類免疫系統(Artificial Immune System, AIS)是人工智慧(artificial intelligence, AI)領域中新興的方法,近年已在許多領域上取得良好的成果。本論文對AIS進行了兩個研究,在第一個研究上,我們針對分離未知混合訊號的獨立成分分析(Independent Component Analysis, ICA)法有著易陷入區域解和分離訊號順序不確定性的問題,我們藉著AIS的特性,提出了AISICA演算法來克服這些缺點。而在第二個研究上,我們對於AI預測工具:支援向量迴歸(Support Vector Regression, SVR)建立預測模式的參數搜尋不易,以及投資者在股價預測上希望同時兼顧較小的數值誤差和較高的漲跌趨勢預測正確,因此我們以AIS為基礎,提出AIS^2 VR兩階段多目標搜尋法來改良參數搜尋並滿足股價預測需求。在AISICA實證上,經過以合成、實際資料實驗並和常用FastICA法比較後,顯示AISICA的分離能力更佳,且分離訊號順序有著固定性。而在透過台灣、美國、香港及上海等四個股票市場的實際資料進行實證研究後,結果顯示AIS^2 VR兩階段多目標搜尋法不僅在參數搜尋上比一般常用的Grid-Search 方法更好,並且在股價預測上能夠在確保較低的數值誤差下兼具較高的漲跌趨勢預測正確率。總結以上,本研究成功地以AIS改善ICA和SVR所遭遇的問題,並增進其效能。
Artificial Immune System is a novel technology in the field of artificial intelligence (AI), and it have been successfully employed in a wide variety of different application areas in recent years. This paper contains two studies. Independent Component Analysis (ICA) is able to recover a set of unknown mutually independent components (ICs) from their observed mixtures without knowledge of the mixing coefficients, but ICA is easily converge to local optimal solutions and the order of recovered source signals is unpredictable. In the first study, we proposed AISICA algorithms to overcome these drawbacks. SVR is a new learning algorithm based on statistical learning theory, which has a good time-series forecasting ability. But parameters determination for a SVR model is competent to the forecasting accuracy. Furthermore, investors need the stock price prediction with lower numerical error and higher accuracy of trend forecasting. For those reasons, we proposed the two-stage multi-objective AIS^2 VR search method to meet the neesd. Experimental results have shown that all of the methods we proposed, by comparing with general methods, like FastICA or Grid Search, are efficient ways to solve the problems of ICA and SVR, and increase their efficiency.