Study On The Short Impending Prediction Method Of GPS Precipitable Water Vapor

Water vapor is an important part of the atmosphere.Although the content of water vapor in the atmosphere is small,it changes quickly.Many natural disasters and extreme weather have an inseparable relationship with the change of water vapor content,such as precipitation,Blizzard and other medium and small scale disaster weather.The Atmospheric Precipitable Water can be retrieved from the ground-based GPS tropospheric delay,which provides a new method for the acquisition of GPS precipitable water.However,there are not many domestic research results on water vapor prediction.At the same time,the time series of GPS Atmospheric Precipitable Water has the characteristics of nonlinear and non-stationary,and the prediction of GPS precipitable water is generally used for half an hour or an hour,and the precision is also millimeter.However,water vapor has the characteristics of rapid change.It is of great significance for short-term weather or climate forecast to improve and improve the prediction accuracy of precipitation if we can grasp the change law ofGPS precipitation in time.In order to improve the effectiveness of the short-term and high resolution GPS precipitation forecast,this paper makes a thorough study of the GPS precipitation prediction method.The main work is:(1)this paper first introduces the principle of inversion of GPS precipitable water,and calculates the tropospheric zenith delay through high precision data processing software GAMIT.Then it combines the meteorological data of the ground station and the precipitation transformation model to calculate the time series of the 5 min interval of the precipitable water,which provides the precondition for the experiment.(2)because the topology structure of the wavelet neural network adopts the structure of BP neural network,there inevitably exist some defects of the BP neural network,such as overfitting,slow convergence,and easy to fall into the local extremum.In view of the above problems,this paper uses genetic algorithm to code the initial weights and threshold parameters of wavelet neural network,and then determines the optimal parameters by selecting cross variation and other operations as the initial input parameters of the wavelet neural network,so as to overcome the blindness and randomness of the threshold selection of the initial weight value of the wavelet neural network.It is easy to fall into local extremum and cause oscillatory effects.The experimental results show that the mean square root error of the genetic wavelet neural network prediction method is 0.124 mm and the mean absolute 100% error is 0.167%.The accuracy of the method is improved obviously compared with the BP neural network and the wavelet neural network method,and it can better reflect the variation characteristics of the precipitable water.(3)a prediction model based on genetic algorithm,wavelet decomposition and least squares support vector machine is established.The method first decomposes the time series of GPS precipitable water into low frequency and high frequency components by using wavelet decomposition,then optimizes the parameters of LSSVM by genetic algorithm,and then sets up a prediction model for each component.Finally,the final results are obtained by superposing and reconstructing the prediction results of each component.The experimental results show that the combined model has good generalization ability,and can effectively solve the problem of the neural network easily trapped in the local minimum,and improve the global prediction accuracy.