以氫量子複數神經模糊方法進行多目標預測之研究;A Study on Multitarget Prediction Using a Hydrogen Quantum Complex Neuro-Fuzzy Approach

NCUIR > School of Management at National Central University > Graduate Institute of Information Management > Electronic Thesis & Dissertation > Item 987654321/98275

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/98275

Title:	以氫量子複數神經模糊方法進行多目標預測之研究;A Study on Multitarget Prediction Using a Hydrogen Quantum Complex Neuro-Fuzzy Approach
Authors:	孫有年;Sun, Yu-Nien
Contributors:	資訊管理學系
Keywords:	多目標預測;量子球型複數模糊集;氫量子鯨群優化演算法;模糊類神經網路;時間序列;multi-target forecasting;quantum sphere complex fuzzy sets;Hydrogen quantum whale optimizer;neuro-fuzzy network;time series
Date:	2025-07-14
Issue Date:	2025-10-17 12:34:21 (UTC+8)
Publisher:	國立中央大學
Abstract:	本研究提出一種創新的智慧型混合預測模型—量子球型複數類神經模糊推理系統（Quantum Sphere Complex Neuro-Fuzzy Inference System, QSCNFIS）。QSCNFIS結合兩項新技術：其一為量子球型複數模糊集（Quantum Sphere Complex Fuzzy Sets, QSCFSs），透過量子概念改良球型複數模糊集（Sphere Complex Fuzzy Sets, SCFSs），以氫原子波函數建構多維模糊隸屬函數，有效提升模型表現；其二為氫量子鯨群優化演算法（Hydrogen Quantum Whale Optimizer, HQWO），該方法結合氫原子波函數中的機率分布特性，改良傳統鯨群最佳化演算法（Whale Optimization Algorithm, WOA）的搜尋與收斂能力。此外，模型亦融合遞迴最小平方估計法（Recursive Least Squares Estimation, RLSE），加速T-S模糊規則中線性參數的訓練效率。為驗證所提方法之有效性，本研究設計三組預測實驗，涵蓋單目標、雙目標與四目標之時間序列任務，所採用資料包含主要金融指數（TAIEX、HSI、N225、DJI、FTSE100、S&P500、KOSPI200）。實驗結果顯示，QSCNFIS在預測準確性與穩定性方面均優於傳統與現有文獻方法，驗證所提方法具備高度彈性與良好擴展性，適用於處理多樣且高維的預測問題。;This study proposes an innovative intelligent hybrid forecasting model—Quantum Sphere Complex Neuro-Fuzzy Inference System (QSCNFIS). QSCNFIS integrates two novel techniques: (1) the Quantum Sphere Complex Fuzzy Sets (QSCFSs), which enhance the traditional Sphere Complex Fuzzy Sets (SCFSs) by incorporating quantum concepts. Specifically, multidimensional fuzzy membership functions are constructed using hydrogen atom wave functions, effectively improving model representation capability; and (2) the Hydrogen Quantum Whale Optimizer (HQWO), which leverages the probabilistic characteristics of hydrogen wave functions to refine the search and convergence behavior of the traditional Whale Optimization Algorithm (WOA). In addition, the model incorporates the Recursive Least Squares Estimation (RLSE) to accelerate the training of linear parameters in Takagi–Sugeno (T-S) fuzzy rules. To validate the effectiveness of the proposed approach, three forecasting experiments were designed, targeting single, dual, and multitarget time series prediction tasks. The datasets include major financial indices such as TAIEX, HSI, N225, DJI, FTSE100, S&P500, and KOSPI200. Experimental results demonstrate that QSCNFIS outperforms traditional and existing methods in both prediction accuracy and stability, verifying that the proposed approach possesses high flexibility and good scalability, making it suitable for handling diverse and high-dimensional prediction problems.
Appears in Collections:	[Graduate Institute of Information Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	30	View/Open

社群 sharing

Loading...