Font Size: a A A

Study Chaotic Time Series Forecasting Model Based On BP Neural Networks

Posted on:2008-12-12Degree:MasterType:Thesis
Country:ChinaCandidate:M ChenFull Text:PDF
GTID:2178360215986014Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
There is no denying the fact that the method of predicting the futurebased on historical data is commonly used in science, economics andengineering. Time Series Forecasting, which constructs time series modelon the basis of historical data and then uses the model to forecast thefuture, is an important research direction in forecasting research area.Asfar as astronomy, hydrology and meteorological phenomena areconcerned, many time series such as sun-spots, amount of runoff, rainfallamount were discovered all including the chaotic character in recentyears.In the face of the chaotic time series largely existed in nature andsocial economic phenomena, the traditional method of statistical analysisperformed badly. Neural Network (NN) posesses excellent non-linearcharacter, which enables it to be extremely suitable to the forecastingresearch in chaotic array. Based on artificial neural network and chaotictheory, the forecasting research has become research hot spot andreceived special attention at present.This dissertation has done systematicand thorough research on the above mentioned problem.Firstly, this dissertation has made an simple introduction to theconcept and distinction method of chaotic time series.In addition,thisdissertation has introduced lorenz model and summarized severalfundamental method about time series chaos recognition.And then, this dissertation has introduced the foundation of chaotictime series: the phase space restructuring theory, namely the time serieshaving chaotic character is rebuilded into one kind of low nonlineardynamics system.Phase space restructuring can find out the lawconcealing in evolution in chaotic attractor, causes the now availabledata to be integrated into the frame describing that kind, and provides anew method for the time series research.During the period ofrestructuring phase space, the appropriate delay time and the embeddingdimension selection are all-important. The main body of this dissertationis emphasized on the GP algorithm. In the GP algorithm, the fractalscaleless band and the fractal scaleless band accuracy usually need to be determined,which will affect the fractal dimension's accuracy directly.This chapter has carried out a quite thorough research on this question,and proposed one simple solution to the choice of the Fractal ScalelessBand.Finally, this dissertation has discussed the model and structure ofthe BP neural network and the BP study rule,constructed forecastingmodel for chaotic time series based on the BP neural network, studied theneural network problems such as scale and popularization and so on,andpredicted two concrete chaotic time series by making use of the BP neuralnetwork model. The results are extremely ideal and illustrates the chaoticTime Series Forecasting model based on neural network has very goodcapability of forecasting and popularization, and verified that the TimeSeries Forecasting model based on neural network has validity andgenerally application.
Keywords/Search Tags:chaos, Phase-space reconstruction, embedding dimension, Correlation Dimension, delay time, BP algorithm
PDF Full Text Request
Related items