Uso de metamodelos na otimização do processo de secagem do carvão.

Abstract

Coal is an input used as a reducer and/or energy source in several processes. This fuel is found in such a way that it cannot be used without previous treatment, requiring operations to reduce granulometry and moisture. Thus, this work aims to develop a phenomenological model of the coal definition process in Aspen Plus to serve as a data generator for the creation of metamodels based on Artificial Neural Networks, and thus, to evaluate routine optimization process scenarios. Together, a Genetic Algorithm was applied to determine ANN architectures, in order to reduce the metamodel prediction error. With the simulation validated from industrial data, it was possible to optimize the architecture of the network, there was a change in the number of layers to two and neurons, contributing to improve the prediction of the process, obtaining average errors from 0.01% to 1.66 % of predicted variables. The optimization of the operating scenario managed to significantly reduce the consumption of fuels used for heating the drying gas, with all process quality variables being met, resulting in a reduction of 15% of blast furnace gas and 80% of natural gas, reducing the energy consumption of the process.