Theoretical investigation on optimization of biodiesel production using waste cooking oil: Machine learning modeling and experimental validation

Abstract

In order to optimize productin of biodiesel from waste cooking oil utilizing Fe-exchanged montmorillonite 12 K10 (Fe-MMT K10) heterogeneous catalyst was applied in this work. The data of batch reaction experiments were collected for optimization considering four inputs and one output. The input parameters included reaction temperature, reaction time, catalyst loading, and ratio of methanol to oil. The model was developed to predict the output which is the production yield of biodiesel (%). For optimization of the process, three ensemble models were utilized as a novel method for the first time in this study: Huber Regression, Decision Trees, and Gaussian process which were all boosted using AdaBoost technique. The R2-Scores for Boosted Huber Regression (ADABOOST-HBR), Boosted Decision Tree (ADABOOST-DT), and boosted Gaussian process (ADABOOST-GPR), respectively, were 0.814, 0.780, and 0.996. The calculated MAE parameter for the models illustrated that the error rates for Boosted Huber Regression ADABOOST-HBR, ADABOOST-DT, and ADABOOST-GPR were 3.84, 5.94, and 1.82, respectively. Indeed, the boosted GPR model has a better accuracy over the two models in optimization of the process. Moreover, applying the input values (X1=145, X2=5.625, X3=4.22, X4=11.73), the recommended methods produced an ideal output value of 96.75% which was considered to be the optimum yield for production of biodiesel.

Keywords

• Optimization
• FAME
• Biodiesel
• Heterogeneous catalyst
• Machine learning
• Waste oil