Using artificial intelligence to improve microalgae growth modeling in continuous photobioreactors

The application of microalgae is attracting increasing interest for waste-to-biomass applications. Notably, the red microalga Galdieria sulphuraria has emerged as a promising candidate due to its metabolic flexibility and its ability to thrive under extreme environmental conditions. The main critical aspect is the development of a quantitative model that accurately describes its behavior under different trophic regimes and operating conditions during continuous cultivation. In particular, the mathematical description of the mixotrophic regime results complex, as the relative contributions of the autotrophic and heterotrophic pathways vary continuously with the operating conditions. The sparsity of experimental data and the limited information available in the literature make the construction of an accurate and interpretable model complex. To address these limitations, a state-of-the-art artificial-intelligence-driven methodology for interpretable hybrid model discovery, known as HyMech, is employed to generate mathematical corrections to a prior first-principles model, inferred from the analysis of experimental data. First, the theoretical foundations of the algorithm and its application to a simulated case study are presented to provide understanding of its functioning and implementation, highlighting the most critical steps and demonstrating the potential of the methodology. Then, experimental data of Galdieria sulphuraria are studied to understand the effects of the process variables on the microalga growth within a bioreactor, enabling the construction of a prior model. Finally, the algorithm is employed to discover appropriate corrections to the backbone fist-principles model that attenuate the observed process-model mismatch. The advantages and limitations of using HyMech are discussed extensively.