Application of kernel extreme learning machine and Kriging model in prediction of heavy metals removal by biochar

Kernel extreme learning machine (KELM) and Kriging models are proposed to predict biochar adsorption efficiency of heavy metals. Both six popular ions (Pb2+, Cd2+, Zn2+, Cu2+, Ni2+, As3+) and single ion are considered to test the accuracy of KELM and Kriging models. Two ways (data selection and fix output value) are attempted to improve the model fitting accuracy and the best R2 can reach 0.919 (KELM) and 0.980 (Kriging). In addition, stepwise regression and local sensitivity analysis show that adsorption efficiency has strong relationship with pHsolute and T. Moreover, the most sensitive parameters are T, pHH2O, r, C and pHsolute. The accurate KELM and Kriging models identify the most important controlling factors on metal adsorption, and ultimately provide some sort of predictive framework that will be useful in selecting appropriate biochar for particular treatment scenarios. This, in turn, will reduce the number of metal-biochar adsorption experiments needed going forward.

Automated summary

The study introduces the use of KELM and Kriging models to predict the heavy metal adsorption efficiency of biochar, improving model fitting accuracy and identifying the strong relationship between adsorption efficiency and pHsolute and T. The most sensitive parameters are T, pHH2O, r, C, and pHsolute. Accurate predictive models help reduce the number of experiments needed for metal-biochar adsorption in the future.