- 學位論文
民生水庫藻類生長影響因子及優養化潛勢預測分析之研究
Analysis of influencing factors towards algal growth and eutrophication potential prediction for drinking water reservoirs
摘要
民生水庫環境因子(如水質及水溫)會影響藻類生長及優養化發生，故如何準確評估影響水質優養化之水質因子，以及篩選最佳因子預測優養化潛勢是水庫水質管理重要課題。本研究自行調查湖山、寶山、鯉魚潭、石門等四座代表性水庫水質，以及收集歷年21座民生水庫水質監測數據(2011~2020年)，先評估藻數與葉綠素a之關聯性，再使用主成分分析(Principal aomponent analysis, PCA)法探討藻類生長影響因子與利用類神經網路(Atificial neural network, ANN)分析優養化潛勢最佳預測輸入因子。 本研究顯示四座代表性民生水庫優勢藻為矽藻與綠藻，其藻數與葉綠素a具高度關聯性(|P|=0.83)，葉綠素a可作為國內水庫藻類生長之專一性指標。歷年水庫水質數據之主成份分析(PCA)結果顯示，鯉魚潭水庫藻類生長因子主要為總磷(TP)、懸浮固體物(SS)；明德水庫藻類生長因子主要為總磷(TP)、氨氮(NH3)、化學需氧量(COD)。在預測鯉魚潭水庫藻類生長程度(Chl-a濃度)及優養化比例(ER)上，使用倒傳遞類神經網路(BPN)模式，最佳輸入參數為水溫(WT)、酸鹼值(pH)、懸浮固體物(SS)、溶氧飽和度(DOS)，其預測準確度高(Chl-a：R=0.69;ER：R=0.66)。然而，在預測明德水庫優養化比例(ER)上，最佳輸入參數為總磷(TP)、化學需氧量(COD)、氨氮(NH3)，其預測準確度(R=0.88)較高，但對藻類生長程度(Chl-a濃度)預測結果差(R<0.3)，主要是預測群組時間範疇內之水質數據波動性高於訓練組群之水質數據。綜合上述，類神經網路(ANN)輸入參數選擇可先透過主成份分析(PCA)推論，找出較合適的輸入因子(如水質因子)，以優化訓練結果及輸出因子數值(如Chl-a濃度及優養化潛勢)預測之準確度。
並列摘要
Environmental factors, such as water quality and temperature, would affect algae growth and etrophication occurrence for drinking water reservoirs. Thus, it is crucial to precisely evaluate the water qulity factors towards eutrophication and the determination of optimum factors for the prediction of eutrophication potential. This study investigated the water quality of four representative reservoirs, including Hushan, Baoshan, Liyutan, and Shimen. Meanwhile, the data about water quality of 21 drinking water reservoirs from 2011 to 2020 has been collected. Then, the relationship bettwen algae count and chlorophyll a (Chl-a) was evaluated, and investigate the influcing factors of algae growth by principal component analysis (PCA) along with the optimum input parameters of artificial neural network (ANN) simulation in the prediction of eutrophication potential. This study has shown that green algae and diatoms are the dominant spceies in four representative reservoirs where the algae count well correlated to Chl-a (|P|=0.83). The Chl-a as a specific indicator is proper to evaluate the growth of algae in Taiwan drinking water reservoris. PCA inference has indicated that algae growth influencing factors include total phosphorus (TP) and suspende solids (SS) for Liyutan reservoir while TP, ammonia (NH3) and chemical oxygen demand (COD) for Mingde reservoir. In the case of predicting algae growth (Chl-a concentration) and eutrophication ratio (ER) for Liyutan reservoir, the pronounced accuracy (Chl-a:R=0.69;ER:R=0.66) is obtained using the back propagation neural network (BPN) mode, in which the optimum input parameters include water temperature (WT), pH, SS and dissolved oxygen saturation (DOS). However, the optimum input parameters, including TP, COD and NH3, reach significant prediction accuracy (R=0.88) for Mingde reservoir, but, the accuracy of prediction in algae growth (Chl-a concentration) is relatively poor (R<0.3) because the variability of data in prediction period is over high. In summry, the determination of input parameters for ANN training preferrencially based on PCA inference, and then the proper input parameters (e.g., water quality factors) to optimize the accuracy of training and prediction parameters (Chl-a concentration and ER).
參考文獻
延伸閱讀
- 許嘉珍（2006）。新山水庫藻類生態模擬及改善優養化工法之初步探討〔碩士論文，國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU.2006.00757
- 黃致源（2013）。以電混凝/浮除法收集優養化水體藻類供製生質能之可行性評估〔碩士論文，國立暨南國際大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0020-0509201321193900
- 洪上尚（2015）。植生分佈及密度對深水型人工溼地水力效率及水流型態影響之研究〔碩士論文，國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU.2015.11363
- 張雅淳（2021）。Effect of nitrogen/phosphorus ratio on algal growth and disinfection by-product precursors〔碩士論文，中原大學〕。華藝線上圖書館。https://doi.org/10.6840/cycu202100701
- 簡鈺晴（2005）。Analyses of Algal Biodiversity and Modeling the Dynamics of Phytoplankton Succession in Feitsui Reservoir〔碩士論文，國立臺灣大學〕。華藝線上圖書館。https://doi.org/10.6342/NTU.2005.01037