Study On The Forecasting Method Of Air Turbulence Which Near Convective Clouds Based On AMDAR

At present,all the measures to prevent convention induced turbulence are to use radar monitoring to avoid the region,so that most aircraft can avoid the convective center.However,due to the strong correlation between aircraft turbulence and airflow velocity,the positions corresponding to airflow and radar reflectivity are usually different,strong turbulence occurs in the safe range recommended by radar instruments occasionally.Therefore,it is very necessary to study a prediction model for this.In this paper,AMDAR is used to obtain the data of turbulence points and compared with satellite cloud images to get the points which are needed.The physical characteristics of turbulence are analyzed and sign the meteorological factors which have influence on turbulence by drawing pictures.Then,using multivariate linear regression and neural network to establish prediction models and evaluate them.The results show that the generalized regression neural network can predict convention induced turbulence satisfactorily.Firstly,the necessity of establishing prediction models for convention induced turbulence and the sources of data acquisition and processing tools are introduced.Then case selection and data processing are carried out.Vector maps of relevant physical quantities are drawn,and it is concluded that turbulence always occurs at the junction of ascending and descending airflow,with large shear or horizontal temperature gradient and strong ascending or sinking airflow.It can be seen that the wind direction,wind speed and temperature at high altitude have a great influence on the aircraft bump,and the independent variables are determined accordingly.Then the linear regression method is used to eliminate the obvious linear relationship between the independent variables and the bump,and to verify that there is no correlation between the independent variables.It is verified that the commonly used BP neural network and RBF neural network can not deal with this situation well.Finally,GRNN and PNN are selected to build prediction models,the prediction time of these is less than 0.11 s,and the highest prediction accuracy is 86.66%.The preliminary design idea of the forecasting system is given,which provides assistant reference for relevant personnel.