Research And Application Of Liuzhou Precipitation Model Based On Genetic Algorithm And Support Vector Machine

The reliability and timeliness of precipitation prediction ability will be of great guiding significance for reducing the economic losses caused by rainstorm and protecting the safety of local people's life and property.In order to establish the optimal precipitation prediction model,this paper needs to overcome the problems of the high dimension of the influence factors and the non-linearity of the precipitation system,so we select the support vector machine model as the main body of this paper.This is because it has a lot of unique advantages,especially in solving the finite sample,nonlinear,high dimensional problem,and pointing at its defects,we use the genetic algorithm to optimize it.The specific research process is as follows:In the first chapter,we first explain the reasons for the study of precipitation forecast in Liuzhou,and then introduce the background and significance of the selected topic.Secondly,the background and recent research situation of the two algorithms are discussed.Finally,the questions and the solutions to problems are expounded.In the second chapter,the definition of the principal component analysis(PCA),and the definition of the kernel principal component analysis(PCA),and the implementation steps and some other preparatory knowledge are given successively.And then introduce the related theories of SVM and GA briefly,the optimization characteristics of the genetic algorithm are also shown and an example is illustrated at the end of this section to explain it.In the third chapter,considering the existing problems in the process of modeling,two analysis methods are introduced,and give the implementation steps to clean and compress the data in detail.Finally,aiming to improve the accuracy of the prediction,the global nature of the spatial exploration of the genetic algorithm is taken into consideration,so we give the detailed operation steps to optimize the parameters of the model.In the fourth chapter,we take the precipitation forecast in Liuzhou as the research object,and compare the effect of the model.In order to make the model to learn a good rule of data,it is necessary to preprocess the data.In this chapter,we compare the standardization of data is reduced and the effect.When applying kernel principal component analysis,we analyze the dimensionality reduction results based on three different kernel functions.Finally,according to the optimization of the parameters of the genetic algorithm in the third chapter,the optimal model is created by using the optimal solution of the parameters,and the results of the prediction of the three different models are compared.Through empirical study,although the KPCA reduction effect based on Sigmoid kernel is the best,the result shows that the generalization ability of the SVM model based on the KPCA extracted by the Gauss kernel function is the best.This indicates that,the data obtained by dimensionality reduction can make the calculation simple,but it also loses some of the laws in the original data,so that reducing the learning ability of the model,which directly leads to the unsatisfactory results.Meanwhile,after comparing the prediction results of the three models,it is concluded that the fitting result of the KPCA-GA-SVM's model is the best and the accuracy is the highest.