Biochar is widely applied as an adsorbent for removing contaminants. Herein, two machine learning (ML) models for biochar preparation and adsorption application were trained based on the eXtreme Gradient Boosting algorithm. Then, the two models were combined by developing a hybrid ML-based optimization framework via Particle Swarm Optimization to design biochar for application requirements. The test determination coefficient for predicting specific surface area (SSA), total volume, and adsorption capacity of biochar were 0.85, 0.88, and 0.97 with root mean square error of 63.10 m2/g, 0.07 cm3/g, and 65.72 μmol/g, respectively. Moreover, SSA was the foremost and positive property of biochar for its adsorption capacity; and pyrolysis temperature and feedstock ash content were the two most important factors affecting SSA, among which the former was positively related and the latter had a negative impact. Optimization results indicated that pine wood pyrolyzed at 500–700 °C could prepare a biochar with higher antibiotic adsorption capacity. This work presents an intelligent strategy to design biochar for adsorbing target pollutants.
Read full abstract