Prediction of Gasoline Blend Ignition Characteristics Using Machine Learning Models

Research Octane Number (RON), among other autoignition related properties, is a primary indicator of the grade spark-ignition (SI) fuels. However, in many cases, blending various gasoline components affects RON final fuel product nonlinear way. Currently, lack precise predictive models for challenges accurate and production commercial SI This study compares popular Machine Learning (ML) algorithms evaluates their potential to develop state-of-the-art able predict key properties. Typical composition was simplified represented by palette seven characteristic molecules, including five hydrocarbons two oxygenated species. Ordinary Least Squares (OLS), Nearest Neighbors (NN), Support Vector Machines (SVM), Decision Trees (DT), Random Forest (RF) were trained, cross-validated, tested using database containing 243 gasoline-like blends with known RON. Best results obtained SVM reproduce synergistic antagonistic molecular interactions. The Mean Absolute Error (MAE) on test set equal 0.9, estimator maintained its accuracy when alterations performed training data set. Linear methods better molar compositions while predictions volumetric basis required satisfactory accuracy. Developed allow one quantify behavior different oxygenates accounting those effects during production. Moreover, these contribute deeper understanding phenomena that will facilitate introduction alternative recipes components.