• An end-to-end ML framework bridges biochar synthesis and dye wastewater treatment. • Framework achieved R2=0.97-0.99 and RMSE=0.01-0.04 with strong accuracy. • Online platform enables rapid biochar design with real-time analysis. • Optimized biochar showed 125.25 mg/g adsorption, 18.71% above literature. Biochar has emerged as an essential material for wastewater remediation due to its tunable physicochemical properties. There is a pressing need to develop rapid and efficient strategies to guide the rational design of biochar, facilitating its transition from laboratory research to practical applications. In this work, we developed a sequentially coupled end-to-end machine learning framework (EEMLF) that integrates an online, user-friendly platform enabling rapid prediction of optimal biochar synthesis and its adsorption capacity without programming expertise. The EEMLF had high and robust predictive accuracy on dye removal end and biochar synthesis end (R 1 2 /R 2 2 = 0.99/0.97; RMSE 1 /RMSE 2 =0.01/0.04). Through EEMLF, pyrolysis temperature and the activator-to-biomass ratio were identified as the dominant factors governing biochar properties, while specific surface area emerged as the primary determinant of adsorption performance. Guided by the outputs of the EEMLF, the waste reed straw was pyrolyzed to produce a biochar (RSCH-800), RSCH-800’s methylene blue adsorption capacity reached 125.25 mg/g, nearly twice that of the unactivated one (RSC-800). For real wastewater (TOC Initial = 672.6 mg/L), RSCH-800 achieved 85% TOC removal at a solids-to-liquid ratio of 1.3 g/L within 2 h. This work delivers a generalizable data-driven framework that enables the rapid development of high-performance biochar for dye wastewater, even broader polluted wastewater remediation.
Gao et al. (Sun,) studied this question.