Research On Flow Field Reconstruction Method Of Complex Terrain Based On Convolution Neural Network

As a part of wind resource assessment,the site selection of wind farm is of great significance to the development and utilization of wind resources.In the wind resource assessment methods,the main use of anemometer tower for data collection and computational fluid dynamics(CFD)simulation of wind speed distribution changes,using these methods need to deal with a large amount of data,will increase the development of wind resources utilization cycle.The utilization of vast amounts of data,such as anemometer tower observations and numerical calculations,has emerged as a pressing issue that needs to be addressed.In this paper,the flow field reconstruction method is applied to wind farm site selection and wind resource assessment,and a flow field reconstruction method based on Convolutional Neural Network(CNN)is proposed.CFD numerical simulation and deep learning are combined in the research of reconstruction method,which provides some reference for the future application of deep learning and neural network in wind resource assessment.The main research work is as follows:(1)A convolution neural network is proposed to reconstruct the velocity field around Bolund Island surface based on wind profile,which enriches the application scenarios of convolution neural network.Convolution neural network can play a key role in studying the problem of flow field reconstruction,Mining the information of flow field characteristics from a large number of jumbled flow field information,extracting the flow field characteristics in convolution layer,reorganizing the flow field characteristics in linear layer,and constantly optimizing them.In this process,CNN can better retain the original flow field information;(2)Based on the standard turbulence model,the neutral atmospheric boundary condition is used in this paper.By introducing the source term into the transport equation of turbulent kinetic energy and dissipation rate,the good self-preservation characteristics of atmospheric boundary layer are realized.Based on the standard model Bolund Island,the CFD numerical simulation of complex terrain wind field under different wind directions is carried out by using high-quality grid,and the calculated results are compared with the field measured data,which proves the effectiveness and reliability of the proposed method;(3)The reconstructed method takes the input and output flow fields of numerical wind profiles as sample labels,and uses the model established by CNN to train,and establishes a flow field reconstruction method based on the mapping model of wind profiles and flow fields.Through the comparison between the reconstructed and CFD flow field,it is proved that the reconstructed method can quickly extract the flow field characteristics and realize the purpose of reconstructing the velocity field based on the wind profile,and can better retain the boundary information of the flow field structure and restore some velocity structure information of the original flow field in the reconstructed process;(4)By setting different sample numbers,the influence on the reconstruction accuracy is discussed.By analyzing the reconstructed wind profiles,it is found that the sample numbers have a great influence on the reconstruction accuracy of the model.Based on the above research,the mapping model uses the test data beyond the training sample set to study the extrapolation ability of CNN.When the flow phenomenon in the working condition section included in the test set is different from the training set range used by the model,the prediction accuracy of the model decreases and the extrapolation ability of CNN is limited.