面對當代時間序列資料高度非線性與多樣化的預測挑戰，本文提出一種結合量子物理概念與複數模糊邏輯系統之創新模型架構。本研究設計量子球型複數模糊集（Quantum Sphere Complex Fuzzy Sets, QSCFSs），以氫原子波函數為基礎，強化模型對複雜資料型態的處理能力，並整合模糊邏輯與神經網路架構以進行多目標預測分析。針對特徵過多的問題，本文採用基於影響資訊熵的多目標特徵選取機制，篩選具預測解釋力的輸入變數，兼顧資訊保留與模型簡化。於模型訓練方面，提出量子斑馬最佳化演算法（Quantum Zebra Optimization Algorithm, QZOA），融合量子位置更新策略與遞迴最小平方估計法（Recursive least squares estimator, RLSE），提升參數收斂速度與預測穩定性。實驗設計涵蓋國際著名指數與加密貨幣市場等多種金融時間序列，評估模型於單一與多重目標預測下之效能。實驗結果顯示，所提出模型於多項預測指標上表現優異，整體在資料維度壓縮、模型精準度與泛化能力上均優於傳統方法與近期文獻之混合模型。此研究驗證融合量子啟發式策略與複數模糊推理系統於處理高維時間序列資料之可行性，為時間序列預測領域提供一具理論深度與應用潛力的解決方案。;Facing the forecasting challenges arising from the high nonlinearity and diversity of contemporary time-series data, this paper proposes an innovative model architecture that integrates quantum physics concepts with complex fuzzy logic systems. This study designs the Quantum Sphere Complex Fuzzy Sets (QSCFSs), based on the hydrogen atom wave function, to enhance the model’s capability in handling complex data types. It further integrates fuzzy logic and neural network architectures to perform multitarget prediction analysis. To address the issue of excessively high amount of features, this paper adopts a multiarget feature selection mechanism based on influential information entropy to filter input variables with predictive interpretability, achieving a balance between information preservation and model simplification. In terms of model training, a Quantum Zebra Optimization Algorithm (QZOA) is proposed, which combines quantum position updating strategies with the Recursive Least Squares Estimator (RLSE) to accelerate parameter convergence and enhance prediction stability. The experimental design covers various financial time-series data, including internationally renowned indices and cryptocurrency markets, to evaluate the model’s performance under both single-target and multitarget forecasting scenarios. Experimental results indicate that the proposed model performs excellently across multiple prediction metrics, consistently outperforming traditional methods and recent hybrid models in data dimension reduction, model accuracy, and generalization capability. This study verifies the feasibility of integrating quantum-inspired heuristic strategies with complex fuzzy inference systems in handling high-dimensional time-series data, offering a solution with both theoretical depth and practical potential in the domain of time-series forecasting.
