本文係應用小波混合方法於降雨量及逕流量之模擬，以提升模擬之精確性。首先應用小波分解將降雨量及逕流量時間序列分解成趨勢項、週期項與隨機項。趨勢項採用倒傳遞模式之類神經網路進行擬合，而週期項則根據小波之類週期特性，分析並挑選出合適之週期，結合逐步最小二乘法，來估計以正弦及餘弦函數表示之週期項近似表達式。最後將趨勢項與週期項之推估值合併，可獲致原始降雨量及逕流量時間序列之推估值。為驗證倡議方法之合適性，本文應用小波混合方法於月降雨量及月逕流量時間序列。研究結果說明該方法比直接採用倒傳遞模式之類神經網路進行擬合之平均誤差小，闡釋小波混合方法之有效性，臻以提供水資源規劃應用之參考。

In this paper, wavelet mixed method was applied to the modeling of rainfall and runoff to enhance the accuracy. Firstly, the wavelet decomposition was applied to decompose the rainfall and runoff time series into the tendency, periodic and stochastic terms. Then, the tendency term can be fitted using the Back-Propagation (BP) Network model of the Artificial Neural Network. The appropriate periods were chose in accordance with the likely periodic property of wavelet in the periodic term. The approximate expression of the periodic term of the hydrological time series with sine and cosine functions can be obtained by using Ordinary Least Square method step by step. Finally, the estimations of tendency and periodic term can be combined into the estimation of original rainfall and runoff time series. To verify the appropriateness of the model adopted, the wavelet mixed method was applied to the modeling of the monthly rainfall and runoff time series in this paper. The results illustrate that the mean relative error obtained by the proposed method is less than that obtained by directly BP Network model. The results also show that the wavelet mixed method can enhance the accuracy of modeling of rainfall and runoff time series, so as to provide the reference for water resource planning and applications.