傳統上,模糊規則庫之規則數,會隨著輸入樣本數目與模糊區間的增加,而呈指數的增加,進而增加了模糊規則庫推理的時間及降低預測的準確度。為了改善此缺點,本研究期望藉由因子的篩選找出最佳的技術指標組合,基於K-means分群技術,建立精簡化的模糊規則庫預測模式(SFR),減少規則數的產生,有效的達到精簡之目的,而規則的參數部分則使用模擬退火法來進行最適化調整,最後將建立好的預測模型運用於股價的預測上,針對大盤及個股之收盤價進行預測。顯示本研究所建構之SFR預測模式在績效評比方面,皆比GANN與K-GAWM兩方法有著較佳的預測效果。以大盤為例,SFR預測模式平均MAPE值可達0.038。由此可知本研究所提出之SFR預測模式優於其他兩種預測模式,且較適用於台灣股價之預測。為了探討模式在實際應用的可行性,本研究嘗試使用三種案例式推演方法來預測技術指標值,並代入SFR預測模式中。3種方法中以方法3之二階段資料處理法有較佳的預測結果,在相對誤差方面,四個探討案例中以大盤的預測結果較佳,其相對誤差約在2%以內。
Traditionally, fuzzy inference system has been used in a wide variety of applications. However, with the number of data and fuzzy sets increasing, there often exist redundant rules or similar fuzzy sets in system. This results in unnecessary structural complexity and decreases the interpretability of the system. In this paper, a rule base simplification method (SFR) is proposed to establish interpretable fuzzy models from numerical data. After identifying the key variables by stepwise regression analysis, K-means clustering technique is used to simplify and find out the optimal number of rule base. When the model is constructed, SA is used to tune the consequent parameters of the rules. The result indicates that the SFR model has better performance than GANN and K-GAWM. Take TSE (Taiwan Stock Exchange) for example, the MAPE attains 0.038. To investigate the feasibility of SFR model, three methods are proposed to predict key variables and combined with SFR model for forecasting. The experimental results show that method 3 is better than others. In four cases, TSE has better performance and the relative error is under 2%.