Research On Prediction Method Of Icing Quality Of Wind Turbine Impeller Based On DNN And SVR Algorithm

Wind energy is a green,clean and renewable new energy source,which is favored by most countries.Most wind turbines are commonly installed in the cold plateau and coastal wind resource-rich areas,due to high altitude,low temperature and complex meteorological conditions,it is very easy to cause icing phenomenon formed by super-cooled water droplets hitting the blades,which in turn leads to lower power generation efficiency of the unit and blade breakage,which seriously restricts the development of wind power industry.Due to the different ice cover area and icing quality,it will make the impeller structure of the inherent frequency change.In order to realize the early prediction of wind turbine icing state,this thesis is aimed at 600 W small wind turbine to carry out the prediction of wind turbine impeller icing quality in different icing states.By establishing a blade ice cover model,conducting blade ice cover tests and impeller test modal tests under natural winter conditions,the relationship equation between the rate of change of intrinsic frequency and icing quality in different ice cover regions is established.Then,a deep neural network model and a support vector regression algorithm model are constructed to predict the icing quality in different ice cover areas of the impeller based on the intrinsic frequency variation rate.Finally,the better-performing ice-cover prediction model method is extended to large wind turbine impeller ice-cover prediction.The details of the study are as follows:Firstly,the model of wind turbine impeller was established by using FARO3 D scanner and Solidworks 3D software,and modal simulation analysis was conducted to compare with the modal parameters of the impeller structure measured by PULSE modal test system.The experimental and simulated modal results of the ice-free impeller are close to each other,the error of the first-6th orders of inherent frequency was within 9%,and the corresponding vibration pattern of each order was basically consistent.Secondly,there is no significant change in the modal shape of the impeller under the thin ice cover compared with the uncovered state,and the absolute value of the rate of change of the intrinsic frequency of each order shows an increasing trend with the increase of the ice cover mass,which indicates that icing has a significant effect on the intrinsic frequency.The relationship between intrinsic frequency variation and icing quality is well fitted with the correlation coefficients above 0.93.Thirdly,it is found that the relative error of prediction increases when the ice cover mass is smaller,but the relative error between the predicted and actual values decreases as the ice cover mass increases.The deep neural network model shows better results in impeller icing quality prediction compared to the support vector regression algorithm model,indicating that the deep neural network model is more suitable for the prediction of wind turbine icing quality.Fourthly,the finite element modal analysis of the ice-covered impeller of a1.3 MW large wind turbine was performed,and the overall average error of the prediction using the deep neural network model was 5.66%.Which was small and showed good prediction capability compared with the method of Gantasala using BP neural network to predict the ice-covered mass of a large wind turbine blade.Therefore,the prediction method proposed in this thesis can achieve the prediction of wind turbine impeller icing quality,which provides a new idea for wind turbine blade icing fault detection as well as deicing decision.