Computational framework for multi-objective optimization of activated biochar properties using machine learning and evolutionary algorithms

Abstract

Abstract Climate neutrality and renewable energy expansion demand multifunctional materials capable of simultaneous carbon sequestration and electrochemical energy storage. Agricultural residue biochar offers dual-function potential, yet conventional pyrolysis does not systematically optimize competing objectives. This study presents a simulation-based computational framework integrating multi-output random forest surrogate modeling with differential evolution algorithms to identify Pareto-optimal process configurations balancing specific surface area, CO 2 adsorption, electrochemical capacitance, and carbon stability. 800 parameter combinations were evaluated across pyrolysis temperature (400–900 °C), residence time (0.5–3.0 h), heating rate (5–50 °C min − 1 ), activation chemistry (KOH, $$\:{\text{H}}_{3}{\text{PO}}_{4}$$ , $$\:{\text{ZnCl}}_{2}$$ , NaOH), and five feedstock classes using calibrated response surfaces. The surrogate model achieved test-set $$\:{\text{R}}^{2}$$ of 0.971 (RMSE = 48.06 m 2 g − 1 for specific surface area and 0.942 (RMSE = 6.67 F g − 1 for specific capacitance on 160 independent samples; CO 2 adsorption yielded $$\:{\text{R}}^{2}$$ = 0.788, while carbon stability index showed moderate fidelity ( $$\:{\text{R}}^{2}$$ = 0.497, RMSE = 0.069), consistent with challenges in modeling recalcitrance from compositional proxies. Multi-objective optimization identified configurations with specific surface area of 1094 m 2 g − 1 , $$\:{\text{CO}}_{2}$$ adsorption of 5.01 mmol g − 1 , and specific capacitance of 114 F g −1 . Pyrolysis temperature was the dominant predictor (48% feature importance); hydrogen-to-carbon ratios below 0.4 demarcate recalcitrant materials suitable for millennial-scale sequestration. Optimized processing achieved feedstock-independent carbon stability (median 0.44–0.46) across all biomass types. Spatially explicit assessment indicates that EU-scale deployment could sequester 53.9 Mt $$\:{\text{CO}}_{2}$$ e year −1 , equivalent to 1.2% of total EU greenhouse gas emissions. Experimental validation of selected configurations constitutes the primary direction for future work.