| With the continuous development of radar remote sensing detection technology,the quantitative estimation of precipitation using weather radar has been more and more widely used.Radar is affected by the complex system and external interference.The measurement accuracy is low.In addition,the spatiotemporal characteristics of rainfall are strong and change rapidly.It is not ideal to estimate the effect of precipitation by establishing a relationship between radar reflectivity factor(Z)and rainfall intensity(I).This paper focuses on improving the accuracy of radar quantitative estimation of precipitation,improving the quality of radar detection data,and using artificial neural network excelent nonlinear problem solving capabilities to find a suitable method for estimating precipitation.The main research contents of this paper are as follows:1.The X-band radar power spectrum data processing method is studied.Firstly,based on the analysis of the time and frequency domain characteristics of meteorological echo,the meteorological echo is simulated;Secondly,use the simulated meteorological echo to quantitatively evaluate three noise level calculation methods;Focusing on the analysis and comparison of the spectral moment estimation performance of the frequency domain estimation algorithm and the pulse pair algorithm,an improved FFT / NS method is proposed for the problem that the spectral moment estimation performance is greatly affected by noise.Simulation experiments show that this method reduces the influence of noise on spectral moment estimation performance and improves anti-interference performance;Finally,the threshold of spectrum points and the threshold of SNR are used to identify the meteorological signal and improve the quality of the meteorological echo signal.2.The application of neural network in radar quantitative estimation of rainfall is studied.Firstly,using the radar detection data and measured data of ground rainfall stations in the southeastern area of Xi'an in June 2019,the reflectivity factor-rain intensity data pair is established according to the principle of spatial and temporal consistency,and the data is quality controlled;Secondly,through Back Propagation and Radial Basis Function neural network combined with data to establish radar estimation precipitation model for surface precipitation estimation,comparison with the Z-I relationship obtained by least squares fitting;According to the comparison of the four indicators of mean error,mean relative error,mean relative mean square error and correlation coefficient,and the cumulative rainfall estimation results show that: the accuracy of the two neural networks in estimating the precipitation model is better than the Z-I relationship;Finally,the study area is located in the Loess Plateau and the terrain has large fluctuations.The input of the model is added to the relative altitude and relative distance of the radar-rainfall station.This method can obtain accurate rainfall estimation values in both hourly rainfall and accumulated rainfall,which improves the accuracy of precipitation estimation.3.Design and implementation of weather radar signal processing driver software.According to the system requirements,the overall structure of the software and the division of functional modules are described;Using the Qt graphic application framework as a development tool,using modular design ideas and multithreaded programming methods,the network communication module,system control module,data acquisition module,information display module,and simulation test module were designed;The software provides a large amount of historical data for neural network training and provides guarantee for improving the quality of basic radar products. |
PDF Full Text Request