Research On Radar Echo Extrapolation Method Based On Fourier Transform

With the rapid development of artificial intelligence technology,the combination of neural network and meteorological field is more closely.In terms of disaster early warning,the short-impending forecast system plays an important role.The main method in the current business is the extrapolation method based on radar echoes,but there is still much room for improvement in the forecasting effect.From the perspective of training data quality control and preprocessing and deep learning model,this thesis studies and improves the prediction method based on U-Net network,and applies the prediction results to the field of civil aviation.The main research content of this thesis is as follows:(1)This thesis first analyzes the problems caused by the radar echo strength networking strategy commonly used in the current business in view of the quality of the model input data,and finds that using the strategy of taking the maximum value in the overlapping area of the network will lead to echoes in the overlapping boundary area The phenomenon of sudden change in intensity,which will cause the model to learn the wrong echo motion law.To solve this problem,an improved networking strategy is proposed.The improved method not only takes into account the data smoothness near the border of the overlapping area,but also preserves the strong echo characteristics.In this way,the data after quality control is used as the training data of the prediction model,which can not only ensure that the model can correctly learn the echo motion characteristics in the overlapping area of the network,but also ensure the ability of the model to retain high-intensity echoes.(2)For radar echo data,it is first proposed to use spectral decomposition to pre-extract multi-scale echo information as the input data of the deep learning network to enrich the feature information of the input data.The main framework of the model adopts U-Net neural network,and a residual connection structure is added to solve the problem of model degradation.Based on the above design,this study proposes the Sp At-Res UNet prediction model.The test results show that compared with the traditional radar echo extrapolation algorithm SPROG and the deep learning network Res UNet model,the model has an overall improvement in the evaluation indicators of the radar echo prediction extrapolation results in the next hour and has a strong echo retention ability score has been enhanced.(3)In order to draw the dynamic dangerous weather flight restricted area more accurately,it is proposed to combine the Sp At-Res UNet prediction model in this thesis with the improved Graham algorithm to predict the approximate range of the dangerous weather flight restricted area at a certain moment in the future,and Combining with the density-based clustering method(DBSCAN)and considering the flight safety,a complete set of procedures for the designation of restricted areas for hazardous weather flights is completed.Finally,the accuracy,redundancy,and deviation are used to evaluate the designated flight restricted area.The results show that compared with the traditional method of designating dangerous weather flight restrictions,the design process of the flight restricted area combined with the radar echo extrapolation method can Significantly improve the above three evaluation indicators.