Font Size: a A A

Study Of Time Series Feature Selection And Prediction

Posted on:2017-02-08Degree:MasterType:Thesis
Country:ChinaCandidate:J F FanFull Text:PDF
GTID:2308330485461836Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
Recently time series prediction problems has been emerging in different appli-cation fields including communication, energy, finance, police and so on. In every specific application, there will be different requirements for the time series processing methodology, where old method won’t work. In these problems time series feature extraction, classification, clustering, modeling and regression prediction are the main focus. Some new methods in time series feature extraction, clustering and regression prediction used in their applications are studied in this paper.In the application of off-line network customers prediction, a new high level time series feature which represents people usage trend is proposed against the user behav-ior. We also present a Group Lasso-based feature selection method to predict the latent off-network customers by analyzing the corresponding multisource teledata. Extensive experiment results show that the proposed approach has the superior performance.In the application of traffic prediction for telecoms, we extract three kinds of fea-tures (Differential, Gaussian and Histogram) from the traffic fields, on which K-means and KNN methods are used to find out the similar community base stations. Then ratio is calculated to predict traffic of the community base stations.In the application of power load forecasting of Jiangsu Province, we compare different granularity of SARIMA model and find a new way of using the time series model. Also we take other facts into consideration and use GRNN model to predict the power load. For the special situation around the Chinese Spring Festival, we find a way to revise the predicted value, which approves good.
Keywords/Search Tags:Time Series Prediction, Feature Extraction, Off-line Customers Predic- tion, Traffic Prediction, Power Load Forecasting
PDF Full Text Request
Related items