The Application Research On Prediction And Classification Based On Artificial Neural Network

The climate change has an important impact on the extreme temperature, precipitation and even human health in the context of global climate change. The realistic significance of extreme minimum temperature and precipitation forecast on the social economy, agricultural production and the city water logging prevention and control. This paper proposed method of the artificial neural network model of the extreme temperature and precipitation. Meanwhile, researching on the classification of cancer gene based on the data of colon cancer gene. This study can provide a reliable theoretical basis for the accurate prediction of the extreme temperature and precipitation accuracy and correct classification of cancer genes.This paper mainly study on two aspects of artificial neural network. On one hand, it is the theoretical study of algorithms of selecting the RBF artificial neural network center. The methods of phase space reconstruction and principal component analysis are used to optimize the artificial neural network and using trial calculation method to select the smooth factor of RBF artificial neural network model. On the other hand, it is some applications of climate forecast and classification of cancer genes and then doing some realistic situation forecast based on model of extreme temperature and precipitation.In this paper, using the climate data of Zhangbei city from 1956 to 2009 as research object.Firstly, establishing extreme temperature model and precipitation model based on several different artificial neural networks, and then comparing the forecast results with different models. Secondly, using the method of phase space reconstruction to determine the input dimension of network and reduce the redundancy independent variable. Therefore, a forecast model, which can overcome the disadvantages of RBF network and improve the forecasting accuracy, based on the combination of phase space reconstruction and RBF neural network is established. Thirdly, the paper uses BP, PAC-BP and PLS methods to establish the precipitation forecast model. Meanwhile, the main affect factors of precipitation is analyzed and the multi variable equation is obtained by the PLS precipitation forecast model. The results show that the PLS algorithm not only solves the instability of the neural network, but also shortens the running time of the network and improves the generalization ability. Finally, the paper introduces Neural network algorithm to the classification of cancer gene and uses BP, SVM, S-Kohonen methods to establish three forecast models of the classification of cancer gene, and to forecast the classification of genetic samples about recurrence and non-recurrence after cancer surgery. The prediction results of comparing these three methods show that the forecast categories based on S-Kohonen method is basically consistent with the test category, which has a better classification effect.