Optimization of biodiesel production from oil using a novel green catalyst via development of a predictive model

Abstract

This study provides a machine learning (ML) modeling method for predicting the production of biodiesel from palm oil through transesterification process. The ensemble decision tree-based algorithms including AdaBoost Regression Tree (ADA + RT), Extra Trees, and Gradient Boosting Regression Tree (GBRT) were used as a potential tool for modeling biodiesel production. The time of reaction (h), methanol to oil (palm oil) molar ratio, and catalyst amount (wt.%, zeolite) were selected as the input variables of models, while the fatty acid methyl esters (FAME) production yield was set as the output for modeling as well as optimization tasks. The performance models were compared using several performance indicators (R2, RMSE, MAE). The obtained MAE standard error rates for ADA + RT, Extra Trees, and GBRT were 1.2, 1.1, and 0.33, respectively. Comparing the RMSE measurements showed that ADA + RT and Extra Trees had error value of about 1.5 and this value for GBRT model was about 0.4. Although all of the models that were generated were robust, the GBRT model was found to be the most robust and accurate in terms of predicting biodiesel output. The optimization of results confirmed that 98.73% yield of production can be achieved at optimal values operating factors (time = 45 h, methanol to oil = 12.0, and catalyst = 2.0 wt%).