Machine learning-driven dual-objective optimization of biomass-derived carbon quantum dots toward high conductivity...

Biomass-derived carbon quantum dots (CQDs) have been reported to effectively improve the efficiency of anaerobic digestion processes. However, simultaneously achieving high CQD yield and high conductivity remains challenging due to the complexity of the process. To address this, a machine learning approach was employed to optimize the preparation of straw-based CQDs. Based on 273 experimentally obtained data points, a predictive model was developed using the AutoGluon framework. Interpretable machine learning analyses were conducted to elucidate the complex relationships between preparation parameters and CQD yield and electrical conductivity. The NSGA-II algorithm was then applied to obtain a set of Pareto-optimal solutions for these dual objectives. Under the high-conductivity scenario, the maximum conductivity of 0.713 mS cm^-1 was obtained at carbon-to-nitrogen ratio (C/N) = 1 : 2, a hydrothermal temperature (T) of 187.7 °C, and a reaction time (t) of 323.18 min. Under these parameters, the corresponding optimal CQD yield was 32.02%. Subsequently, we conducted practical anaerobic digestion scenario tests on the optimized CQDs, demonstrating excellent methane productivity, process stability, and acid buffering capacity. These results demonstrate that machine learning can effectively guide the preparation of CQDs with both high conductivity and high yield, and that the prepared CQDs can practically optimize anaerobic digestion application processes.