Abstract
The emission of soot particles from fossil fuel combustion has a major negative effect on the environment, human health, and the global climate. The negative impacts of these have led to ongoing efforts to minimize the soot formation from burning carbon-based fuels. To understand the means of effectively reducing soot formation, it is necessary to understand the relationship between chemical structure and soot formation. A significant interest in machine learning algorithm capabilities to predict chemical and physical phenomena led to the development and deployment of various predictive models. In this study, two models were developed to predict the yield sooting index (YSI) (i.e., Artificial Neural Network (ANN) and Adaptive-Network-based Fuzzy Inference Systems (ANFIS) models). YSI measurements of two hundred and ninety-four fuels were used for preparing a dataset. The features consist of eight functional groups, molecular weight (MW), and branching index (BI). The functional groups are paraffinic CH groups, paraffinic CH2 groups, paraffinic CH3 groups, olefins, naphthene, alcohol, aromatics, and ethers. The developed ANN model has one hidden layer that consists of 8 neurons. This model was validated against 15% of the dataset and the coefficient of determination (R-2) obtained was equal to 0.9922. Similarly, the ANFIS model was validated against 15% of the dataset and resulted in 0.9543. Comparing both models to each other, although the ANFIS model generated a highly descriptive model for the training set, the ANN managed to generalize better on the test set. The prediction results for the ANN model show good accuracy concluding that the use of these features is effective in predicting YSI.