The Study of Applying Regression Model and Artificial Neural Network Model to the Cost Function of Wastewater Treatment Plant

• Main Authors: Tian-Yow Tsui, 崔天佑
• Other Authors: Terng-Jou Wan
• Format: Others
• Language: zh-TW
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/17018561785939179997

碩士 === 國立雲林科技大學 === 環境與安全工程技術研究所 === 87 === Traditional way of wastewater treatment is time-consuming and laborious. Since the quantity and quality of the wastewater can not be precisely predicted, designers often decide the value of design parameters based on their experience and intuition, and without considering problems like the cost and benefit. Moreover, there is lack of researches in this field. Therefore, it is an urgent need to build a reasonable and complete database of cost function parameters of wastewater treatment plant for the domestic industries. This research aims to, by analyzing the cost structure of wastewater treatment process, study the effect of each treatment unit on the total cost of constructing a wastewater treatment plant. This research adopts multiple regression and back-propagation network (BPN) prediction model, in which fourteen operational unit designing parameters, i.e. total design flow, biological treatment unit volume and secondary sedimentation basin volume, are used as the input variables, to stimulate and analyze the construction cost functions of several wastewater treatment plants which were built after 1991. Then, overall comparison of this method with the traditional simple regression function was carried out to figure out a proper model of predicting the cost of wastewater treatment plant. It is found that it is difficult to obtain the best results from traditional simple regression function because the cost of wastewater treatment plant is unstable. Artificial neural network is a simple, quick and precise tool which performs well in predicting the cost of waste water treatment plants. The best results of predicting the cost of constructing waste water treatment plant can be obtained from BPN, in which root mean square error(RMS) and coefficient of correlation(R) is used as standards, and the volume of sterilizing basin as input variables. From the models we have constructed, it is found that based on the efficiency of prediction, models can be classified from the best one to the worst one as follows: neural network、simple linear regression model、simple non-linear regression model、multiple non-linear regression model and multiple linear regression model.