| As global warming and environmental problems become more and more critical,wind energy have been increasingly used as clean and renewable energy.Vertical axis wind turbines(VAWT)are a type of small-scale wind power equipment that can adapt to harsh wind farm environments.Due to the uniform cross-section of its blades,the improvements of wind energy utilization are mainly contributed by optimizing the cross-section of wind turbine rotor.Traditional rotor design and optimization use analytic expressions based on mathematical definitions to describe the cross-section shape.However,due to the limitation of the theoretical knowledge of mathematics and aerodynamics,some rotors with novel shapes and excellent power generation efficiency cannot described directly through pre-designed mathematical expressions.One possible method is to use computer vision technology to extract image features to parameterize the description of the rotor,but the extracted features are often too verbose and complex to be emploied for rotor optimization.This paper proposes a sparse image encoding and decoding algorithm for wind turbine rotor section parameterization,a novel rotor generation method for gengeration innovative rotors,and a rotor automatic optimization system to verify that the methods can generate a rotor section with better performance.The main work and innovations of this paper are as follows:1.A wind turbine rotor cross-section parameterization method is proposed,namely Residual-Recursive Autoencoder(RRAE).Compared with the general encoding and decoding algorithms designed for natural images,RRAE can take full advantage of the sparseness and regularity of the cross-sectional image,thereby compressing the high-resolution cross-sectional image into a set of latent code with a higher compression rate.And the original image can be decoded according to a set of latent code with high precision.This process is called rotor section parameterization,and the latent code is the rotor section parameters.In addition,in this paper,the RRAE is extended to other publicly available sparse image datasets.According to the different datasets and latent code dimension,this method improves the PSNR of the reconstructed image by 3.4% to 27.18% compared with the standard autoencoder.In addition,the residual recursive algorithm does not rely on specific network structures,and has a performance improvement from 2.5% to 14.11% even with a simple structure like fully connected network.2.This paper proposes a method to imporve the novelty of the generated blades.This method uses a cross-section random disturber to add a set of random noise to the blades in the frequency domain.The strength of the noise is controled by the distrubance intensity.By inverse transforming from the frequency domain,a novel cross-section will be generated and used to train the RRAE to improve the richness of cross-section modeling in the coding space.3.A wind turbine rotor section optimization method based on RRAE is designed to automatically find a section shape with higher performace.The optimization method can adaptively select the appropriate evolution direction in the coding space using genetic algorithm.By applying the above-mentioned parameterization method and novelty-constraint method to the rotor optimization system,the optimal torque coefficient is 12.67% higher than the optimal one in the training dataset. |
