Effects of dynamic inflow on biological nutrient removal of an A2O system

• Main Authors: Jiahui.Huang, 黃家輝
• Other Authors: Tsai Yung-Pin
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/96212528989640743475

碩士 === 國立暨南國際大學 === 土木工程學系 === 94 === Abstract Activated sludge system is a non-linear and complicated system. For biological nutrient removal (BNR), dynamic loading of the influent results in dynamic variation of effluent. For operators of wastewater treatment plants, they have to daily maintain the stability of the activated sludge system under dynamic loading of influent. It is also necessary to prevent the system from sludge bulking, washout and the decrease of treatment efficiencies. If the activated sludge system is operated basing on steady-state, it is difficult to well control the dynamic characteristics of effluent and to achieve the objective of a BNR system. In this research, an A2O pilot plant was used to acclimatize activated sludge under different influent patterns and concentration for simulating domestic wastewater treatment plant. The pattern of recyclable nitrified water and recyclable sludge were simultaneously varied to control sludge pre-recycle for BNR. The artificial neural network and grey theory were further used to describe the metabolism of microorganism and to predict the effluent of BNR under dynamic loading of influent. Then, the prediction abilities of the artificial neural network and grey theory were further compared in the study. Basing on type1, type2 and type3 control strategies, the results indicated that the removal efficiencies of soluble COD were 92%, 94% and 92%, respectively; the removal efficiencies of ammonia nitrogen were 78%, 90% and 95%, respectively; the denitrification rates were 77%, 74% and 83%, respectively; the removal efficiencies of phosphate were 22%, 57% and 14%, respectively. The type2 control strategy was the best one. On the type2 control strategy, it was found that MLSS concentration during higher loading periods was higher than overall average MLSS concentration. During lower loading periods, MLSS concentration was lower than overall average MLSS concentration. The Type2 control strategy achieved the objective of sludge pre-recycle control. The data set of water quality obtained from all kinds of control strategy was used to train and test both of the models (the artificial neural network and grey theory). Coefficient of correlation (R), mean absolute percentage (MAPE) and root mean square (RMS) between experimental data and predicted data were used to evaluate the fitness of the model. It was obviously found that the artificial neural network was more appropriate for predicting the dynamic behaviors of the A2O system. Then, the sensitivity analysis was further used to comprehend the influence of each variable on the effluent and the internal variation of the A2O system.