Desenvolvimento de estratégia para projeto de controlador preditivo baseada em deep learning

Abstract

Thinking of the search for industrial development through efficient processes, Industry 4.0 has been gaining space with operations in different areas. In the chemical industry, the development of interconnected processes can be highlighted, with more stable, efficient and robust control systems. This contributes to greater security of operations, adaptation to product specifications, as well as increased profitability of processes. Different types of drivers can be selected to keep the process variables in the desired conditions, but due to non-linearities in highly complex systems, the use of Model-Based Predictive Control (MPC) has grown. In MPC, basically, process conditions are applied in a dynamic model to predict future behavior and obtain better control actions. Machine Learning (ML) algorithms, such as Artificial Neural Networks (RNA), have gained attention in the Industry 4.0 context in a range of applications, especially in the use of predictive models for MPC. However, to deal with generalization difficulty and the non-applicability in some systems, in the current specialized literature can perceive the growing investigation for the use of Deep Learning (DL), type of ML characterized by having structures with a higher density of neurons. In this context, the aim of this work is to develop strategies for building an MPC controller using ML and DL algorithms as an internal model based on the Python programming language, which has been gaining prominence in the industry and in the scientific community for being easy to use, it has libraries that open fans for different applications and for having a free license. The constructed strategies were evaluated and validated through the application in a reactive process in a CSTR reactor (continuous stirred tank reactor) used in the control verification. From the driver structures, systems with greater stability were sought from the Deep Learning structures (recurrent networks), contributing to greater process stability in the Industrial 4.0 context.