to first determine the basic structural framework of the wind turbine. The Savonius wind turbine mainly consists of three parts: blades, a rotating shaft, and end plates, among which the geometry of the blades plays a decisive role in aerodynamic performance. This embodiment specifies the number of blades. Blade twist angle and blade overlap ratio These are defined parametrically as three core design variables. Specifically, the number of blades... This indicates the number of blades evenly distributed along the circumference of the wind turbine, and its value is an integer ranging from 2 to 5; the blade twist angle. Defined as the cumulative rotational offset angle of the blade along the rotor axis from the bottom section to the top section, i.e., the total angle of rotation layer by layer along the blade height direction, with a value ranging from 0° to 180°; blade overlap ratio. Defined as the overlap distance between adjacent blades in the radial direction of the wind turbine. With wind turbine outer diameter The ratio, that is The value ranges from 0 to 0.3.

[0015] Preferably, in the process of constructing the parametric model, the blade cross-sectional profile adopts a semi-circular arc shape with a radius of curvature. According to the outer diameter of the wind turbine and blade overlap ratio This is determined through the following relationship: ,in, The radius of curvature of the blade cross section is expressed in meters (m). The outer diameter of the wind turbine is in meters (m). In this embodiment, it is taken as... m; The overlap distance between adjacent blades in the radial direction, in meters (m), is a value that... ; The number of blades is a dimensionless integer. This formula ensures that the radius of curvature of the blade cross-section can be adaptively adjusted under different combinations of blade numbers and overlap ratios to maintain a reasonable blade spacing and airflow channel width. When the number of blades increases from 2 to 5, the radius of curvature decreases accordingly, and the circumferential arc occupied by each blade decreases, thus changing the ratio of the concave windward area to the convex leeward area, directly affecting the torque characteristics of the wind turbine.

[0016] Furthermore, for helical blades with non-zero torsion angles, their three-dimensional geometry is achieved using a cross-section sweeping method. Specifically, this involves setting equal intervals along the rotor's rotation axis... A cross-sectional layer, in this embodiment, is taken as... Each cross-section has a section shape that is the aforementioned semi-circular arc, but adjacent cross-sections are rotated by an incremental angle around the rotation axis. The incremental angle is: ,in, This represents the rotational increment angle between adjacent sections, in degrees. This represents the total blade twist angle, in degrees. The number of cross-sectional layers along the axial direction is a dimensionless integer. By sequentially connecting the rotated cross-sectional profiles of each layer using SolidWorks' loft feature, a three-dimensional blade solid with helical torsion characteristics can be generated. Preferably, the sweep path uses equidistant helices instead of simple straight lines to ensure a smooth transition of the blade surface in the axial direction and avoid stress concentration areas.

[0017] In one embodiment of the present invention, the parameters of the wind turbine end plate are also included in the modeling scope. End plate diameter The diameter is set to 1.05 to 1.15 times the outer diameter of the wind turbine; in this embodiment, it is preferably [missing information]. ,Right now m; end plate thickness The setting is 2mm to 5mm; in this embodiment, it is taken as... mm. The main function of the endplate is to suppress leakage of three-dimensional vortices at the blade tips and reduce aerodynamic losses caused by the end effect. Rotor height. Set to 0.3m, thus the height-to-diameter ratio of the wind turbine is... This value is based on the recommended range for the height-to-diameter ratio of the Savonius wind turbine in existing studies. The diameter of the rotating shaft is set at 10mm, and the material is selected from 6061-T6 aluminum alloy.

[0018] To achieve automated batch modeling, this embodiment developed a parameter-driven script based on the SolidWorks API. This script uses the number of blades as a parameter. , torsion angle and overlap ratio As input parameters, the script automatically invokes SolidWorks' parametric dimension modification function to update all associated dimensions, including blade section curvature radius, adjacent section rotation increment angle, and blade spacing. It also automatically performs lofting, arraying, and Boolean operations, ultimately outputting a STEP format wind turbine geometry solid file. During the optimization process, for each new combination of design variables generated, the script can automatically update the corresponding 3D model and export the file within 30 to 60 seconds, significantly improving the efficiency of parameter space traversal.

[0019] It is worth noting that the parameter association design of the 3D parametric model is crucial to ensuring the robustness of automated modeling. In one embodiment of the present invention, when the number of blades... When changes occur, not only does the radius of curvature of the blade section need to be adjusted in conjunction with the aforementioned formula, but the angular spacing of the blades in the circumferential direction of the wind turbine also needs to be updated synchronously. For example, when At that time, the three blades are arranged at equal intervals of 120°; when At that time, the four blades are arranged at equal intervals of 90°. Furthermore, the overlap ratio... The change in overlap ratio directly affects the relative position of the blades in the radial direction. When the overlap ratio increases, the concave areas of adjacent blades form a larger overlap space, allowing more high-speed airflow flowing from the concave surface of the windward blade to be guided to the inner side of the concave surface of the leeward blade, thereby enhancing the propulsive torque of the leeward blade to a certain extent. However, when the overlap ratio exceeds a certain critical value, the excessively large overlap area will increase the backflow blockage effect of the airflow, leading to a decrease in the overall wind energy utilization coefficient. Therefore, there is a significant interactive coupling relationship between the overlap ratio and the number of blades, which is the fundamental technical motivation for this invention to adopt multi-parameter collaborative optimization rather than single-factor optimization.

[0020] Furthermore, twist the angle The impact on the three-dimensional morphology of the blade is more complex. When At this time, the blades degenerate into a traditional straight blade configuration, with the same azimuth angle at all height sections; when As the blades gradually increase in size, they exhibit a spiral upward shape. During the rotation of the wind turbine, the blade cross-sections at different heights sequentially enter the windward position, resulting in a more uniform torque output and reducing the torque fluctuation problem inherent in traditional Savonius wind turbines. In this embodiment, the preferred torsion angle search resolution is 5°, that is, 37 candidate torsion angle values ​​are generated in 5° increments within the range of 0° to 180°, in order to achieve a reasonable balance between search accuracy and computational cost.

[0021] Step S2: Multi-condition CFD flow field numerical simulation. The core objective of this step is to perform multi-condition flow field simulation on the wind turbine geometry generated in Step S1 using computational fluid dynamics methods, obtain aerodynamic performance data of various design variable combinations under different wind speed conditions, and provide quantitative evaluation basis for the construction of subsequent optimization models.

[0022] In one embodiment of the present invention, MATLAB is used as the main control platform, and the ANSYS Fluent solver is automatically invoked via scripts to perform CFD simulation calculations. MATLAB is responsible for configuring preprocessing parameters, scheduling batch processing of the Fluent solver, and automatically extracting and collecting post-processing data. Preferably, the entire simulation process is automated by using MATLAB's system command to call Fluent's Journal file interface, requiring no manual intervention.

[0023] Regarding the construction of the computational domain, this embodiment employs a dual-region sliding mesh strategy. A Cartesian coordinate system is established with the wind turbine's rotation center as the origin. The rotation domain is set as a coaxial cylinder with a diameter ranging from 1.5 to 2.0 times the outer diameter of the wind turbine; in this embodiment, it is preferably [missing information]. The height is 0.3m, the same as the wind turbine height. The external flow field is set as a cuboid region, and the length in the incoming flow direction is 20 times the outer diameter of the wind turbine. m, where the width is 15 times the outer diameter of the wind turbine. m, the height is taken as 5 times the outer diameter of the wind turbine. m. Data is transferred between the rotating domain and the external flow field through a sliding interface to simulate the actual rotational motion of the wind turbine.

[0024] For mesh generation, an unstructured tetrahedral mesh is used within the rotating domain, and 15 prism boundary layer meshes are set near the blade wall. The thickness of the first layer is set to 0.05 mm to ensure the wall surface... The value is controlled below 1 to meet the near-wall mesh accuracy requirements of the k-ω SST turbulence model. A structured hexahedral mesh is used in the external flow field, gradually thinning out from the rotating domain interface towards the far field to save computational resources. In this embodiment, the total mesh size of the rotating domain is approximately 1.2 million to 1.8 million elements, the external flow field mesh is approximately 800,000 to 1.2 million elements, and the total mesh size of the overall computational domain is approximately 2 million to 3 million elements. To ensure the mesh independence of the calculation results, independence verification is performed using three different mesh densities before the formal calculation: coarse mesh (approximately 1.5 million elements), medium mesh (approximately 2.5 million elements), and fine mesh (approximately 3.5 million elements). The wind energy utilization coefficient between adjacent mesh sets is... A deviation of less than 2% is considered to meet the computational accuracy requirements.

[0025] The k-ω SST model was selected as the turbulence model. This model automatically switches to the k-ω model in the near-wall region to capture the turbulent characteristics within the boundary layer, while switching to the k-ε model in the free flow region far from the wall to improve far-field prediction accuracy. It is particularly suitable for complex flow fields such as the Savonius wind turbine, which exhibit large-scale flow separation and vortex shedding. The pressure-velocity coupling was performed using the SIMPLE algorithm. The spatial discretization scheme used a second-order upwind scheme for both the momentum and turbulence equations, while the time discretization used a second-order implicit scheme.

[0026] In one embodiment of the present invention, the wind speed conditions covered by the simulation include three categories: the low wind speed condition has a wind speed range of 2 m / s to 4 m / s, and in this embodiment, the preferred wind speed is... The wind speed range under rated operating conditions is 6 m / s to 10 m / s. In this embodiment, the preferred wind speed is... m / s, the wind speed range for high-wind conditions is 12 m / s to 16 m / s, and in this embodiment, the preferred wind speed is... m / s. Tip speed ratio under each wind speed condition. The range is 0.2 to 1.4, with intervals of 0.2. This means that simulations need to be performed at 7 different tip speed ratio points for each wind speed condition. Each simulation must run for at least 10 rotations of the wind turbine to ensure the flow field reaches a periodic steady state. The time step is set to correspond to the time interval for each 1° rotation of the wind turbine, under rated wind speed conditions. m / s, tip speed ratio For example, the rotational angular velocity of the wind turbine It can be determined by the following relationship: ,in, The angular velocity of the wind turbine is expressed in rad / s. The tip speed ratio is dimensionless. The incoming air velocity is expressed in m / s. This refers to the outer diameter of the wind turbine, in meters (m). , m / s Substituting m into the equation yields... rad / s, then the time required to complete one revolution. s, corresponding to the time step of each 1° rotation. s.

[0027] After the simulation is completed, the torque coefficient for each time step is extracted from the Fluent solution. The wind energy utilization coefficient is calculated according to the following formula. : ,in, The wind energy utilization coefficient is a dimensionless coefficient that characterizes the efficiency of a wind turbine in converting incoming wind energy into rotational mechanical energy. The torque coefficient is defined as the aerodynamic torque acting on the wind turbine. With reference torque The ratio of, among which Take air density kg / m 3 , Wind turbine frontal area , The radius of the wind turbine; The tip speed ratio is taken as the time average of the torque coefficient during the last 5 revolutions after the periodic steady state. A representative value is used to eliminate the random effects of torque fluctuations within a single rotation cycle.

[0028] The MATLAB post-processing script automatically reads the simulation result files for each working condition and performs calculations. Characteristic curves and extraction of maximum wind energy utilization coefficient and its corresponding optimal tip speed ratio Simultaneously, the static moment coefficients of the wind turbine at various azimuth angles in a stationary state are extracted from the simulation results under low wind speed conditions. All azimuth angles The minimum wind speed corresponding to all values ​​being greater than zero is defined as the starting wind speed. The above data will be used as input for optimizing the model in step S3.

[0029] In one embodiment of the present invention, an adaptive mesh refinement strategy is introduced to improve the accuracy and reliability of CFD simulation results. Specifically, a complete unsteady-state simulation is first performed on the initial mesh to extract the pressure and velocity gradient distributions on the blade surface. Then, the mesh is automatically refined in local regions where the pressure or velocity gradient exceeds a preset threshold, with a refinement ratio of 0.5 times the original mesh size. Finally, the simulation is re-executed on the refined mesh. This adaptive refinement strategy effectively improves the simulation accuracy of key locations such as flow separation regions, vortex shedding regions, and blade leading-edge stagnation points without significantly increasing the overall mesh size.

[0030] Preferably, for each set of design variable combinations, in addition to extracting the moment coefficient and wind energy utilization coefficient, the simulation results also extract the pressure distribution cloud map and streamline distribution information of the blade surface. The pressure distribution data is used for the evaluation of structural strength constraints in subsequent step S3. Specifically, the aerodynamic pressure distribution on the blade surface is applied as a load boundary condition to the finite element structural model of the blade, and the stress distribution at each location on the blade is calculated through static analysis to determine the maximum equivalent stress value. The streamline distribution information is used to analyze the influence of different design variable combinations on the flow field structure around the wind turbine, such as the improvement effect of helical twisted blades on suppressing end vortex leakage compared to straight blades, and the influence of different overlap ratios on the airflow guidance efficiency between blades. These auxiliary analysis results help engineers to deeply understand the physical mechanism of the optimization scheme and provide a reference for subsequent engineering implementation.

[0031] Step S3: Construction of a multi-objective collaborative optimization model. The core objective of this step is to establish a multi-objective collaborative optimization mathematical model based on the multi-condition aerodynamic performance data obtained in step S2, with the maximization of wind energy utilization coefficient as the core objective and structural safety and start-up capability as constraints.

[0032] In one embodiment of the present invention, the objective function is designed to comprehensively consider wind energy capture performance under different wind speed conditions. Given the varying frequencies of different wind speed ranges in actual wind fields, this embodiment uses a weighted synthesis approach to define the objective function: ,in, The value of the comprehensive objective function is dimensionless. , , These are the maximum wind energy utilization coefficients under low wind speed, rated wind speed, and high wind speed conditions, respectively. , , The weighting coefficients for each working condition satisfy... In this embodiment, considering that the low wind speed range and the rated wind speed range are the main operating ranges of the Savonius wind turbine, the weights are set to... , , This weighted strategy prioritizes improving aerodynamic efficiency at rated wind speed while also considering low-wind-speed start-up capability and high-wind-speed stability.

[0033] The constraints include both structural strength constraints and start-up wind speed constraints. Regarding structural strength constraints, this refers to the equivalent stress borne by the blade root section when the wind turbine operates at its maximum design speed under high wind speed conditions. The allowable stress of the material shall not be exceeded. ,Right now: ,in, Von Mises equivalent stress at the blade root section, in MPa, can be obtained by combining the aerodynamic load obtained from CFD simulation with the finite element static analysis of the blade structure. This represents the allowable stress of the material, expressed in MPa. The yield strength of the blade material is specified. In this embodiment, the blade is made of 304 stainless steel, and its yield strength is specified. MPa; For safety factors, the value ranges from 1.5 to 2.5; in this embodiment, it is taken as... Corresponding allowable stress MPa. The starting wind speed constraint requires the wind turbine to be able to self-start under low wind speed conditions while stationary, i.e.: ,in, The starting wind speed of the wind turbine is expressed in m / s and is determined by the static moment coefficient analysis in step S2. To set the wind speed threshold for activation, this embodiment takes... m / s, to ensure that the wind turbine can start reliably in low wind speed environments.

[0034] Based on the above objective function and constraints, the multi-objective collaborative optimization model can be expressed as:

[0035] ,

[0036] ,

[0037] ,

[0038] ,

[0039] This model is a mixed-integer nonlinear optimization problem, where For integer variables, and The variables are continuous. Since both the objective function and constraint functions need to be evaluated through CFD simulation and do not have analytical forms, a gradient-free intelligent optimization algorithm is required to solve them.

[0040] In one embodiment of the present invention, the weight coefficients of each operating condition in the objective function are not arbitrarily set, but are calculated based on the Weibull wind speed probability distribution characteristics of the target wind field. Specifically, the annual average wind speed distribution data of the target installation site is first obtained, and its Weibull distribution parameters, i.e., shape parameters, are fitted. and scale parameters Then, the probability integral values ​​of the low wind speed segment, rated wind speed segment, and high wind speed segment under the Weibull distribution are calculated separately. These probability integral values ​​are normalized and used as the weight coefficients for the corresponding operating conditions. This weight determination method based on wind field statistical characteristics enables the optimization direction to automatically adapt to the wind resource conditions of different installation sites, avoiding subjective biases that may be introduced by manually setting weights.

[0041] Furthermore, this embodiment introduces a simplified finite element estimation method in handling structural strength constraints to avoid the enormous computational overhead of performing a complete structural finite element analysis for each fitness assessment. Specifically, based on a simplified cantilever beam mechanical model of the blade, the aerodynamic pressure on the blade surface obtained from CFD simulation is equivalent to a distributed load along the blade height direction. The maximum bending moment and equivalent stress at the blade root section are quickly estimated using analytical formulas. For helical blades, the contribution of the additional torque component introduced by the torsional angle to the stress at the root section is also considered. Comparison and verification of this simplified estimation method with the results of the complete finite element analysis show that the deviation of the equivalent stress estimation between the two methods does not exceed 10% within the range of design variable values, fully meeting the accuracy requirements for constraint screening during the optimization iteration process. Only in the final closed-loop verification stage is a complete three-dimensional finite element structural analysis performed on the optimal solution to confirm its structural safety.

[0042] Step S4: Adaptive Hybrid PSO-GA Intelligent Optimization. The core objective of this step is to design a hybrid optimization strategy that combines the advantages of particle swarm optimization and genetic algorithm to efficiently solve the multi-objective collaborative optimization model established in step S3.

[0043] In one embodiment of the present invention, the overall architecture of the hybrid optimization strategy adopts a two-layer collaborative mechanism of PSO-led and GA-assisted. Specifically, the outer loop is a particle swarm optimization (PSO) framework, responsible for the global position search and velocity update of the population; the inner layer embeds crossover and mutation operations of a genetic algorithm to enhance population diversity and help particles escape local extreme value regions. The fusion of the two algorithms is achieved as follows: after each round of particle swarm velocity-position update, the crossover and mutation operations of the genetic algorithm are performed on the bottom 30% of the particles in the current population, and the generated new individuals replace the original low-fitness particles, and then the next round of particle swarm iteration begins.

[0044] The velocity update formula in the particle swarm optimization algorithm is: ,in, For the first The particle in the first The velocity vector at the next iteration; For the first The particle in the first The velocity vector at the next iteration; For the first The particle in the first The position vector at the next iteration has components that correspond to... , , Three design variables; For the first The historical optimal position of each particle; The globally optimal position; and As the learning factor, in this embodiment, we take... ; and For interval Uniformly distributed random numbers within; For the first Inertia weights in the next iteration.

[0045] The key innovation of this invention lies in the inertial weight. An adaptive adjustment strategy is proposed. Traditional particle swarm optimization algorithms often employ linearly decreasing inertia weights. However, in high-dimensional multimodal search spaces, linear decreasing strategies can easily lead to insufficient early global search or slow convergence in later stages. This embodiment proposes a nonlinear adaptive inertia weight calculation method based on the cosine function: ,in, For the first The inertial weight in the next iteration is dimensionless. The initial maximum value of the inertia weight is taken in this embodiment. ; The minimum value for the inertial weight termination is taken in this embodiment. ; This represents the current iteration number; To determine the maximum number of iterations, in this embodiment, we take... Compared to the linear decreasing strategy, the cosine decreasing function maintains a higher inertia weight in the early stage of iteration to enhance global exploration capability, decreases rapidly in the middle stage to accelerate the aggregation to high-quality regions, and slowly approaches the termination value in the later stage to perform fine local search, thus achieving a better balance between global exploration and local development.

[0046] In the crossover operation of a genetic algorithm, the crossover probability A strategy of adaptive adjustment based on population fitness variance is adopted. Population fitness variance. It reflects the current level of population diversity and is defined as: ,in, The normalized variance of population fitness is dimensionless. For population size, this embodiment takes... ; For the first The fitness value of each particle; The average fitness of the population; As the normalization factor, take To avoid division by zero. When When the population diversity is high, the crossover probability is set to a low level. To maintain existing high-quality individuals; when When this occurs, it indicates that the population is tending to converge and may be trapped in a local extremum. In this case, the crossover probability is increased to [a certain value]. This increases population perturbation and helps particles escape local traps. In this embodiment, the variance threshold is set to... The mutation operation uses Gaussian mutation, with a mutation probability of... The value is fixed at 0.1, and the variation amplitude decreases linearly with the iteration process to maintain stability in the later stages.

[0047] Regarding the number of leaves This integer variable, the position vector, is involved in particle swarm optimization and genetic operations. The components remain continuous real numbers in the calculation, but when evaluating fitness, they are rounded to the nearest integer value, thus embedding integer constraints into the continuous optimization framework. Furthermore, to handle constraints, this embodiment employs an external penalty function method, multiplying the fitness value of solutions that do not meet the constraints by a penalty coefficient to attenuate the fitness. Pick to .

[0048] For population initialization, the Latin hypercube sampling method is used to generate populations in the three-dimensional design variable space. The initial population is selected to ensure a uniform distribution in the parameter space. The termination condition for the optimization process is reaching the maximum number of iterations. The improvement in the global optimal fitness value after 30 consecutive iterations is less than [a certain value]. .

[0049] In one embodiment of the invention, an elite retention strategy is introduced to prevent the loss of the optimal solution during optimization. Specifically, after each iteration, the top 10% of the particles in terms of fitness in the current population are directly copied to the next generation population, without participating in the crossover and mutation operations of the genetic algorithm, thus ensuring that the historical optimal solution is always preserved during population evolution. Simultaneously, to prevent the search area from narrowing due to excessive aggregation of elite individuals, the neighborhood range of elite particles is monitored. When the Euclidean distance between two elite particles in the design variable space is less than a preset minimum distance threshold, the particle with higher fitness is retained, and the other particle is randomly re-initialized to an unexplored region of the search space. This strategy ensures that the optimal solution is not lost while maintaining sufficient dispersion of the population in the design variable space.

[0050] Preferably, regarding the number of blades This discrete integer variable and continuous variable , Given the mixed characteristics, this embodiment also designs a differentiated mutation strategy. For integer variables... The mutation operation uses a random replacement method, that is, based on the mutation probability. Replace the current value with another random integer within the range of values; for continuous variables and The mutation operation employs a Gaussian perturbation method, which involves adding a Gaussian random increment with zero mean and a standard deviation that decreases linearly with the iteration process to the current value. This differentiated mutation strategy, designed for different variable types, effectively avoids problems such as invalid searches for integer variables or excessive jumps in continuous variables that may be caused by a uniform mutation operation, thus improving search efficiency in the mixed variable space.

[0051] Step S5: Closed-loop feedback verification and parameter backtracking correction. The core objective of this step is to establish a closed-loop feedback mechanism from optimization prediction to CFD verification and then to model update, to ensure that the optimal design parameter combination output is verified by actual flow field simulation, and to eliminate surrogate model bias or numerical accumulation error that may be introduced during the optimization process.

[0052] In one embodiment of the present invention, the optimal combination of design variables output in step S4 is... and its corresponding optimized predicted wind energy utilization coefficient First, the data is returned to step S1, where the corresponding wind turbine geometry is automatically generated by the 3D parametric modeling script. Then, this geometry enters the complete CFD simulation process in step S2, where independent verification simulation calculations are performed under three wind speed conditions to obtain the verification wind energy utilization coefficient. .

[0053] The core criterion for closed-loop validation is the relative deviation between the optimized predicted value and the CFD validation value. : ,in, The percentage is the relative deviation. The optimal integrated objective function value obtained from the hybrid PSO-GA optimization in step S4; This refers to the corresponding comprehensive objective function value obtained from the CFD verification simulation in step S2. If... ,in To preset the convergence threshold, this embodiment takes... If the result is satisfactory, then the optimization result is considered reliable. The optimal design scheme as the final output; if If the error is not clear, it indicates that there is a significant deviation in the optimization process, and the following backtracking correction process needs to be executed.

[0054] The specific operation of backtracking correction is as follows: The results obtained from this round of verification simulation... The corresponding design variable combinations are used as new training samples and added to the optimization model database in step S3 to update the response surface or interpolation model of the objective function. Then, the hybrid PSO-GA optimization process in step S4 is re-executed. Preferably, the global optimum position from the previous round can be used during the re-optimization. This serves as initial population guidance information for the next round, accelerating convergence. In this embodiment, the maximum number of iterations for closed-loop feedback is set. This is to prevent infinite loops caused by inherent biases in the model.

[0055] Through the aforementioned closed-loop feedback mechanism, step S5 forms a data loop with the output of step S4 and the verification from steps S1-S2, and propagates the verification deviation information back to the optimization model in step S3 for correction, enabling the entire optimization system to have self-correcting capabilities. In actual implementation, the relative deviation typically converges to within 5% after 2 to 3 rounds of closed-loop iterations, effectively ensuring the reliability of the final optimization scheme.

[0056] Reference Figure 2 As shown, this embodiment of the invention also provides a Savonius wind turbine multi-parameter collaborative optimization system. This system corresponds one-to-one with the five steps in the above method embodiment, including a three-dimensional parametric modeling module 1, a multi-condition CFD simulation module 2, a multi-objective optimization construction module 3, an adaptive hybrid optimization module 4, and a closed-loop feedback verification module 5. The modules achieve bidirectional data flow through standardized data interfaces, forming a complete closed-loop collaborative architecture from modeling to simulation, from optimization to verification, and from verification to model updating.

[0057] The 3D parametric modeling module 1 corresponds to step S1 in the method embodiment. This module is configured to build a 3D parametric geometric model of the Savonius wind turbine based on SolidWorks software and its API interface. This module receives a combination of design variables from the adaptive hybrid optimization module 4 or the closed-loop feedback verification module 5, including three parameters: the number of blades, the blade twist angle, and the blade overlap ratio. It automatically calls SolidWorks' parametric drive engine to update the geometric dimensions, performs lofting, arraying, and Boolean operations to generate the corresponding wind turbine geometric entity, and outputs it to the multi-condition CFD simulation module as a STEP format file. Preferably, this module has a built-in parameter validity verification subunit. When a new combination of design variables is received, it first checks whether each parameter is within a preset value range. For parameters that exceed the range, it automatically truncates them to the nearest boundary value to avoid modeling failure due to parameter anomalies. The module's single model update time is approximately 30 to 60 seconds, and it has the automation capability to support large-scale parameter space traversal.

[0058] The multi-condition CFD simulation module 2 corresponds to step S2 in the method embodiment. This module is configured to import the wind turbine geometry output by the 3D parametric modeling module 1 into the MATLAB and Fluent co-simulation platform, completing all CFD preprocessing and solution tasks, including mesh generation, boundary condition setting, turbulence model selection, and unsteady-state solution. This module contains three functional components: a mesh generation subunit, a solver scheduling subunit, and a post-processing data extraction subunit. The mesh generation subunit is responsible for meshing the imported wind turbine geometry into the rotational domain and external flow field domain, employing the aforementioned dual-region sliding mesh strategy and boundary layer mesh configuration scheme. The solver scheduling subunit calls Fluent's Journal file interface through MATLAB's system command, sequentially starting batch simulation tasks according to the preset wind speed condition sequence and tip speed ratio parameters, achieving unattended automated calculation. The post-processing data extraction subunit automatically reads the time-series data of the torque coefficient after each simulation task, calculates key aerodynamic performance indicators such as wind energy utilization coefficient, maximum wind energy utilization coefficient, optimal tip speed ratio, and start-up wind speed under each operating condition, and transmits the results to the multi-objective optimization construction module in a structured data format. The specific technical parameters regarding grid density, turbulence model selection, and time step settings in the aforementioned method embodiments are applicable to the configuration of this module and will not be repeated here.

[0059] The multi-objective optimization construction module 3 corresponds to step S3 in the method embodiment. This module is configured to receive aerodynamic performance data transmitted by the multi-condition CFD simulation module 2, and establish a multi-objective collaborative optimization model with the number of blades, torsion angle, and overlap ratio as decision variables. The objective function is to maximize the weighted comprehensive wind energy utilization coefficient, and the constraints are that the equivalent stress of the blades does not exceed the allowable stress of the material and the starting wind speed does not exceed a preset threshold. Internally, this module maintains a dynamically updated aerodynamic performance database. Whenever new verification data is transmitted from the closed-loop feedback verification module 5, this database is automatically expanded and the response surface is refitted, enabling the optimization model to continuously absorb new simulation information to improve prediction accuracy. The specific parameter values ​​regarding the objective function weight setting, allowable stress calculation, safety factor selection, and starting wind speed threshold in the aforementioned method embodiment are all applicable to this module and will not be repeated here.

[0060] The adaptive hybrid optimization module 4 corresponds to step S4 in the method embodiment. This module is configured to use a hybrid optimization strategy that combines particle swarm optimization (PSO) and genetic algorithm to solve the multi-objective collaborative optimization model. Internally, this module integrates three core components: a PSO velocity-position update engine, a GA crossover and mutation engine, and an adaptive parameter regulator. The PSO engine is responsible for the global search of the population particles in the design variable space; the GA engine is responsible for performing crossover and mutation operations on low-fitness particles to enhance population diversity; and the adaptive parameter regulator calculates the inertia weight and crossover probability in real time based on the current iteration progress and the population fitness variance. All specific algorithm parameters and implementation details in the aforementioned method embodiment, including the cosine decreasing inertia weight formula, adaptive crossover probability strategy, population size, learning factor, mutation probability, integer variable handling method, and termination condition, are applicable to the configuration of this module. The output of this module is the optimal combination of design variables that satisfies the constraints and its corresponding optimization prediction objective function value.

[0061] The closed-loop feedback verification module 5 corresponds to step S5 in the method embodiment. This module is configured to receive the optimal design variable combination and optimized prediction value output by the adaptive hybrid optimization module 4, send the design variable combination back to the 3D parametric modeling module 1 to trigger remodeling, and then perform verification simulation through the multi-condition CFD simulation module 2 to obtain the verified wind energy utilization coefficient and compare it with the optimized prediction value. This module contains two functional components: a deviation calculation subunit and a convergence decision subunit. The deviation calculation subunit calculates the deviation between the predicted value and the verified value according to the aforementioned relative deviation formula; the convergence decision subunit compares the deviation with a preset convergence threshold. If the deviation does not exceed the threshold, the final optimized scheme is directly output; if the deviation exceeds the threshold, the verification data is sent back to the multi-objective optimization construction module 3 to update the database and triggers the adaptive hybrid optimization module to re-execute the optimization, thereby realizing closed-loop iteration. The parameters such as the convergence threshold value and the maximum number of closed-loop iterations in the aforementioned method embodiment are also applicable to this module.

[0062] The five modules mentioned above are interconnected through data interfaces to form a closed-loop system architecture. The 3D parametric modeling module 1 and the multi-condition CFD simulation module 2 constitute the physical simulation layer; the multi-objective optimization construction module 3 and the adaptive hybrid optimization module 4 constitute the intelligent optimization layer; and the closed-loop feedback verification module 5 serves as a bridge connecting the physical simulation layer and the intelligent optimization layer, ensuring consistency between the optimization results and the physical reality. In a complete optimization run, the system typically performs approximately 200 to 500 CFD simulation calls. The total computation time depends on the duration of each CFD simulation and the available parallel computing resources. In a workstation environment equipped with a 16-core CPU and 32GB of memory, the total runtime is approximately 48 to 72 hours.

[0063] In one embodiment of the present invention, data interaction between the modules of the above system is transmitted using a unified structured file format. The wind turbine geometry output by the 3D parametric modeling module 1 is stored in STEP format, along with a JSON parameter configuration file containing all design variable values. After completing the simulation calculation, the multi-condition CFD simulation module 2 writes the torque coefficient time series, wind energy utilization coefficient characteristic curve, and start-up wind speed data under each condition into a structured CSV data file, and generates a log file containing key simulation configuration parameters for traceability. The multi-objective optimization construction module 3 reads aerodynamic performance indicators from the CSV data file and updates its internal database, while the adaptive hybrid optimization module 4 efficiently exchanges data with the PSO engine and GA engine by reading and writing shared memory areas. The closed-loop feedback verification module 5 fully records the design variable combinations, optimization prediction values, and verification values ​​for each round, generating a closed-loop iteration tracking report to facilitate engineers' evaluation of the convergence quality of the optimization process.

[0064] Preferably, the system also features parallel computing scheduling and management capabilities. When the workstation is equipped with a multi-core processor, the multi-condition CFD simulation module can allocate simulation tasks for different wind speed conditions or different tip speed ratio conditions to different CPU cores for parallel execution, thereby reducing the simulation time per round from approximately 45 minutes in serial mode to approximately 15 minutes. Simultaneously, during the particle swarm optimization phase, combinations of design variables corresponding to multiple particles can also be submitted to the modeling and simulation process for parallel evaluation, further improving overall optimization efficiency. By rationally configuring parallelism and task scheduling strategies, the system can compress the complete optimization runtime to approximately 24 to 36 hours in a 24-core workstation environment.

[0065] To verify the effectiveness of the method of this invention, an optimization experiment was conducted using a Savonius wind turbine with an outer diameter of 0.3m and a height of 0.3m as the experimental object. The search range for the design variables in the experiment was set as follows: number of blades 2 to 5, twist angle 0° to 180°, and overlap ratio 0 to 0.3. The rated wind speed was 8m / s, and the benchmark was a traditional, unoptimized two-bladed semi-circular Savonius wind turbine.

[0066] After synergistic optimization using the method of this invention, the optimal design scheme is as follows: 3 blades, a torsion angle of approximately 105°, and an overlap ratio of approximately 0.12. The aerodynamic performance indicators under various operating conditions are as follows: At a low wind speed of 3 m / s, the maximum wind energy utilization coefficient is 0.198, an improvement of approximately 21.5% compared to the traditional design's 0.163; at a rated wind speed of 8 m / s, the maximum wind energy utilization coefficient is 0.268, an improvement of approximately 16.0% compared to the traditional design's 0.231; and at a high wind speed of 14 m / s, the maximum wind energy utilization coefficient is 0.242, an improvement of approximately 12.6% compared to the traditional design's 0.215. The starting wind speed is reduced from approximately 3.5 m / s in the traditional design to 2.6 m / s, a reduction of approximately 25.7%, meeting the design target of below 3.0 m / s. The maximum equivalent stress at the blade root is 78.3 MPa, far below the material's allowable stress of 102.5 MPa, indicating sufficient structural safety margin.

[0067] The method of this invention was compared with single-factor stepwise optimization and single genetic algorithm optimization methods. The single-factor stepwise optimization method first fixed the number of blades to 3 and the overlap ratio to 0, optimizing to obtain a twist angle of approximately 45°. Then, fixing the number of blades and the twist angle, it optimized to obtain an overlap ratio of approximately 0.15. The final maximum wind energy utilization coefficient at rated wind speed was 0.243, which is about 9.3% lower than the method of this invention. The single genetic algorithm optimization method, using the same three-dimensional parameter space and constraints, optimized to obtain a maximum wind energy utilization coefficient of 0.251 at rated wind speed after 200 generations of population evolution, which is about 6.4% lower than the method of this invention, and its convergence speed was significantly slower. These comparative results show that the multi-parameter collaborative optimization method and the PSO-GA hybrid algorithm of this invention are superior to existing single-factor optimization and single-algorithm optimization schemes in terms of optimization accuracy and convergence efficiency.

[0068] Furthermore, the effectiveness of the closed-loop feedback verification mechanism was analyzed. In the first round of optimization, the relative deviation between the optimal predicted objective function value output by the PSO-GA hybrid algorithm and the CFD verification value was 6.8%, exceeding the 5% convergence threshold. After the second round of closed-loop iteration, the deviation decreased to 3.2%, meeting the convergence requirement. Compared with the direct optimization scheme without closed-loop verification, its final CFD-verified wind energy utilization coefficient was 0.258, which is about 3.7% lower than the 0.268 of the closed-loop verification scheme. This indicates that the closed-loop feedback mechanism effectively eliminated model bias in the optimization process and improved the reliability of the final scheme.

[0069] The experimental results above fully demonstrate that this invention achieves globally optimal design of the number of blades, torsion angle, and overlap ratio of the Savonius wind turbine by deeply coupling five aspects: three-dimensional parametric modeling, multi-condition CFD simulation, multi-objective collaborative optimization, adaptive hybrid PSO-GA optimization, and closed-loop feedback verification to form a closed-loop collaborative system. The optimized three-bladed helical torsion angle wind turbine has a power coefficient that is about 12% to 18% higher than that of the traditional design at rated wind speed, and the starting wind speed is reduced to below 2.8 m / s, significantly expanding the application range in low wind speed environments and demonstrating good engineering practical value.

[0070] Furthermore, to further examine the contribution of each core step in the method of this invention to the final optimization effect, ablation experiments were also conducted. When the closed-loop feedback verification step was removed and the scheme of the first round of optimization output was directly adopted, the wind energy utilization coefficient at rated wind speed decreased to 0.258, which is about 3.7% lower than that of the complete method, indicating that the closed-loop verification mechanism can effectively correct the model bias in the optimization process. When the adaptive hybrid PSO-GA algorithm was replaced with the standard PSO algorithm, after the same 200 iterations, the wind energy utilization coefficient at rated wind speed was 0.249, which is about 7.1% lower than that of the complete method. Moreover, the standard PSO showed convergence stagnation at about the 120th iteration, while the hybrid strategy of this invention still maintained a slow but stable fitness improvement at the 180th iteration, demonstrating the positive effect of GA crossover and mutation operation on enhancing population diversity and helping to escape local extrema. When the cosine decreasing inertia weights are replaced with traditional linear decreasing inertia weights, the final wind energy utilization coefficient is 0.261, which is about 2.6% lower than the complete method, indicating that the cosine decreasing strategy does have an advantage in balancing global search and local fine search.

[0071] From a computational efficiency perspective, a complete optimization run on a workstation equipped with an Intel Xeon E5-2680v4 processor and 32GB of memory takes approximately 56 hours, with each CFD simulation averaging about 15 minutes. The total number of CFD simulations called during the 200 generations of particle swarm optimization is approximately 350. Compared to the thousands of CFD simulations required for exhaustive searching of the 3D design variable space using the full factorial traversal method, this invention reduces computational costs by approximately an order of magnitude while maintaining effective capture of the global optimum.

[0072] Furthermore, the optimized wind turbine design was validated through physical prototype manufacturing and wind tunnel experiments. The wind tunnel experiments were conducted in a low-speed open wind tunnel with a cross-sectional area of ​​1.2m x 1.2m, with wind speeds ranging from 2m / s to 15m / s. A six-component balance was used to measure the aerodynamic torque of the wind turbine. Experimental results showed that the measured maximum power coefficient at the rated wind speed of 8m / s was 0.253, a deviation of approximately 5.6% compared to the CFD simulation prediction of 0.268. This deviation is within the acceptable range for engineering applications and is mainly attributable to wind tunnel wall effects and manufacturing accuracy errors. The experiments also verified that the wind turbine can reliably self-start under low wind speeds ranging from 2.5m / s to 3.0m / s, which is basically consistent with the simulation prediction of a starting wind speed of 2.6m / s. These wind tunnel experimental results further confirm the effectiveness and reliability of the method of this invention in the multi-parameter collaborative optimization of Savonius wind turbines.

[0073] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. The Savonius method for multi-parameter collaborative optimization of wind turbines, characterized in that, Includes the following steps: Step S1, 3D parametric wind turbine modeling step: Based on SolidWorks software, a 3D parametric geometric model of the Savonius wind turbine is established. The 3D parametric geometric model uses the number of blades, blade torsion angle, and blade overlap ratio as adjustable design variables. The blade torsion angle is defined as the cumulative rotational offset angle of the blade from the bottom section to the top section along the wind turbine rotation axis. The blade overlap ratio is defined as the ratio of the overlap distance of adjacent blades in the radial direction of the wind turbine to the outer diameter of the wind turbine. The wind turbine geometric entities corresponding to different combinations of design variables are generated through parametric driving. Step S2, Multi-condition CFD flow field numerical simulation step: Import the wind turbine geometric entity generated in step S1 into the MATLAB and Fluent joint simulation platform, establish a computational fluid dynamics model including the rotation domain and the external flow field domain, perform unsteady flow field numerical simulation on the wind turbine corresponding to each combination of design variables under multiple preset wind speed conditions, and extract the torque coefficient and wind energy utilization coefficient under each condition as aerodynamic performance evaluation indicators. Step S3, Multi-objective collaborative optimization model construction steps: Taking the maximization of the weighted comprehensive value of the wind energy utilization coefficient obtained in step S2 as the objective function, and taking the equivalent stress of the wind turbine blades not exceeding the allowable stress of the material and the starting wind speed not exceeding the preset threshold as the constraints, a multi-objective collaborative optimization model with the number of blades, torsion angle and overlap ratio as decision variables is established. Step S4, Adaptive Hybrid Intelligent Optimization Step: The multi-objective collaborative optimization model established in Step S3 is solved using a hybrid optimization strategy that combines particle swarm optimization and genetic algorithm. In the particle swarm optimization, an adaptive inertia weight that decreases nonlinearly with the number of iterations is introduced to balance global exploration and local development capabilities. In the genetic algorithm, a crossover probability that is adaptively adjusted with the population fitness variance is set to maintain population diversity. The optimal combination of design variables that satisfies the constraints is output.

2. The method according to claim 1, characterized in that, In step S1, the number of blades is an integer ranging from 2 to 5, the blade torsion angle is ranging from 0° to 180°, and the blade overlap ratio is ranging from 0 to 0.

3.

3. The method according to claim 1, characterized in that, In step S2, the multiple preset wind speed conditions include low wind speed condition, rated wind speed condition and high wind speed condition. The wind speed range of the low wind speed condition is 2 m / s to 4 m / s, the wind speed range of the rated wind speed condition is 6 m / s to 10 m / s, and the wind speed range of the high wind speed condition is 12 m / s to 16 m / s.

4. The method according to claim 1, characterized in that, In step S3, the preset threshold is 3 m / s, and the allowable stress of the material is determined by dividing the yield strength of the blade material by a safety factor, wherein the safety factor ranges from 1.5 to 2.

5.

5. The method according to claim 1, characterized in that, In step S4, the adaptive inertia weight is calculated as follows: the inertia weight is set to decrease from the initial maximum value to the final minimum value according to the cosine decreasing function with the number of iterations. The initial maximum value ranges from 0.8 to 0.95, and the final minimum value ranges from 0.2 to 0.

4.

6. The method according to claim 1, characterized in that, In step S4, the adaptive adjustment of the crossover probability is determined as follows: when the population fitness variance is greater than a preset variance threshold, the crossover probability is set to a first crossover probability value; when the population fitness variance is not greater than the preset variance threshold, the crossover probability is increased to a second crossover probability value, and the second crossover probability value is greater than the first crossover probability value.

7. The method according to claim 1, characterized in that, In step S2, the unsteady flow field numerical simulation adopts the k-ω SST turbulence model. The time step corresponds to the time interval that the wind turbine rotates by 1° to 2°. The diameter of the rotation domain of the computational fluid dynamics model is 1.5 to 2.0 times the outer diameter of the wind turbine, and the length of the external flow field in the direction of the incoming flow is 15 to 25 times the outer diameter of the wind turbine.

8. The method according to claim 1, characterized in that, In step S1, the three-dimensional parametric geometric model also includes wind turbine end plate parameters, which include end plate diameter and end plate thickness. The end plate diameter is 1.05 to 1.15 times the outer diameter of the wind turbine.

9. The method according to claim 1, characterized in that, It also includes step S5, a closed-loop feedback verification step: the optimal design variable combination output in step S4 is fed back to step S1 to regenerate the corresponding wind turbine geometry, and CFD verification simulation is performed in step S2. The wind energy utilization coefficient obtained from the verification is compared with the optimized prediction value in step S4. If the relative deviation exceeds the preset convergence threshold, the verification result is used as a correction sample to update the optimization model in step S3 and the optimization process in step S4 is re-executed until the relative deviation converges to within the preset convergence threshold, which is 3% to 5%. The maximum number of iterations for closed-loop feedback verification does not exceed 5.

10. A Savonius multi-parameter collaborative optimization system for wind turbines, used to implement the method described in claim 9, characterized in that, include: The three-dimensional parametric modeling module is configured to build a three-dimensional parametric geometric model of the Savonius wind turbine based on SolidWorks software. The three-dimensional parametric geometric model uses the number of blades, blade torsion angle and blade overlap ratio as adjustable design variables, and generates wind turbine geometric entities corresponding to different combinations of design variables through parametric driving. The multi-condition CFD simulation module is configured to import the wind turbine geometric entity generated by the three-dimensional parametric modeling module into the MATLAB and Fluent joint simulation platform, establish a computational fluid dynamics model, perform unsteady flow field numerical simulation under multiple preset wind speed conditions, and extract the torque coefficient and wind energy utilization coefficient. The multi-objective optimization construction module is configured to maximize the weighted comprehensive value of the wind energy utilization coefficient obtained by the multi-condition CFD simulation module as the objective function, and establish a multi-objective collaborative optimization model with blade equivalent stress and start-up wind speed as constraints. The adaptive hybrid optimization module is configured to solve the multi-objective collaborative optimization model by using a hybrid optimization strategy that combines particle swarm optimization and genetic algorithm, and introduces adaptive inertia weights and adaptive crossover probabilities to output the optimal combination of design variables. The closed-loop feedback verification module is configured to send the optimal design variable combination back to the three-dimensional parametric modeling module to regenerate the wind turbine geometry. The verification simulation is performed by the multi-condition CFD simulation module. The wind energy utilization coefficient obtained from the verification is compared with the optimized prediction value. If the deviation exceeds the preset convergence threshold, the multi-objective collaborative optimization model is updated and the adaptive hybrid optimization module is triggered to solve the problem again until the deviation converges.

Images