Modeling of Adsorption Isotherm Constants Using Regression Analysis & Neural Networks

This paper deals with modeling the suitability of novel adsorbents in terms of the isotherm constants and temperature. Two different models viz., Linear Regression Model (LRM) and Model based on Normalized Sum of Squares (NMSS) are used. A database comprising of 120 data points (DB-1) pertaining to Freundlich constants is compiled from literature to find out the trends of n and k on Temperature (T). The accuracy of the prediction is compared using Standard Deviation (SD) based on expected values. The results are not satisfactory but indicated that the SD is lower for MNSS than that for LRM. The database is sorted out into three other data sets viz., having n values in the range 0.9 to 10 (comprising 94 data points), two clustered datasets for removal of Cr & Pb comprising of 12 and 14 data points respectively and fitted using both models. The SD values are comparatively lesser than those obtained with DB-1, but still not encouraging indicating wide spread in the database.