Shohreh Mirzaei - Chemical Engineering Department, Faculty of Engineering, Ferdowsi University of Mashhad, Mashhad, Iran
Akbar Shahsavand - Chemical Engineering Department, Faculty of Engineering, Ferdowsi University of Mashhad, Mashhad, Iran
Ali Ahmadpour - Chemical Engineering Department, Faculty of Engineering, Ferdowsi University of Mashhad, Mashhad, Iran
Ali Garmroodi Asil - Chemical Engineering Department, Faculty of Engineering, University of Bojnord, Bojnord, Iran

In the present article, a performance of optimized regularization network (RN) and adaptive neuro-fuzzy inference system (ANFIS) are compared with a conventional MATLAB multi-layer back propagation toolbox for prediction of carbon active iodine number as a function of impregnation ratio (2.5-4), activation temperature (660-850 ○C) and residence time (60-150 min). The networks were trained by resorting several sets of experimental data ordered with Design Expert software. All the anthracite-based samples were chemically activated with KOH at various preparation conditions for methane storage applications. An efficient LOOCV criterion were employed for training the optimal isotropic Gaussian regularization network. Around 20% of data sets were randomly selected for validation performance of all networks. Error analysis along with correlation coefficient demonstrate that ANFIS and MLP networks have a superior performances over entire training exemplars, while, optimized regularization network shows excellent performances on both training andvalidation exemplars with R2 0.9994 and 0.9970 and root mean squared error of 6.66 and 32.96 respectively. Hence, RN can be considered as a reliable tool for predicting iodine number of activated carbons versus several preparation factors for methane storage purposes.

Active carbon, Iodine Number, Regularization network, ANFIS, Methane storage.