Machine Learning Prediction of the Yield and BET Area of Activated Carbon Quantitatively Relating to Biomass Compositions and Operating Conditions

Although
activated carbon’s yield (quantity index) and BET
area (quality index) are crucial to its application, the two indexes
must be accurately predicted. Herein, biomass compositions (ultimate
analysis, proximate analysis, and chemical analysis), operating conditions
(mass ratio, carbonization time, carbonization temperature, activation
time, and activation temperature) under physical activation (CO2 and steam), and chemical activation (H3PO4, KOH, and ZnCl2) conditions as input parameters
were used to predict the two indexes of activated carbon simultaneously
through the random forest (RF) method for the first time. In total,
the samples (>1500 data) identified from experiments in the literature
were used to train, validate, and test the RF models. The results
show that the model built on ultimate analysis is more suitable for
predicting the BET area and yield of activated carbon prepared by
both physical and chemical activation. Therein, the R2 values of activated carbon’s yield and BET area
under the H3PO4 activation condition were the
highest, which were 0.98 and 0.97, respectively. In addition, the
influence of various factors and interactions on the target variables
was analyzed. The results show that the hydrogen content has a large
impact on the yield under physical activation conditions, and the
mass ratio has the most contribution to the BET area under chemical
activation conditions. This study affords achievable hints to the
quantitative prediction of porous materials affected by multiple compositions
of raw materials and different operating conditions.