Forecasting Methods Based on Seasonal Chaotic Time Series
碩士 === 東海大學 === 工業工程與經營資訊學系 === 103 === Chaos is one type of systems with structure and regularities. On appearance, it may look disorderly. Yet eventually it will appear to return to equilibrium and wait for the next disorder, cyclically. Characteristics of chaotic systems thus include: 1. changea...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | zh-TW |
Published: |
2015
|
Online Access: | http://ndltd.ncl.edu.tw/handle/53967643670979226481 |
id |
ndltd-TW-103THU00030024 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-TW-103THU000300242016-08-19T04:10:36Z http://ndltd.ncl.edu.tw/handle/53967643670979226481 Forecasting Methods Based on Seasonal Chaotic Time Series 以季節性混沌時間序列為基礎之預測方法研究 Yuan-Chih Wang 王元志 碩士 東海大學 工業工程與經營資訊學系 103 Chaos is one type of systems with structure and regularities. On appearance, it may look disorderly. Yet eventually it will appear to return to equilibrium and wait for the next disorder, cyclically. Characteristics of chaotic systems thus include: 1. changeable, nonlinear, 2. random-look disorderly yet with certain regularity order, and 3. sensitive to initial conditions. Thus, grasping and understanding such systems may unfold research-worthy directions for development of forecasting systems. In recent years along with the developments of related forecasting-models, researches and applications in various areas, mutual combinations of various theories also boost the forecast accuracy. How to capture the chaotic-phenomena resultant variations and the related researches thereby increasing the forecasting accuracy also have been proposed. And forecasting methods based on the chaotic approach also have been introduced such as full-range and local approximation methods of reconstructed phase space based on the embedding theorem, largest Lyapunov exponent (LLE) approach, weighted largest Lyapunov exponent (WLLE) approach, and that combines with various artificial neural networks (such as support vector regression (SVR), back-propagation neural network (BPN), radial basis function neural network (RBFNN), etc.). All of them have applied areas and good performance. Yet further improvements also exist. In this research, therefore the aforementioned, or in recent years, researches regarding the investigation of forecasting result distortion resulting from chaotic phenomena and how to analyze and capture such variations for reducing forecasting errors, would be incorporated. Besides, the research will focus on another variation factor that causes forecasting result distortion in some type of data: seasonal variation. Seasonal variation has the characteristics of regularity and cyclic repeats. So far, related research regarding combinations of chaotic forecasting and seasonal variation is still very few. But, in addition to variations of chaotic phenomena, seasonality effects on forecasting are also an essential and non-ignorable variation. Thus, this research observes that there are affect result by different data processing and methods. Therefore, this research will propose and investigate forecasting systems, which can be classified into five types. Based on that, different data processing and methods. (I) Not doing the data processing before implement the forecasting methods. (II) Only separate seasonal index. (III) Only doing the chaotic reconstructed-phase-space. (IV) Adjust seasonal index then reconstructed-phase-space. (V) Reconstructed-phase-space and execute forecasting methods before adjust seasonal index. Also, the research results will be executed with the comparisons of the proposed methods with other methods (such as conventional seasonal data forecasting method (the decomposition method), different ANN-type forecasting methods (such as SVR, GRNN, RBFNN, BPN, Elman, NARX neural networks, etc.) especially designed and applied for seasonal data forecasting, chaotic forecasting, seasonal autoregressive integrated moving average (SARIMA). In addition, this research will investigate data from different areas (such as energy demand, traffic flow, tourist arrival, raw material output, public database for forecasting, etc.) for validation of the proposed forecasting models. Presently preliminary experiment indicates that in most situations this research’s methods can have good performance. Ping-Teng Chang Tsueng-Yao Tseng 張炳騰 曾宗瑤 2015 學位論文 ; thesis 94 zh-TW |
collection |
NDLTD |
language |
zh-TW |
format |
Others
|
sources |
NDLTD |
description |
碩士 === 東海大學 === 工業工程與經營資訊學系 === 103 === Chaos is one type of systems with structure and regularities. On appearance, it may look disorderly. Yet eventually it will appear to return to equilibrium and wait for the next disorder, cyclically. Characteristics of chaotic systems thus include: 1. changeable, nonlinear, 2. random-look disorderly yet with certain regularity order, and 3. sensitive to initial conditions. Thus, grasping and understanding such systems may unfold research-worthy directions for development of forecasting systems.
In recent years along with the developments of related forecasting-models, researches and applications in various areas, mutual combinations of various theories also boost the forecast accuracy. How to capture the chaotic-phenomena resultant variations and the related researches thereby increasing the forecasting accuracy also have been proposed. And forecasting methods based on the chaotic approach also have been introduced such as full-range and local approximation methods of reconstructed phase space based on the embedding theorem, largest Lyapunov exponent (LLE) approach, weighted largest Lyapunov exponent (WLLE) approach, and that combines with various artificial neural networks (such as support vector regression (SVR), back-propagation neural network (BPN), radial basis function neural network (RBFNN), etc.). All of them have applied areas and good performance. Yet further improvements also exist.
In this research, therefore the aforementioned, or in recent years, researches regarding the investigation of forecasting result distortion resulting from chaotic phenomena and how to analyze and capture such variations for reducing forecasting errors, would be incorporated. Besides, the research will focus on another variation factor that causes forecasting result distortion in some type of data: seasonal variation. Seasonal variation has the characteristics of regularity and cyclic repeats. So far, related research regarding combinations of chaotic forecasting and seasonal variation is still very few. But, in addition to variations of chaotic phenomena, seasonality effects on forecasting are also an essential and non-ignorable variation. Thus, this research observes that there are affect result by different data processing and methods.
Therefore, this research will propose and investigate forecasting systems, which can be classified into five types. Based on that, different data processing and methods. (I) Not doing the data processing before implement the forecasting methods. (II) Only separate seasonal index. (III) Only doing the chaotic reconstructed-phase-space. (IV) Adjust seasonal index then reconstructed-phase-space. (V) Reconstructed-phase-space and execute forecasting methods before adjust seasonal index. Also, the research results will be executed with the comparisons of the proposed methods with other methods (such as conventional seasonal data forecasting method (the decomposition method), different ANN-type forecasting methods (such as SVR, GRNN, RBFNN, BPN, Elman, NARX neural networks, etc.) especially designed and applied for seasonal data forecasting, chaotic forecasting, seasonal autoregressive integrated moving average (SARIMA). In addition, this research will investigate data from different areas (such as energy demand, traffic flow, tourist arrival, raw material output, public database for forecasting, etc.) for validation of the proposed forecasting models. Presently preliminary experiment indicates that in most situations this research’s methods can have good performance.
|
author2 |
Ping-Teng Chang |
author_facet |
Ping-Teng Chang Yuan-Chih Wang 王元志 |
author |
Yuan-Chih Wang 王元志 |
spellingShingle |
Yuan-Chih Wang 王元志 Forecasting Methods Based on Seasonal Chaotic Time Series |
author_sort |
Yuan-Chih Wang |
title |
Forecasting Methods Based on Seasonal Chaotic Time Series |
title_short |
Forecasting Methods Based on Seasonal Chaotic Time Series |
title_full |
Forecasting Methods Based on Seasonal Chaotic Time Series |
title_fullStr |
Forecasting Methods Based on Seasonal Chaotic Time Series |
title_full_unstemmed |
Forecasting Methods Based on Seasonal Chaotic Time Series |
title_sort |
forecasting methods based on seasonal chaotic time series |
publishDate |
2015 |
url |
http://ndltd.ncl.edu.tw/handle/53967643670979226481 |
work_keys_str_mv |
AT yuanchihwang forecastingmethodsbasedonseasonalchaotictimeseries AT wángyuánzhì forecastingmethodsbasedonseasonalchaotictimeseries AT yuanchihwang yǐjìjiéxìnghùndùnshíjiānxùlièwèijīchǔzhīyùcèfāngfǎyánjiū AT wángyuánzhì yǐjìjiéxìnghùndùnshíjiānxùlièwèijīchǔzhīyùcèfāngfǎyánjiū |
_version_ |
1718378720255803392 |