(475f) A Hybrid Mechanistic-Machine Learning Approach for Identifying Governing Dynamical Equations of Algal Biodiesel Production Networks

Conversion of waste nutrients and CO₂ to biodiesel via algal biomass is considered a key technology for building a sustainable future for waste-to-energy systems. The process of converting waste to bio-diesel is a network of biological and chemical conversion processes that are driven by its own mechanistic dynamics such as chemical kinetics, mass transfer and separation. However, the driving parameters for the rate of conversion of waste to bio-diesel are hidden in this complex sequence of mechanistic processes, which will help identify how fast can the CO₂ be sequestered or waste nutrients re-utilized to meet the goal of global CO₂ reduction. Therefore the understanding of governing dynamics of this system is vital for the sustainability goals.

In this work, we address this gap of lack of dynamical equations for waste to the algal-bio-diesel production network. For achieving this, we propose a hybrid mechanistic-Machine Learning (ML) approach that utilizes data from mechanistic dynamic process models for the network with ML algorithms to identify governing dynamical equations for key state variables. We split the algal-bio-diesel plant into tractable individual production blocks. Preliminary results of the production blocks show 79% and 82% test accuracy for Symbolic Regression and Sparse Identification of Non-Linear Dynamics, respectively. The combined system analysis is achieved through solving coupled differential equations of these blocks. We also show how the hybrid approach gives kinetic equations for different types of reactor and use of these equations for sustainability decision making. We also argue on the applicability and limitations of different machine learning approaches in identifying the governing dynamics of the manufacturing systems.