本論文以獨立元素分析技術與時間序列分析中的自我回歸模型以建構時間序列預測 模型。與傳統的自我回歸移動平均模型所不同的是,以輸入的時間序列所解開的內 部因素序列作為我們預測模型的基準。 我們的預測模型中有元素混淆性、時間序列相關性與平均數差異效應等三大主 要不利於時間序列預測的情況。本論文也提出相對應的解決方案以克服預測困難。 在線性時間複雜度的運算時間內,我們所提出的解決方案可以有效地解決此三大問 題而達成預測的目的。
Building a time series forecasting model by independent component analysis mechanism presents in the paper. Different from using the time series directly with the traditional ARIMA forecasting model, the underlying factors extracted from time series is the forecasting base in our model. Within component ambiguity, correlation approximation and mean difference problems, independent component analysis mechanism has intrinsic limitations for time series forecasting. Solutions for those limitations were purposed in this paper. Under the linear time complexity, the component ambiguity and mean-difference problem was solved by our proposed evaluation to improve the forecasting reward. The empirical data show that our model exactly reveals the flexibility and accuracy in time series forecasting domain.