面臨大數據時代，影響股票市場的各種因素使股票預測有些複雜和困難，準確的股票指數預測能幫助決策者採取正確的行動來發展更好的經濟。大多數傳統的時間序列模型在預測中只使用一個變數，多數使用收盤價格單一變數預測隔日收盤價格，而本研究所採用5個變數作為模型輸入，預測模型應該使用更多變數來提高預測的準確性，本研究提出一個以模糊類神經系統(Neuro-Fuzzy System, NFS)為架構，其結合Takagi-Sugeno模糊系統形成本研究模型，使其與傳統的類神經模型進行比較。在參數學習，以粒子群演算法(Particle Swarm Optimization, PSO)結合遞迴最小平方演算法(Recursive Least Squares Estimator, RLSE)，成為PSO-RLSE複合型演算法，進行參數的優化，發揮效用。本研究以三個實驗使用多種真實世界驗證模型的效能與研究理論，實驗一為台股指數預測與利潤計算，實驗二為恆生指數預測與利潤計算，實驗三為日經指數預測與利潤計算，實驗結果說明本研究模型在時間序列預測上有良好效能。;Faced with the era of big data, various factors affecting the stock market make stock forecasting which are complicated and difficult. In order to obtain accurate stock index forecasts, we hope to help decision makers take the right actions to develop a better economy. Most traditional time series models use only one variable in the forecast. Most use the single variable of the closing price to predict the closing price of the next day. In this study, three variables are used as input to the model. The forecasting model should use more variables to improve the forecast. This study proposes a Neuro-Fuzzy System (NFS) architecture that combines the Takagi-Sugeno fuzzy system to form the model structure of this study, which is compared with the traditional neural network model. In the parameter learning, Particle Swarm Optimization (PSO) combined with Recursive Least Squares Estimator (RLSE) is used as a PSO-RLSE composite algorithm to optimize parameters. This study used a variety of real-world data sets to validate the model′s efficacy and research theory in three experiments. The results of individual experiments are compared with the previous literature. The research results show that the model has good performance in time series prediction.
