Data-driven recurrent neural networks modeling of cyanobacteria growth in bubble columns reactors under sparging with CO2-enriched Air

Abstract

Cultivating carbohydrate-enriched microalgae in photobioreactors is a recognized method for converting atmospheric carbon dioxide (CO2) into valuable resources, offering a solution to combat industrial CO2 emissions. Achieving global carbon neutrality by 2050 necessitates an annual reduction of approximately 1.4 gigatons of CO2. To address this challenge, we developed a machine learning-based black box model using data from the Institute of Advanced Materials for Sustainable Manufacturing at ITESM. This dataset encompasses hourly operational parameters collected over eleven months, including temperature, pressure, and flow rate. These parameters were essential for a microalgal cultivation experiment using the cyanobacterium Desertifilum tharense and CO2-enriched air. Two 100-L photobioreactor systems were employed. Leveraging the sequential dataset, we deployed a long short-term memory network (LSTM) to predict CO2 percentages at reactor outlets, indirectly indicating algae biomass production. Seven input variables were selected on the basis of their impact on system behavior. Using advanced machine learning time series mathematical models can boost process intensification tasks through advanced process optimization, simulation, and control goals.