Research On Time Series Prediction Based On RBF Neutral Network

Prediction is one of the application of time series analysis. people analyze a large amount of observation data in order to predicting the characteristic of the data in the future, in order to deal with or control the characteristic. It is analyzed in terms of pure mathematics that traditional time series analysis is, but because the time series in practical application has non-linear characteristics such as irregular, chaos and so on. It is very difficult to set up ideal model to the system, predict more powerless. And the artificial neural network has learning ability organizing, since, approaches ability non-linearly because of it, used for the time series and predicted more effective.The thesis has introduced time series methods for predicting at present and various prediction models at first. Then introduced neural network application and good characteristic and advantage shown in non-linear prediction model rising in recent years, especially introduced the superior characteristic of RBF network appears in the time series in detail, point out the defect of the algorithm of RBF network at the same time, improve it on this basis. Use covering cluster algorithm of improved for carrying on the cluster to the sample data, and optimize cluster's number with orthogonal least squares learning algorithms, raise the processing speeds to the sample cluster, and then has guaranteed study of network is fast. The experiment has shown the improved validity of algorithms.Finally, the endowment insurance income was predicted by the improved RBF network. It is compared with the RBF model, BP network model and actuarial science model precisely. The experimental result indicates that the prediction result of improved RBF network is better than RBF network, BP network and actuarial science model, then verified the feasibility of the model.