Methods for imputing missing values in water quality database of wastewater treatment plant

碩士 === 立德管理學院 === 資源環境研究所 === 95 === Water qualities of the wastewater treatment plant are completely affected by its operation conditions. And thus, with regard to imputing missing values of water qualities, it's impossible to process data by simply considering the data itself and, instead by...

Full description

Bibliographic Details
Main Authors: Fang-Mao Yang, 楊芳茂
Other Authors: Chun-Sheng Wu
Format: Others
Language:zh-TW
Published: 2007
Online Access:http://ndltd.ncl.edu.tw/handle/48294509268788776698
Description
Summary:碩士 === 立德管理學院 === 資源環境研究所 === 95 === Water qualities of the wastewater treatment plant are completely affected by its operation conditions. And thus, with regard to imputing missing values of water qualities, it's impossible to process data by simply considering the data itself and, instead by taking into account of operation conditions. An appropriate way for imputing missing values both with the statistics�{ Principal Components Analysis and computing instrument�{ Artificial Neural Networks, PCA + ANN, was developed in this study. The developed method of PCA + ANN will be used for imputing missing values in water quality database of wastewater treatment plant. Three years' data of operating and experimental water quality of the wastewater treatment plants located at the industrial park in southern Taiwan was applied in this study. Firstly, 10 variables (inflow rate, influent COD and SS concentration, MLSS, SVI, F/M ratio, HRT, COD volumetric loading rate, SRT, and pH) related operational conditions of the activated sludge system were conducted with PCA. Four principal factors�{ “water quality factor”, “water volume factor”, “MLSS factor”, and “SVI and aeration basin COD volumetric loading factor”, each combined with the 10 variables, were then simplified. Secondly, ANN training task was performed by inputting the four principal factors integrated from 720 datasets of COD and SS beforehand, and followed by ANN simulation with 180 datasets of COD and SS. Finally, to verify the simulating results, both relative coefficient (R) and relative error between experimental data and results simulated by PCA + ANN were compared with those of simulating by ANN, linear interpolation method (LIM), periodical average method (PAM). The results were shown as follows: R values of COD (= 0.736) and SS (= 0.751) by using PCA + ANN method were respectively higher than those simulated by ANN with inputting 10 variables (= 0.685 and 0.653, respectively), ANN with 4 variables�{ SVI, F/M ratio, SRT, and COD volumetric loading (= 0.666 and 0.627, respectively), LIM (= 0.723 and 0.748, respectively), and PAM (= 0.722 and 0.741, respectively). The relative errors of COD (= 0.1%) and SS (= 9.9%) by using PCA + ANN method were respectively lower than those of 2.1% and 10.4% by ANN (inputting 10 variables), 1.4% and 10.1% by ANN (inputting 4 variables), 0.7% and 12.7% by LIM, and 0.7% and 65.8% by PAM. Accordingly, PCA + ANN method should be superior in imputing missing values in water quality database of wastewater treatment plant.