Intelligent predictive modeling for the optimization of advanced algal photobioreactors in greenhouse gas capture and utilization.

Approximately 76 % of global greenhouse gas emissions are attributed to carbon dioxide (CO2), highlighting the need for effective mitigation strategies. In this context, smart photobioreactors (PBRs) utilizing microalgae have been identified as a promising carbon capture technology. Moreover, developing advanced predictive models can enhance biomass production, optimize carbon sequestration, and improve the sustainability of PBR systems. This study investigated the performance of different data-based machine learning prediction models for CO2 removal efficiency (RE) and Chlorella vulgaris growth under a smart PBR system. A 13-15-2 feed-forward backpropagation neural network (FFBP NN) and a 7-component partial least squares (PLS) were developed to predict multiple response variables. Results showed that FFBP NN was the optimum model by demonstrating superior performance (R2: ≥0.933 CO2 RE, ≥0.980 C. vulgaris growth; Root Mean Square Error: ≤4.730 % for CO2 RE, ≤37.80 mg L-1 for C. vulgaris growth) compared to PLS model due to its capacity to process larger datasets and ability to deal with the high variations. Meanwhile, PLS only relied on collinearity, but it could reveal variable importance and interactions. For instance, pH and inlet pressure highly affected CO2 RE, while nitrogenous compounds and phosphorus were highly related to algal growth. The dual focus of the intelligent models highlights an original concept in both reducing greenhouse gas emissions to promote environmental sustainability and advancing a circular economy through the production of algal biomass.