Catalyst Design by Machine Learning and Multiobjective Optimization

The computer technologies of machine learning and multiobjective optimization were introduced to develop the catalyst for fluid catalytic cracking (FCC). Response surface methodology was applied a training set consisting 1000 data points with varied compositions which consist variety catalysts compositions, feedstock properties, pseudo-equilibrium conditions, performance test conditions as input parameters results outputs. At first, response model (RSM) obtained four approximation methods, among radial basis function (RBF) method found give highest score accurate RSM smallest average error coefficient determination them. Then virtual experiments carried out genetic algorithm (MOGA) optimize design considering multiobjective; yield less bottoms, coke, more gasoline gas. After 5000 out, we that pareto front obtained. Finally, optimum selected from designs on front. As result, showed 2.7 % higher confirmed show excellent over conventional FCC catalyst.