非線性系統廣泛存在,幾乎所有實際系統(包括降雨-逕流歷程)都是非線性,因此研究非線性系統之模擬十分必要。因Volterra模式可視為非線性系統之通用表示式,其為一種多項式模式,且其輸出與核心之間為線性關係,因此,本研究採用非線性之二階Volterra模式來模擬非線性降雨-逕流歷程。另套配Kalman濾波以即時校正Volterra模式參數,可更進一步模擬非線性降雨-逕流歷程之時變特性。上述方法經應用於鹽水溪流域豐化橋上游集水區之降雨-逕流歷程分析,研究結果顯示,具時變性之二階Volterra模式,普遍優於非時變性之二階Volterra模式。此乃因為套配線性Kalman濾波於模式參數,可使其具有時變性,能進一步地模擬及反應非線性降雨-逕流歷程之時變特性,進而提高水文預報之精確度。
Almost all practical systems, including rainfall-runoff processes, are nonlinear. The modeling of a nonlinear system must be studied. The Volterra model, which is one kind of the polynomial models, offers an important general representation of a nonlinear time-invariant system. One benefit of Volterra filters is that the output of the systems is linearly related to the Volterra kernels. The nonlinear second-order Volterra model is adopted for approximating nonlinear rainfall-runoff processes herein. Kalman filters are then utilized to online estimate the parameters of the Volterra kernels and thus model the time-varying nonlinear rainfall-runoff processes. To demonstrate the feasibility of applying Kalman filters and the Volterra model to the modeling of rainfall-runoff processes, the Feng-Hua Bridge watershed in the Yan-Shui River basin is selected as the study area. The average of validation results indicates that the developed time-varying second-order Volterra model outperforms the time-invariant one, because the time-variation of parameters is associated with the adoption of a Kalman filter, and thus simulate and respond to the time-variation of nonlinear rainfall-runoff processes. Validation results also reveal that the resulting method improves the accuracy of hydrological forecasting.