類神經網路在環境單元操作最適化之應用

CNU IR > College of Sustainable Environment and Recreation > Dept. of Environmental Engineering and Science (including master's program) > Dissertations and Theses > Item 310902800/6132

Please use this identifier to cite or link to this item: https://ir.cnu.edu.tw/handle/310902800/6132

Title:	類神經網路在環境單元操作最適化之應用 Application of neural network on the environmental unit operation optimization
Authors:	蕭振文 Chen-wen Hsiao
Contributors:	殷堂凱廖志祥嘉南藥理科技大學：環境工程與科學系碩士班
Keywords:	倒傳遞網路類神經網路 Neural network Back-Propagation neural networks
Date:	2005
Issue Date:	2008-10-31 16:14:56 (UTC+8)
Abstract:	本研究主要是利用類神經網路(Neural Networks)的多輸入多輸出系統的能力，運用於生活污水廢水處理廠的放流水質預測模擬與元素鐵配合二氧化碳曝氣去除水中硝酸根的實驗程序模擬。類神經網路含有多種網路結構，在本研究使用可運用於運算非線性資料的倒傳遞網路(Back-Propagation neural networks)，其可運算非線性數據原理是多了隱藏層。廢水廠部份是根據實廠的實際操作參數和進流水質，以類神經網路做批次式運算，預測實際的放流水質；實驗程序部份是以實際實驗所得的數據及實驗條件，以類神經網路運算，預測實際的放流水質。廢水廠的模擬部份，資料取自Montréal污水處理廠，以2001年水質資料訓練網路，預測2002年的放流水質資料。當測得廢水廠當天進流水質時，運用本研究所建構最佳化之模式架構，能迅速的運算而預測當天的放流水質，幫助現場操作人員即時調整操作參數，以使放流的水質能合乎排放標準。化學處理實驗程序的模擬部份可以假設實驗的初始條件，以模式運算而得放流水質數據，可以幫助節省冗長的實驗時間、實驗初始條件的調整及節省藥品的消耗以及幫助設計實驗程序。研究結果顯示，資料先經正規化處理有助於模式的預測效能並能辨識是否含有不良變數，當含有不良的變數時，正規化後模式的預測效能將會降低。相關係數(Correlation Coefficient)可以使用於判斷線性變數間的相關性，而非線性變數的相關性就需要繪圖來判斷。單一輸出變數的網路架構可以達到不錯的預測效能，故以此架構的預測效能為目標，建立其它種類的網路架構比較。 This research is to use ability of multi-input and multi-output from artificial neural network for environmental application. It is applied to predict effluent quality in municipal wastewater plant and to simulate the experiment of nitrate removal by zero-valent iron coupled with carbon dioxide bubbling. Neural network contains many kinds of network structures, of which Back-Propagation neural networks was employed in this research. In the case of wastewater treatment plant, it is based on parameters such as the practical operation and influent wastewater quality. The batch operation was used to predict effluent wastewater quality discharging into the river by using neural network. In 2001, water quality data obtained from Montréal wastewater treatment plant was carried out to train the model. According to the raw data of 2002, the model was used to predict effluent wastewater quality by using the input data from the influent quality. Therefore, the optimal model that predicts the operation parameter is useful and easy to operate in the real plant. In the case of chemical treatment process, the data taken from experimental measurement was used to predict the output using the same type of neural network model. The optimal model was conducted to simulate the parameters such as nitrate, ferrous iron and ammonium. The advantage of this study can reduce the cost of material and time for research as well as help design the treatment process. As a result, the normalization could assist to predict performance of model. Moreover, the correlation coefficient can determine the correlation of variables except non-line variables. In addition, the single output model can achieve good prediction. Besides, setting up a single output model can support in comparing with other model structure.
Relation:	校內馬上公開，校外一年後公開
Appears in Collections:	[Dept. of Environmental Engineering and Science (including master's program)] Dissertations and Theses

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	1680	View/Open

Loading...