結合類神經網路及多目標複演進全域優選法建立降雨－逕流預報模式之研究

A rainfall-runoff Model Based on the Neural Network Incorporating the Multi-objective Complex Evolution Global Optimization Method

陳儒賢(Lu-Hsien Chen)

《台灣首府大學學報》 1期 (2010/09) Pp. 149-167

https://doi.org/10.6607/BTSU.2010.1.7

摘要

本研究嘗試利用多目標複演進全域優選法（The Multi-objective Complex Evolution Global Optimization Method，簡稱MOCOM-UA）取代傳統倒傳遞類神經網路(back-propagation neural network)之參數檢定方法（最陡坡降法），針對以倒傳遞類神經網路為基礎之降雨－逕流預報模式之參數作檢定，進而建立一個即時的降雨－逕流預報模式。關於類神經網路之研究，已經有多種形型式之網路架構及學習方法被提出，在眾多類神經網路中，又以倒傳遞類神經網路最廣為被熟知及應用，但是倒傳遞類神經網路於訓練過程中會有許多問題，尤其是它容易收斂到區域最小值。由於MOCOM-UA結合controlled random search, competitive evolution, Pareto ranking和多目標最陡坡降演進法(multi-objective downhill simplex search)等四項優選策略，能快速且有效率地找出全域解(global solution)，其求解方法為平行地求出給定之柏拉圖解之數目，亦即不需如傳統方法求出所有目標值可能的分佈，使得演算次數大幅降低，所以大幅縮短求解時間。本研究將結合類神經網路及多目標複演進全域優選法所建立之降雨－逕流預報模式，應用於石門水庫集水區之流量預測上，分別預測領前二小時與領前四小時時刻之流量。經由四種評鑑指標計算之結果發現，本研究所建立之模式對於洪水流量之預測具有相當良好之效果，且較傳統使用最陡坡降法作參數檢定之類神經網路模式有較佳之預測精確度，因此建議可將此類神經網路應用於降雨－逕流預報上。

關鍵字

類神經網路 ； 多目標複演進全域優選法 ； 降雨－逕流預報模式

並列摘要

In this paper, the multi-objective complex evolution global optimization method (MOCOM-UA) is used to construct a rainfall-runoff model, and gradient steepest descent method is presented for the parametric estimation of the network. Regarding the flood forecasting, many studis used the back-propagation network model for river flood forecasting. However, the back-propagation network has some disadvantages. For example, it tends to yield local solutions, the learning rate is slow and the network structure is difficult to develop. The MOCOM-UA strategy combines the strengths ”controlled random search” with the ”competitive evolution”, Pareto ranking, and a newly developed strategy of multiobjective downhill simplex search, so it can find the global solution fast and efficiently. The proposed methodology is finally applied to an actual reservoir watershed to find the two and four-hour ahead runoff. Based on our study data, the results show that the proposed model can be successfully applied to build the relationship between rainfall and runoff. Moreover, the proposed network provides better training and testing accuracy than the traditional network calibrated by global solution does. Therefore, the proposed model is recommended as an alternative to the rainfall-runoff forecasting model.

並列關鍵字

artificial neural network ； multi-objective complex evolution global optimization method ； rainfall-runoff forecasting model