| DC 欄位 |
值 |
語言 |
| DC.contributor | 資訊管理學系 | zh_TW |
| DC.creator | 鄭淨文 | zh_TW |
| DC.creator | Ching-wen Cheng | en_US |
| dc.date.accessioned | 2012-7-22T07:39:07Z | |
| dc.date.available | 2012-7-22T07:39:07Z | |
| dc.date.issued | 2012 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=994203001 | |
| dc.contributor.department | 資訊管理學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 本論文提出了輸入變數選擇(Input selection)演算法及SPSO-RLSE複合式學習演算法於複數類神經模糊系統之建模上，並將本論文方法應用於時間序列預測。複數類神經模糊系統(Complex neuro-fuzzy system, CNFS)是將複數模糊系統結合於類神經網路的一種計算模型。複數模糊集利用複數來描述隸屬程度，能增加CNFS的彈性與映射能力，在處理非線性問題如時間序列預測有良好的能力。在時間序列預測的研究中，輸入變數選擇為一重要的議題，透過輸入變數選擇，我們能找出變數之間的相依性，進而篩選出有用與重要的資訊，改善預測的準確度。本論文將實作輸入變數選擇於複數類神經模糊系統上，探討輸入變數選擇在時間序列預測中的影響。在模型的學習上，本論文使用了SPSO-RLSE複合式學習法，其將標準粒子群演算法(Standard particle swarm optimization, SPSO)用於更新系統的前鑑部參數，以及遞迴最小平方估計法(Recursive least squares estimation, RLSE)用於後鑑部參數之更新，能夠在訓練時快速的將模型最佳化已達精準預測。本論文用了5個不同的實驗來檢驗本論文方法在時間序列預測上的表現，每個實驗中都根據輸入變數選擇的結果設計了不同的輸入變數做比較以觀察輸入變數選擇的影響。實驗結果皆呈現我們提出的方法其預測準確度優於其他比較文獻的方法，證實本論文的方法在時間序列預測上有良好的表現。
| zh_TW |
| dc.description.abstract | In this thesis, a new modeling approach has been presented, where the modeling theory of complex neuro-fuzzy system (CNFS), a Brock-Dechert-Scheinkman (BDS) based method for input selection and a SPSO-RLSE hybrid learning method for the parameter estimation of CNFS are used in the study for the problem of time series forecasting. For CNFS, complex fuzzy sets (CFSs) are embedded in the neuro-fuzzy model structure to enhance the flexibility in adaption and the non-linear ability of input-output mapping that is good for non-linear problems such as time series forecasting. A CFS is an advanced fuzzy set whose membership degrees are complex-valued and defined in the unit disc of the complex plane. For modeling and forecasting, input selection is very important, for which a variable-dependability index is used. By the results of such indices, we can find out the variable dependencies, so to select dependent variables that are related to the target variable. Afterwards, for the modeling of CNFS the SPSO-RLSE hybrid learning method applies, where the method of standard particle swarm optimization (SPSO) is used to adjust the premise parameters of CNFS and the method of recursive least squares estimation (RLSE) is used to update the consequent parameters of CNFS. This hybrid learning method can rapidly optimize the CNFS to get accurate prediction. Five examples for time series forecasting were used in the study to examine the proposed approach, whose results are compared with those by other approaches. Through the experimental results, the proposed approach shows very promising performance and outperforms the compared approaches.
| en_US |
| DC.subject | 複數模糊集 | zh_TW |
| DC.subject | 複數類神經模糊系統 | zh_TW |
| DC.subject | 標準粒子群最佳化演算法 | zh_TW |
| DC.subject | 時間序列預測 | zh_TW |
| DC.subject | 輸入變數選擇 | zh_TW |
| DC.subject | 遞迴最小平方估計法 | zh_TW |
| DC.subject | complex fuzzy set (CFS) | en_US |
| DC.subject | complex neuro-fuzzy system (CNFS) | en_US |
| DC.subject | standard particle swarm optimization (SPSO) | en_US |
| DC.subject | recursive least squares estimator (RLSE) | en_US |
| DC.subject | time series forecasting. | en_US |
| DC.subject | input selection | en_US |
| DC.title | 變數選擇在智慧型系統與應用之研究 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | A Study on Input Selection for Intelligence Systems and Applications | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |