Data-Driven Recurrent Neural Networks Modeling of Cyanobacteria Growth in Bubble Columns Reactors under Sparging with CO2-enriched Air

Abstract

The Institute of Advanced Materials for Sustainable Manufacturing at ITESM conducted a 10-month experiment on biomass production using cyanobacteria in bioreactors. This study highlights the robust predictive capabilities of LSTM models, which surpass traditional approaches. The research unfolds through key stages, emphasizing the importance of dataset preprocessing, including outlier identification, data scaling, and feature selection using statistical methods. Recognizing the dataset’s sequential nature, we transform the time series into supervised learning, enabling effective training and network design. The LSTM model architecture, detailed in Appendices A and B, efficiently captures temporal dependencies, offering an effective alternative to dynamic models. Eval uation metrics, such as mean absolute error and mean absolute percentage error, underscore the accuracy of the model, which is crucial for precision-demanding applications. This work significantly contributes to reaction system analysis by presenting a streamlined approach for forecasting biomass production. We introduce an innovative monitoring approach using CO2 concentration [%], yielding valuable insights for sustainable industrial practices. Looking forward, our LSTM models hold promise for optimizing algae biomass production systems. Future research will explore Bayesian optimization techniques, hyperparameter tuning, and real-time sensor data integration. This holistic approach aligns with ongoing efforts in industrial biotechnology and sustainability, bridging the oretical research with practical applications.