Título: Machine Learning -Driven Architecture for Real -Time Optimization in Industrial Processes
Autores: Filipe Andersonn Teixeira da Silveira, Elcio H. Shiguemori & Filipe A. N. Verri
Resumo: This paper presents a modular framework and a predictive modeling approach to enhance Real-Time Optimization (RTO) strategies in industrial combustion processes using Machine Learning (ML). The proposed architecture defines a generalizable methodology for integrating ML models into RTO systems, supporting scalable and adaptable applications in energy-intensive operations. A case study on continuous furnaces is used to demonstrate the methodology, where predictive models are developed to estimate optimal initial setpoints for the Lower Calorific Value (LCV) of the fuel mixture. Historical process data are used to train and compare several supervised learning algorithms, aiming to reduce reliance on manual configurations and improve the efficiency of the optimization system. Among the tested models, the Random Forest regressor achieved the best performance, combining low prediction error with high explanatory power. These predictions contributed to a noticeable reduction in natural gas consumption when used to initialize the RTO. The results confirm the potential of ML to improve optimization outcomes and support sustainable, data-driven decision-making in industrial processes.
Palavras-chave: Real-Time Optimization; Machine Learning; Predictive Modeling; Modular Architecture; Industrial Energy Efficiency.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1173977
Artigo em PDF: CBIC_2025_paper1173977.pdf
Arquivo BibTeX:
CBIC_2025_1173977.bib