Machine Learning Techniques Applied to a Complex Polymerization Process

This chapter discusses the use of machine learning in modeling and optimizing free radical polymerization processes. Artificial neural networks, static and dynamic, with various configurations, used individually or aggregated in stack, are presented in different types of applications: direct and inverse modeling, soft sensors, or optimal control. A particular aspect is represented by neuro-evolution, by combining neural networks with evolutionary algorithms (genetic algorithms and differential evolution), with applications in determining optimal neural models or in optimizing chemical processes. In most cases, the selected examples, many of which are the author's own contributions, show the gradual improvement in performance of the applied method. Polymerization processes were chosen as case studies as they have complicated phenomenology, which gives rise to significant modeling difficulties. Machine learning techniques, which are capable of overcoming many of these disadvantages, provide satisfactory results.