Framework Based on Artificial Intelligence to Increase Industrial Bioethanol Production

Increasing the use of bioethanol fuel is an option for reducing greenhouse gas emissions, although there is still room for improvement in the industrial production processes. Simulation tools can assist in improving industrial operations, but the models presented so far in the literature to describe bioethanol production from sugar cane were developed under conditions far from industrial reality, at the bench scale and with highly controlled fermentation variables. This hampers the use of these models to find optimized input conditions to increase industrial bioethanol production. In the present work, a framework based on artificial intelligence techniques combining the Artificial Neural Network (ANN) model and Particle Swarm Optimization (PSO) algorithm was developed to optimize industrial bioethanol production. Industrial data from a mill in the state of São Paulo (Brazil) were used to train the ANN. The databank comprised 3400 experimental values (200 operation days) of the whole fermentation unit. The trained ANN model was able to predict the bioethanol concentration at the end of the process with high accuracy and enabled the optimization of the input variables to maximize bioethanol production of the mill using the PSO algorithm. The results showed an increase of about 10% in bioethanol concentration and production at the end of fermentation per harvest for the industrial unit taken as a case study. This new approach enhances the knowledge about the industrial ethanol fermentation process and can become a tool to guide new studies regarding the increase of biofuels production, mainly the ones which present behavior with high nonlinearity.