Study Of The Prediction And Optimization In Aerodynamic Performance Of Vertical Axis Wind Turbine Based On Artificial Intelligence

Wind energy is benefitial to carbon peaking and carbon neutrality goals due to its large reserves,easy access,pollution-free and renewable characteristics.Vertical axis wind turbine(VAWT)is a promising type of equipment for wind energy utilization,which has the advantages of no yaw,stable structure and easy maintenance.However,at present,the aerodynamic mechanism of VAWT is complex and the efficiency is relatively low,which posed challenges to its large-scaling and the offshore engineering applications.Therefore,the research and aerodynamic optimization of VAWT is an urgent problem to be solved.In recent years,artificial intelligence(AI)methods have been developed rapidly.On the premise of guaranteeing accuracy,the computing efficiency has been greatly improved when they are combined with the traditional methods such as the wind tunnel test or numerical simulation.However,the research on AI methods in wind turbine aerodynamic performance prediction and optimization is still limited.In this paper,the aerodynamic performance prediction and optimization problems of VAWT are studied.By combining deep learning,machine learning and other AI methods with computational fluid dynamics,innovative achievements have been made various aspects,including: short-term wind speed prediction of wind farm,the lift-to-drag evaluation of the blade airfoil,VAWT shape parameter optimization and the performance prediction and analysis of twin-VAWT system.The main contents are summarized as follows:(1)Under large wind history,based on the feture minig and the deep learning algorithms,the models for short-term wind speed prediction of wind farms,including EEMD-GA-LSTM(one-point)and CNN-LSTM(regional),are established.Compared with traditional models(such as persistence method,ANN,etc.),the mean absolute errors of the proposed models are reduced by an average of 56.25% and 30.75%,respectively.The proposed model handled the problem that under large-scaled(millions)wind history,the short-term wind speed can hardly be accurately predicted because of the strong non-linear characteristics.The study can provide reference for wind farm early-warning,wind resource management and wind turbine control.(2)Based on a probabilistic machine learning model,RF-SFS-GPR,the key mechanical parameter,max lift-to-drag ratio,of the wind turbine airfoil is presented.The accurate predictions,including the point prediction and the interval prediction,to the target airfoil can be achieved within 7 minutes.Compared with the observed values from wind tunnel tests and numerical simulations,the point prediction accuracy of the proposed model is up to94.1%,and the relative standard deviation from the interval prediction is less than 2%.The proposed model handled the problems that traditional methods facesd,including the considerable time budget,the high cost and the difficulty of accurately quantifying the reliability of the results.The study can provide a more comprehensive reference for the mechanical design of the VAWT blade airfoil.(3)Based on the aerodynamic analysis model coupling with the optimization algorithm,DMST-CMAES,a high-efficient shape of the Trposkein Darrieus turbine under a given inlet wind speed range is obtained within a few minutes.Compared with the original model:Sandia 5m,the peak power coefficient of the optimized model increased by 12.5% at the inlet wind speed of 7.68 m/s.Also,the performance is holistic optimized under other inlet wind speed condition within the given range(5 m/s to 10 m/s).The study overcomes the problem of global and rapid optimization of the VAWT shape under a certain wind speed range.(4)Based on the CFD data and the neural network surrogate model: CFD-FNN,the aerodynamic torque of twin-VAWT with variable blade pitch angle is predicted.Compared with the pure CFD simulations,the total time cost is reduced by 77.56%,and the total prediction accuracy is as high as 99.76%,which effectively balances the contradiction between calculation accuracy and efficiency by using traditional methods.Meanwhile,the law of the performance of the twin-VAWT changing with the pitch angle is revealed,and the the mechanical mechanism of the blade-flow field interaction is further clarified.Based on the wind turbine theory,CFD,data science and computer science,this paper investigates the potential applications of AI methods in aerodynamic prediction and optimization of VAWT.The research results can not only provide technical support to VAWT wind farm,but also provide scientific and engineering references for interdisciplinary fields including civil engineering,wind engineering,cyber-physical system and digital twin.