改良式類神經網路方法於水文系統之研究

Study on Improved Neural Network Approaches in Hydrosystem

陳谷榕(Guo-Rong Chen)

指導教授 ： 林國峰

國立臺灣大學/工學院/土木工程學研究所/博士(2005年)

https://doi.org/10.6342/NTU.2005.01793

摘要

類神經網路（artificial neural network）模擬水文過程（hydrological process）的潛力已經被大量的應用實例所肯定，然而，由於大部份的類神經網路模式缺乏物理機制（physical mechanisms），因此在某些水文問題的應用上遭致失敗。此外，傳統上使用試誤法（trial and error procedure）來建構類神經網路，不但相當耗時，使用上也不方便。所以類神經網路在水文問題的應用亟需一套能夠提升效能的方法，而本論文的目的就在於建立一套有效的方法，使得類神經網路效能提升。
本論文提出兩個概念，第一個概念是根據已知的物理機制來設計類神經網路，而第二個概念是只用高度相關的輸入項來建構類神經網路。在第二章與第三章中，吾人以第一個概念來設計倒傳遞類神經網路（back-propagation neural networks）與幅射基底函數網路（radial basis function networks），並將其應用在地下水含水層參數檢定之問題。吾人所設計的改良式類神經網路係根據已知的物理機制來設計，因此與現有的類神經網路最大的不同就在於輸入項與輸出項的設計。根據1000組隨機資料的測試結果，改良式類神經網路之效能比現有類神經網路更加卓越。
在第四章中，吾人則以第二個概念來建立降雨-逕流類神經網路模式。為了只保留高度相關的輸入項，因此本文提出一套系統化的方法來消除不相關的輸入項。本方法所建構的降雨-逕流類神經網路模式成功地應用於翡翠水庫集水區，其應用結果亦顯示本方法比傳統上所使用的試誤法更具優點，因此本方法對於建立降雨-逕流類神經網路模式有很大的助益。

關鍵字

降雨-逕流模式 ； 地下水含水層參數檢定 ； 幅射基底函數網路 ； 倒傳遞類神經網路 ； 類神經網路

並列摘要

Artificial neural networks (ANNs) have found increasing applications in various aspects of hydrology and previous studies have shown the potential of ANNs for modeling hydrological processes. However, ANN models failed to be applied to some hydrological problems, because the ANN architectures are usually lack of physical mechanisms. In addition, the ANN models were constructed by a trial and error procedure, which requires amount of time. Hence applications of ANNs in hydrology cry for approaches to the construction of ANN models, which are capable of improving the performance of ANN models. The object of this thesis is to establish effective approaches to the construction of ANN models in different problems of water resources and hydrology.
In this thesis, two concepts for constructing ANN models in hydrology are presented. The first concept is to construct ANN models based on known physical mechanisms and the second concept is to construct adequate ANN models only included highly relevant inputs. In Chapters 2 and 3, two ANN approaches, Back-propagation neural networks (BPNs) and radial basis function networks (RBFNs) approaches, based on the first concept are established to determine aquifer parameters from pumping test data. The major difference between the existing and the proposed ANN approaches is the design of ANN input and output components. The proposed ANNs are designed according to the analytical solutions, which express known physical mechanisms. Testing the existing and the proposed ANN approaches by 1000 sets of synthetic data demonstrates that our design of ANNs is better than the existing ANN approach.
In Chapter 4, a systematic approach based on the second concept is used to construct ANN rainfall-runoff models. In order to construct adequate ANN models only included highly relevant inputs, the irrelevant inputs will be trimmed by the systematic approach. An application to the Fei-Tsui Reservoir Watershed in northern Taiwan shows that the proposed ANN rainfall-runoff model has advantages over those obtained by the trial and error procedure. The proposed approaches will be helpful to hydrologist to construct adequate ANN-based hydrological models.

並列關鍵字

Artificial neural networks ； The determination of aquifer parameters ； Rainfall-runoff model ； Back-propagation neural networks ； Radial basis function networks

參考文獻

1.Aziz, A.R.A., Wong, K.V., 1992. A neural-network approach to the determination of aquifer parameters. Ground Water. 30(2), 164-166.

2.Balkhair, K.S., 2002. Aquifer parameters determination for large diameter wells using neural network approach. Journal of Hydrology. 265, 118-128.

4.Chang, F.J., Liang, J.M., Chen, Y.C., 2001. Flood Forecasting Using Radial Basis Function Neural Networks. IEEE Transactions on Systems, Man, and Cybernetics—PART C: Applications and Reviews. 31(4), 530-535.

5.Chang, L.C., Chang, F.J., 2001. Intelligent control for modelling of real-time reservoir operation. Hydrological Process. 15, 1621-1634.

6.Chen, C.S., Chang, C.C., 2002. Use of cumulative volume of constant-head injection test to estimate aquifer parameters with skin effects: Field experiment and data. Water Resources Research. 38(5), 10.1029/2001WR000300.