Intelligent Meteorological Prediction And Analysis Based On Situation Field Data

With the continuous development and improvement of computer technology,the meteorological data acquisition technology is becoming more and more mature,which brings mass meteorological data information and opens up a new research direction for meteorological prediction and analysis.The traditional meteorological prediction and analysis method is based on statistics,and the prediction and analysis results are highly dependent on the historical data information,so it is difficult to excavate the deep characteristics of the data.As a new research method of meteorological related problems,intelligent meteorological prediction analysis has strong ability to extract features from large-scale data sets,which is conducive to discovering potential meteorological laws and has significant effect on mining deep features of data.The topic of this thesis is derived from the actual business needs of Liaoning Province Meteorological Disaster Warning Center.Deep learning method is used to establish a meteorological analysis model based on historical situation field data,mainly for pre-processing of situation field data,clustering analysis of situation field image,and enhancement of extreme disaster weather situation field data.Alex Net deep learning method is used to analyze precipitation grade and establish visual query system.First of all,in the aspect of data pretreatment,the data processing work of historical actual data and historical situation field data is mainly completed,and the historical situation field data and historical actual data are matched based on longitude and latitude range,national weather station number and time stamp.Ten channel situation field data,such as height field,temperature field,humidity field and sea level pressure field,are matched and normalized to form "field-real" data set,which solves the problem of data mismatch.In addition,Fuzzy C-means(FCM)clustering algorithm is used to segment the situation field weather map to solve the problem of contour boundary misclassification caused by noise and regional inhomogeneity in the process of parsing the situation field data into upper-air weather map.In order to solve the problem of limited sample data of extreme disasters in Liaoning,Deep Convolution Generative Adversarial Networks(DCGAN)generation method was used to enhance the data samples of heavy precipitation,gale and high temperature situation.In addition,a precipitation classification analysis model based on the improved Alex Net method is studied,which is committed to solving the optimization study of Alex Net network structure under the input of 10-channel data.This method can classify the precipitation grade well according to the characteristics of situation field data.Then,in view of the over-fitting problem that is prone to occur in Alex Net learning process,regularization method and Adam optimizer are used to optimize the prediction analysis model in the optimization process,which can better improve the accuracy of the model.Finally,in order to facilitate users to operate a large number of historical real data and visualization of the situation field data,this thesis built a simple and easy to use visualization of the situation field data query system.At the end of the thesis,the progress of the work is summarized,the shortcomings are evaluated,and the prospect of the follow-up work is put forward.