Utilização de redes neurais artificiais para o controle de um reator do tipo CSTR

Abstract

Continuous Stirred Tank Reactors (CSTRs) present significant control challenges due to their nonlinear dynamics and susceptibility to operational disturbances. In this study, an approach based on Artificial Neural Networks (ANNs) was developed to determine the tuning parameters of PID controllers applied to a CSTR. Initially, a mathematical model of the reactor was implemented in Python, considering the mass and energy balances of the saponification reaction between ethyl acetate and sodium hydroxide. A total of 100 simulations were performed using step, pulse, and ramp disturbances, generating a dataset based on PID parameters optimized through the Sequential Least Squares Programming (SLSQP) algorithm. These data were use to train a Multilayer Perceptron (MLP) neural network, whose architecture was optimized using the Optuna framework. Model performance was evaluated through the Mean Absolute Error (MAE) metric and closed-loop validation tests. The results showed that the neural network was able to predict the control parameters with hig accuracy, presenting dynamic behavior similar to that obtained by the conventional optimization method. Therefore, this study highlights the potential of ANNs as a tool for automating controller tuning in complex chemical processes.