Machine learning-driven optimization of argon oxygen decarburization slag recycling for enhanced microalgal carbon sequestration

Abstract

The sustainable management of hazardous argon oxygen decarburization (AOD) slag demands urgent attention owing to its calcium-magnesium-silicon leaching risks in landfill scenarios. This study presents an innovative strategy for waste valorization by repurposing three modified AOD slag variants (raw, aged, and carbonated) as nutrient supplements for Chlorella pyrenoidosa cultivation. Moreover, process parameters in microalgae cultivation, such as algal characteristics and complex operational conditions, will affect its yield and productivity. Traditional methods struggle to enable comprehensive understanding and application. Thus, quantitative prediction was conducted using 96 sets of total CO₂ carbon sequestration data (80% for the training set and 20% for the test set). Combined with three machine learning models and the Shapley Additive explanation (SHAP) algorithm, the intrinsic mechanisms by which five leaching elements (Ca, Mg, Al, Si, and Cr) regulate the efficient carbon sequestration of microalgae were analyzed. Notably, the random forest model excelled well in predicting CO₂ storage and elemental leaching, with performance metrics exceeding 0.87. This approach integrating solid waste recycling, utilization and model development achieves three objectives: (1) establishing a circular economy pathway for metallurgical waste, (2) reducing microalgal cultivation costs through waste-derived nutrient substitution, and (3) providing a machine learning blueprint for hazardous waste valorization process optimization. The research results provide guidance for implementing a sustainable strategy of biocarbon capture while reducing industrial waste.