Machine Learning no controle preditivo de processos

Abstract

The popularization of the concept of Industry 4.0, caused by technological advances and the expansion of the use of Artificial Intelligence techniques, seeking increasingly efficient processes, has made process control evolve more rapidly in the last decade. This happened with the implementation of advanced control techniques, as is the case of model predictive control (MPC). Developed in the 1980s to control power plants and oil refineries, MPC can now be found in the chemical and food industries. This type of control strategy has the advantage of being able to be applied in nonlinear systems, which have a more complicated approach in classical control, and to deal with the presence of operational restrictions. The MPC requires a computationally more complex implementation than the classic control techniques. In this context, the Python programming language has shown promise in several applications in science and industry, a fact that is mainly due to its wide range of libraries, which are free and free, making it a great alternative to paid software such as MATLAB®. With the increase in process automation, thanks to cheaper monitoring equipment, there is greater availability of data. This fact makes it possible to approach the problem with the use of internal models in the MPC controller based on Machine Learning. Machine Learning techniques, one of the pillars of Industry 4.0, study the construction of computational algorithms that enable the development of models based on the analysis of the data provided and, with this, the generalization of the information learned in the form of predictions. In the chemical industry, one of the most common types of equipment is the Continous Stirred Tank Reactor (CSTR). For this reason, this work addressed the construction of strategies for the use of Machine Learning in the MPC applied in the case of a CSTR with a nonlinear reactive system of the catalytic reaction of ethylene hydrogenation, also evaluating the applicability of Python and its respective specific libraries for control and Machine Learning, available for free.