Data-driven predictive control models for manufacturing system

Abstract

Model Predictive Control (MPC) has become a cornerstone technique for the optimization of complex industrial processes, enabling the determination of optimal control actions over a receding horizon. In the context of micro-algae production, the original system model defined in [1] relies on a logistic growth approximation, based on methodological assumptions provided by the manufacturing company. While this model provides a baseline for optimization, deviations from real operational data suggest that its predictive accuracy may be limited, potentially impacting the efficiency of both production and maintenance schedules. This work aims to explore a Data-Driven Predictive Control (DDPC) framework that leverages operational datasets collected over more than one year, covering multiple cultivation columns and providing hourly measurements of column height level, pH, and algae density. By analyzing and pre-processing these datasets, we identify patterns and deviations from the original logistic model and construct data-driven growth predictors that more accurately capture the dynamics of micro-algae cultivation. These predictors are then validated and reviewed to identify the best data-driven method for micro-algae biomass evolution. The project also includes a comprehensive review of state-of-the-art MPC and DDPC techniques, selection of appropriate data-driven modeling approaches, and their integration into a predictive control strategy tailored for micro-algae production. By combining real-time data analysis, model development, and receding-horizon optimization, this study demonstrates the potential of data-driven methods to enhance traditional control strategies.