Water Vapor Forecast Method Research Based On Wavelet Neural Network

Real-time water vapor distribution and forecast has the extremely important theory significance and applications in the area of weather analysis and forecasting, microclimatology, global climate change, severe weather forecast and so on. So using GNSS for getting and forecasting the water vapor is of great significance. Getting water vapor has been mature at present, but the water vapor prediction remains a matter of research. Water vapor forecast are not uniformity in both time and geographical distribution that it’s difficult to forcast accurately, thus many scholars adopt various methods to explore accurately forecast the water vapor. The BP neural network has been research hot spot in the steam prediction scholars both at home and abroad, also has been used to obtain a certain achievements in the field of water vapor forecast. But it also has limitations which cannot get higher forecast precision. Wavelet neural network is an improved neural network which regards wavelet basis function as the excitation function of neurons in the neural network model, it combines the multi-scale analysis of performance in wavelet analysis and the advantages of neural network self-learning and adaptive.This thesis has used the high-precision processing software called PLAOD to calculate the tropospheric delay. Combined the station data with meteorological data, using water transformation model,we can calculate the precipitable water vapor data which meet some specific requirements as the sample data of the experiments of this thesis. Then, using wavelet neural network forecast model to forecast the precipitable water vapor. Comparing it with the experimental data, we can analyze that the wavelet neural network in water vapor precipitation are better than traditional BP neural network in convergence, fault tolerance, better approximation ability, thus we can make a conclusion that the predicted precision is higher. When we analyze different sampling rate data in the same time, it is concluded that the30s sampling rate data of wavelet neural network to predict water vapor is best. The precipitation residual value can reach0.001mm, the average absolute value is0.018mm, so its’ certain application reference value.