不同模式时距对模式预测能力影响之研究.PDF

不同模式時距對模式預測能力影響之研究 (1) (2) (2) 陳鴻烈 蔡大偉 胡慧蘭 摘 要 本研究是以德基水庫為研究對象，利用時間序列模式來預測其未來的優養化變化趨勢，協 助該區優養化問題的經營管理。研究分兩部分來說明各組模式的變化，第一部分是利用三種由 傳統時間序列模式演變而成的預測模式，依據不同的「模式時距」來做優養化預測分析。第二 部分則是選用三種模式評比方法來檢視各個模式最佳時距的選擇。研究顯示，以虛擬變數法進 行預測時，其最佳模式時距為「14 年」，而相加法與相乘法則同為「8 年」。其中又以相加法 的「8 年」來作為模式時距時，會得到最佳結果，此時預測值和觀測值最為接近，亦即能呈現 水庫未來最真實的優養化變化趨勢。 關鍵詞( ：模式預測、時距選擇、優養化) Influence of Model Time Periods on Model Predictive Ability Paris Honglay Chen Professor, Department of Soil and Water Conservation, National Chung-Hsing University, Taichung, Taiwan 402, R.O.C. David D-W. Tsai Hui-Lan Hu Graduate Student, Department of Soil and Water Conservation, National Chung-Hsing University, Taichung, Taiwan 402, R.O.C. ABSTRACT This research was to predict the trophic status of Te-Chi reservoir by the time series model. We hope the results could enhance and help the management of eutrophication in reservoir. The study was divided into two parts to explain the variations of each model. First part was to predict the eutrophication conditions according to the different model time periods of three prediction models derived from traditional time series model. Second part was to find out the best model time periods of each model by three model evaluation methods. The results showed that the best model time periods was “14 years” in dummy variable model, as well as “8 years” in additive and multiplicative model. Among them we noticed that the best results were obtained from model time periods of additive (1)國立中興大學水土保持學系教授 (2)國立中興大學水土保持學系碩士班研究生 141 水土保持學報 3