改良式倒傳遞類神經網路於水庫入流量預報之研究

Improved back-propagation networks for reservoir inflow forecasting

鄭家豪(Jia-Haur Jeng)

指導教授 ： 林國峰

國立臺灣大學/工學院/土木工程學研究所/碩士(2008年)

https://doi.org/10.6342/NTU.2008.01346

摘要

效率對於類神經網路（Neural Networks）而言是一個很重要的議題，但是在水文領域卻很少受到重視。採用誤差倒傳遞演算法（Error Back-Propagation Algorithm）進行訓練的倒傳遞類神經網路（Back-Propagation Networks）是最常被使用的一種類神經網路，然而訓練過程卻相當費時。為求提升效率，本研究提出一種改良式倒傳遞類神經網路，而改良式倒傳遞類神經網路採用一種新型的詢問式學習法（Query Learning Approach）進行訓練。詢問式學習法可以由訓練資料中選取有用的資訊，使改良式倒傳遞類神經網路能僅以部分訓練資料完成訓練，進而大大縮短訓練時間。為了展示改良式倒傳遞類神經網路的優勢，吾人以翡翠水庫為研究案例，分別應用改良式倒傳遞類神經網路及傳統倒傳遞類神經網路架構水庫入流量預報模式，並針對兩種類神經網路模式的預報結果進行比較。研究結果顯示兩種類神經網路模式的預報表現幾乎相同，但改良式倒傳遞類神經網路模式卻很明顯地縮短了訓練時間。與傳統倒傳遞類神經網路模式相比，改良式倒傳遞類神經網路模式僅需要50%的訓練時間。基於效率的觀點，改良式倒傳遞類神經網路模式可成為現有倒傳遞類神經網路模式的替代模式。

關鍵字

類神經網路 ； 倒傳遞類神經網路 ； 詢問式學習法 ； 水庫入流量預報

並列摘要

The efficiency is an important issue for neural networks-based models, but the issue has received little attention in the hydrologic domain. Back-propagation networks (BPNs) are the most frequently used convectional neural networks (NNs). However, BPNs are trained by the error back-propagation algorithm which is a very time-consuming iterative process. To improve the efficiency, improved BPNs which are trained by a novel query learning approach are proposed. The proposed query learning approach is capable of selecting informative data from all training data. Then the improve BPNs can be efficiently trained with partial data. An application is conducted to demonstrate the superiority of the improved BPNs. Two kinds of BPN-based (the improved and the conventional BPN-based) reservoir inflow forecasting models are constructed and the comparison between the improved and the conventional BPN-based model is made. The results show that the performance of the improved BPN-based models is as good as that of the conventional BPN-based models, but the improved BPN-based models significantly required less training time than the conventional BPN-based models. As compared to the conventional BPN models, only about 50% of training time is required for the improved BPN-based models. The improved BPN-based models are recommended as an alternative to the existing models because of their efficiency.

並列關鍵字

neural networks ； back-propagation neural networks ； query learning approach ； reservoir inflow forecasting

參考文獻

1. Bae, D.H., Jeong, D.M., Kim, G., 2007. Monthly dam inflow forecasts using weather forecasting information and neuro-fuzzy technique. Hydrological Sciences Journal 52(1), 99-113.

2. Campolo, M., Andreussi, P., and Soldati, A., 1999. River flood forecasting with a neural network model. Water Resources Research 35(4), 1191-1197.

3. Chang, F.J., Chang, Y.T., 2006. Adaptive neuro-fuzzy inference system for prediction of water level in reservoir. Advances in Water Resources 29(1), 1-10.

4. Chang, Y.T., Chang, L.C., Chang, F.J., 2005. Intelligent control for modeling of real-time reservoir operation, part II: artificial neural network with operating rule curves. Hydrological Processes 19 (9), 1825–1837.

5. Chaves, P., Kojiri, T., 2007a. Deriving reservoir operational strategies considering water quantity and quality objectives by stochastic fuzzy neural networks. Advances in Water Resources 30(5), 1329–1341.