Reliable connectionist tools to determine biodiesel cetane number based on fatty acids methyl esters content

Abstract

The self-ignition ability of fuels is characterized using a property called cetane number (CN). The main objective of this research work is to develop accurate models using smart algorithms to predict the CNs associated with 149 samples of biofuels collected from the literature. In this study, we use two machine learning (ML) algorithms, namely least squares support vector machine (LSSVM) and extra trees (ET) and an experimental dataset containing a variety of fatty acids methyl esters (FAME). To assess the predictive performance of the developed models, various statistical parameters including average relative deviation in percentage (ARD%), average absolute relative deviation in percentage (AARD%), and coefficient of determination (R2) are calculated. The statistical and graphical analyses presented in this study reveal that the ET model outperforms the LSSVM model in predicting the CN associated with the biodiesel systems from the collected dataset so that the value of AARD% is less than 3% for the ET method. It is also found that Linoleic (18:02) is the most important factor affecting the CN value of the biodiesels. According to the sensitivity analysis, the CN value decreases upon an increase in the concentration of Palmitoleic (16:01), Oleic (18:01), Linoleic (18:02), and Linolenic (18:03); this is in agreement with the experimental studies. The smart modelling techniques used in this study can be utilized as alternative methods to direct measurements for determination of biodiesel CN.