摘要
隨著國際貿易的頻繁，匯率對企業乃至
個人的影響與日俱增，為此本研究利用高斯
RBF 與BP 神經網路兩種預測模型分別對人
民幣、美元、日圓、港幣、韓圜進行預測，
並利用不同輸入期數（2 期、3 期、4 期、5
期、6 期），探討樣本數對於匯率預測的準確
度影響，以MATLAB 軟體執行BP、高斯
RBF 類神經網路的運算，我們得到多組預測
結果，將各組預測結果列表比較後，得各種
貨幣所對應最高準確度的預測結果。透過上
述的最適預測模型研究，獲得有效提高匯率
預測的效率與準確度的預測模式，並以此模
式作為協助掌握匯率變動的可靠依據，進而
減少決策錯誤，以達到避險與減少匯兌損失
的目的。
關鍵字：匯率預測、類神經網路、時間序列、
國際貿易
ABSTRACT
With the rapid progress of international
commerce, the influence of currency exchange
rate has become more and more significant for
enterprises or individual persons. Therefore,
this research aims at finding the optimal
forecasting models for foreign currency
exchange rates such as Renminbi, US dollar,
Japanese yen, Hong Kong dollar, and Korea
dollar based on Gaussian RBF and BP neural
networks. Under different input periods such
as two, three, four, five, or six periods,
forecasting accuracy is analyzed. The neural
networks are built in MATLAB environment.
According to the results obtained in this paper,
the optimal forecasting models for foreign
currency exchange rates are obtained. The
models can be used to avoid making wrong
decisions and to reduce exchange loss.
Keywords: Exchange Rate Forecast;
Artificial Neural Network;TimeSeries;
International commerce
|