(63e) Neural Modeling of a Continuous Alcoholic Fermentation Process and Its Optimization by Successive Quadratic Programming

Brazilian annual production capacity of ethanol is currently in 18 billions of liters and, with the increase in demand, the idea is to multiply this capacity through the installation of new factories and optimization of the operation of the existing ones. Thus, there is an intensified interest in the study of all the steps involved in ethanol production. This work focuses on the process operation aspects using model-based optimization of an continuous extractive alcoholic fermentation process. A data-driven identification method based on Multilayer Perceptron Neural Network (MLPNN) and optimal design of experiments was development. The neural model is optimized using Successive Quadratic Programming (SQP) to find out the optimal operational conditions so that the conversion is maximized to the defined allowable limits. In order to check the validity of the computational modeling, the results were compared to the optimization of a deterministic model, whose kinetic parameters were experimentally determined. It was observed that the values for productivity and conversion obtained using the MLPNN models are similar to that obtained using the deterministic model. MLPNN modeling is a powerful and flexible tool and its use in the alcoholic fermentation process is advantageous because updating the neural network parameters is a simpler procedure than reestimating kinetic parameters of the deterministic models. Frequent reestimation of kinetic parameters (or in the case of this work updating the neural network parameters) is necessary due to changes in the dynamic behavior caused by fluctuations in the quality of the raw material and changes in microorganism metabolism, among other factors.