On the rational formulation of alternative fuels: melting point and net heat of combustion predictions for fuel compounds using machine learning methods

Abstract

We report the development of predictive models for two fuel specifications: melting points (T(m)) and net heat of combustion (Δ(c)H). Compounds inside the scope of these models are those likely to be found in alternative fuels, i.e. hydrocarbons, alcohols and esters. Experimental T(m) and Δ(c)H values for these types of molecules have been gathered to generate a unique database. Various quantitative structure-property relationship (QSPR) approaches have been used to build models, ranging from methods leading to multi-linear models such as genetic function approximation (GFA), or partial least squares (PLS) to those leading to non-linear models such as feed-forward artificial neural networks (FFANN), general regression neural networks (GRNN), support vector machines (SVM), or graph machines. Except for the case of the graph machines method for which the only inputs are SMILES formulae, previously listed approaches working on molecular descriptors and functional group count descriptors were used to develop specific models for T(m) and Δ(c)H. For each property, the predictive models return slightly different responses for each molecular structure. Therefore, models labelled as 'consensus models' were built by averaging values computed with selected individual models. Predicted results were then compared with experimental data and with predictions of models in the literature.

Chemical kinetic and combustion characteristics of transportation fuels

“…At the same time, advances in applying quantitative structure property relationship (QSPR) and quantitative structure activity relationship (QSAR) techniques have led to improved predictions of cetane number [143][144][145], flashpoint [144], density and viscosity [146], melting point [147], heat of combustion [147], and laminar burning velocities of biofuels [148]. Recently, Saldana et al [149] have suggested QSPR based methods for formulating alternative fuels for specific applications.…”

Accurate prediction of the standard net heat of combustion from molecular structure

“…The data sets used in the literature to model the DH o c of pure compounds (Cardozo, 1986;Gharagheizi, 2008;Pan et al, 2011;Cao and Wang, 2013;Seaton and Harrison, 1990;Hshieh, 1999;Hshieh et al, 2003;Diallo et al, 2012;Wang and Li, 2000;Van Krevelen, 1990;National Technical Information Service (NTIS), 2001;Gharagheizi et al, 2011;Cao et al, 2009;Saldana et al, 2013;Albahri, 2013a) are so different that these models cannot be compared to one another. However, Table 2 shows that our proposed SGC-ANN model gives good results compared to the other models used to calculate DH o c .…”

“…Saldana et al (Saldana et al, 2013) combined several methods to create a consensus model for predicting the DH o c of 1624 hydrocarbon-based compounds and 1143 oxygenates (alcohols and esters) using QSPR. Various approaches were investigated from linear modeling: Genetic Function Approximation (GFA) and Partial Least Squares (PLS) to nonlinear models, such as FeedForward Artificial Neural Network (FFANN), General Regression Neural Networks (GRNN), Support Vector Mechanics (SVM), and Graph Machines (GM).…”

“…All of the individual models have AAE of less than 2% except for the GRNN based model, which has an AAE of 4%. Therefore, Saldana et al (2013) developed a consensus model by averaging the values computed with selected individual models to improve the generality and predictive power compared to individual predictive models. The robust consensus model, which is more accurate than each individual model alone, can predict DH o c with R ¼ 0.999, AAE ¼ 0.7%, and AAD ¼ 33.2 kJ/mol.…”

Fast learning method for RAAM based on sensitivity analysis

“…15 A different model, the Graph Machines, 16 was developed for QSAR 17 and was recently used for predicting fuel combustion properties. 18 This article presents a combination of the fundamental graph processing connectionist model, RAAM, with the Sensitivity Based Linear Learning Method, which resulted in a model that can be efficiently trained in a very few iterations…”