PREDICTION OF DIOXIN EMISSION FROM MUNICIPAL SOLID WASTE INCINERATION BY ARTIFICIAL NEURAL NETWORK AND ADSORPTION MODEL OF DIOXIN DERIVATIVES

Abstract

Dioxin (PCDDs) emission from incineration was predicted by an artificial neural network (ANN). The prediction was based on four-year monitoring data received from an incineration in Taiwan under a capacity of 90 tons a day. The result indicated that the prediction using model based on a back-propagation neural network, a promising method to deal with a complex and non-linear data with statistical analyses in the selection of useful variables for modeling was satisfied. The suitable architecture of an ANN for using in the dioxin prediction consisted of 5 input factors, 3 basic layers with 8 hidden nodes. Five important variables included amount and frequency of activated carbon injection, concentration of hydrogen chloride in the flue gas at the stack emission, temperature at the mixing chamber, and temperature of final fuel gas emission. The correlation factor (R2) was approximately equal to 0.998 in both training and testing steps. Activated carbon injection frequency was found as the most sensitive factor for PCDDs formation and emission. Laboratory experiment was carried out to remove the derivatives of dioxin compounds using activated carbon (AC). Benzene, chlorobenzene, dichlorobenzene, and O-chlorophenol were used as adsorbates. These chemicals were adsorbed onto four types of ACs. The adsorption experiments were carried out in a fixed-bed reactor with WHSV equal to 1.2 m3/kg-hr. The highest bed capacities in all experiments were obtained from AC derived from coconut shell (ACC). The adsorption behaviors were examined by thermal gravimetric analysis (TGA). O-chlorophenol showed the highest activation energy for desorption, 107 kJ/mol, implying chemisorption behavior. The adsorptions of the derivatives of dioxin compounds onto ACC were selected for the modeling of breakthrough curves. Isotherm parameters for selected chemicals can be listed as qe = (IP1Pi)/(1+IP2Pi) where qe is equilibrium adsorption (mg adsorbate/g agsorbent), and Pi is equilibrium concentration (mg adsorbate/L); Benzene: IP1 = 204 mg/g adsorbent, IP2 = 1.20 L/ml, Chlorobenzene IP1 = 358 mg/g adsorbent, IP2 = 1.46 L/ml, Dichlorobenzene IP1 = 349 mg/g adsorbent, IP2 = 1.51 L/ml, and O-chlorophenol IP1 = 603 mg/g adsorbent, IP2 = 1.28 L/ml.