Machine learning prediction of density of fatty acid methyl ester mixed with alkanes biodiesel over a wide range of operating conditions

Abstract

Biodiesel is observed as more environmentally friendly than fossil fuels because it contains no sulfur, produces low carbon oxide emissions when burned, and has a high oxygen content that promotes thorough combustion. Biodiesel is often blended with fossil diesel to meet the required properties for use as a fuel. Decane and dodecane are commonly utilized as substitutes for fossil diesel due to their prevalence in fossil diesel. The goal of this study is to employ different machine learning techniques in order to develop predictive models for the density of fatty acid methyl esters mixed with alkanes as biodiesel using experimental data. The machine learning methods utilized include Adaptive Boosting (AB), Random Forest (RF), Decision Tree (DT), Convolutional Neural Network (CNN), Ensemble Learning (EL), Multilayer Perceptron Artificial Neural Network (MLP-ANN) and Support Vector Machine (SVM). Various statistical metrics and visual methods serve as indicators of accuracy performance. The findings indicate that almost all the collected data points are appropriate for building the model. It is shown that decane mole fraction is the most influential factor on the density. The assessment demonstrated that CNN and SVR are the most precise intelligent models due to the emerged highest R-squared values (0.999, and 0.998, respectively, for the testing phase), lowest mean square error (1.41 and 2.76, respectively, for the testing phase), lowest average absolute relative error (0.125 % and 0.136 %, respectively, for the testing phase), and accurate trend forecasting of density as a function of input parameters. The developed models of CNN and SVR also outperform Tammann−Tait correlation in terms of accuracy and robustness.