EVOL UTIONARY BASE OPTIMIZATION METHOD TO DESIGN OF ARTIFICIAL NEURAL NETWORK FOR MODELING OF CLAUS REACTION FURNACE

in this paper, an evolutionary approach is used to optimize the topology and characteristics of a feed forward Artificial Neural Network (ANN) in order to predict Claus reaction furnace effluents (SO2 and S2) mole fractions. Input parameters include temperature, reactant (H2S) mole fraction and residence time. The ranges of input data vary from 950 to 1250 °C, 17.91% to 31.29% and 0.5 to 2 second, respectively. Two optimum multilayer feed-forward ANNs were developed separately to predict SO2 and S2 mole fractions at reactor outlet using Genetic Algorithm. Design of the optimum ANN includes determination of number of neurons in each hidden layer, neuron transfer functions and connection pattern through neurons. It can be concluded that using black box modeling provides an accuracy of more than 90% which shows a good improvement comparingwith available kinetic modeling.

EVOL UTIONARY BASE OPTIMIZATION METHOD TO DESIGN OF ARTIFICIAL NEURAL NETWORK FOR MODELING OF CLAUS REACTION FURNACE Keywords:

Artificial Neural Network , Genetic Algorithm , Claus Reaction Furnace , Reactor Modeling , Optimization