建立河川水質之推估模式與最佳水質控制策略

Constructing an Estimation Model and the Optimal Control Strategy on Water Quality in Urban River Basin

蔡宇軒(Yu-Hsuan Tsai)

指導教授 ： 張斐章

國立臺灣大學/生物資源暨農學院/生物環境系統工程學研究所/碩士(2013年)

https://doi.org/10.6342/NTU.2013.01105

摘要

大漢溪流域下游流經桃園、新北市區河段，過多的汙水排放造成水質迅速惡化，然而台灣山勢陡峻，降雨量分佈不均，加上特有的颱風暴雨等季節性氣候特徵，驟降的雨量導致河川水文環境變化劇烈，使得受汙染的河水中各水質因子變動幅度極大，此現象易造成河川生態衝擊。穩定河川水質變化程度一直是環境保育的重要課題之一，因此，本研究藉由探討各河川水質因子之環境特性，尋找此地區最具代表性之水質因子，並藉由廣泛領域可能之影響因素建立其推估模式，進而透過控制河川流量與制定水質目標達到控管水質的目的。

本研究針對石門水庫大壩下游大漢溪流域數個水質測站，探討各水質因子之相互關係和空間分佈特性，以及水質因子與雨量和放流量間之相關性，分析結果推論氨氮為眾多河川水質因子中重要且易被放流量控制之關鍵因子。隨後利用Gamma Test統計分析，從不同環境領域中萃取出影響河川氨氮之代表性因子組合，以多種類神經網路尋找最合適之推估模式，結果顯示具有輸出層回饋項的非線性自回歸類神經網路(NARX)，針對河川氨氮之推估結果最佳。在水質控管部分，利用訓練好之NARX網路搭配連續方程式與邊界條件建立控制模式，求取鳶山堰在不影響正常操作下，使氨氮低於設定之水質標準之最佳放流操作，提供給上游石門水庫在汛期與非汛期的放流參考。結果顯示透過此程序，在25%和50%削減量下，皆可以有效降低河川中氨氮之濃度與變異程度，達成穩定河川水質指標的目的，維持整體河域的生態穩定性。

關鍵字

水質 ； 河道管理 ； 氨氮 ； 非線性自回歸類神經網路 ； Gamma 檢定

並列摘要

Water quality stabilization in rivers is an important task to preserve the health of ecosystems. The water quality of the Dahan River (the upstream of the Danshuei River) in Taiwan has degraded rapidly due to heavy pollutants transported from cities and their surrounding urban areas. Moreover, Taiwan endures contrasting seasonal variations in river flow and water quality because of the short duration and severe intensity of storms and typhoons that attack Taiwan. Sudden changes in river flow may easily result in serious deteriorations in water quality and huge impacts on ecosystems. Therefore, this study attempts to identify the crucial water quality variable of the Dahan River and establish its concentration estimation model by incorporating relevant factors from various fields. Consequently river water quality management can be conducted by controlling the discharge of the upstream dam of the Dahan River.

In this study, correlation coefficient analyses discover the characteristics and spatial distributions of various water quality parameters. The results reveal that ammonia nitrogen (NH3-N) is the crucial variable in judging the contamination level over the study area and NH3-N can be easily influenced by discharge and rainfall. The Gamma test, a statistical analysis method, is used to extract the most relevant factors from various NH3-N-related environmental fields, and thus the most sensitive combination of factors determined is used as inputs to different artificial neural networks (ANNs) for constructing the estimation model of NH3-N concentration. The results show that the NARX network calibrated by the cross validation method adequately utilizes the information of model outputs through recurrent connections to the network itself for estimating NH3-N concentration within shorter time interval without conducting field sampling. For water quality stabilization, this study collaborates the constructed NARX network with mass balance equations and boundary conditions to obtain outstanding results in reducing NH3-N concentration below current water quality goals (i.e., 25% and 50% decreases in concentration). Subject to the regular operation and water demands of the Yuanshan Weir, both 25% and 50% decrease operations can significantly level down NH3-N concentrations with smaller variances, and thus effectively maintain stable water quality in the Dahan River. The results can provide useful information for reservoir operation managers to sustain acceptable water quality throughout different seasons.

並列關鍵字

Water quality ； River basin management ； Ammonia nitrogen (NH3-N) ； NARX neural network ； Gamma test (GT)

參考文獻

Chang, C.L., Liao, C.S., Lo, S.L., Hu, C.Y., Ma, J.L. (2011) The Correlation Analysis of Taiwan's Main Water Treatment Plants between Upstream of the Water Turbidity and Hydrological Characteristics, Journal of Taiwan Agricultural Engineering Vol. 57, No. 2, June 2011.

Yang, C.P., Lung, W.S., Liu, J.H., Hsiao, W.P. (2009) Establishment and Application of Water Quality Model of Hypoxic Stream, Journal of Taiwan Agricultural Engineering Vol.55, No. 1, March 2009.

Heathwaite, A. L., Johnes, P. J., & Peters, N. E. (1996). Trends in nutrients. Hydrological Processes, 10(2), 263-293.

Agalbjorn S, Koncar N, Jones A.J. (1997) A note on the gamma test, Neural Comput. Appl., 5 131-133.

Ahmad, S., Simonovic, S.P. (2005) An artificial neural network model for generating hydrograph from hydro-meteorological parameters. Journal of Hydrology 315 (1–4), 236–251.