Prediction of Gasoline Blend Ignition Characteristics Using Machine Learning Models

Abstract

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

Advanced Progress and Prospects for Producing High-Octane Gasoline Fuel toward Market Development: State-of-the-Art and Outlook

“…In addition, the increased compression ratio of a gasoline direct injection engine causes another problematic effect�an increase in the requirements for the octane number of gasoline. 166 A high compression ratio in the cylinder leads to a reduction in the physical volume of the combustion chamber, which increases its sensitivity to the formation of carbon deposits, the accumulation of which further increases the actual compression ratio and causes a gradual increase in the requirements for fuel knock resistance. Direct injection and turbocharging technologies continue to dominate the market for new gasoline vehicles.…”

“…At the same time, a complicated and more expensive fuel supply system is noted as a disadvantage of a direct injection engine due to the presence of a high-pressure fuel pump, which leads to greater sensitivity to fuel quality in terms of lubricating properties and the presence of mechanical impurities. In addition, the increased compression ratio of a gasoline direct injection engine causes another problematic effectan increase in the requirements for the octane number of gasoline . A high compression ratio in the cylinder leads to a reduction in the physical volume of the combustion chamber, which increases its sensitivity to the formation of carbon deposits, the accumulation of which further increases the actual compression ratio and causes a gradual increase in the requirements for fuel knock resistance.…”

A Machine Learning Model for Predicting Composition of Catalytic Coprocessing Products from Molecular Beam Mass Spectra

“…Here, we focus on predicting the composition of VPU products with fluid catalytic cracking (FCC) catalysts, both in a fixed bed and Davison Circulating Riser (DCR) reactor equipped at National Renewable Energy Laboratory (NREL). , These methods are easily generalizable to predict the detailed composition of hydrocarbon types of VPU products ( n -paraffin, iso -paraffin, olefin, naphthene, and aromaticPIONA) from hot vapor phase mass spectra. PIONA analysis is a technique widely used in refineries to classify hydrocarbons using gas chromatography (GC). ,, Previous studies suggested that the PIONA composition can be used in ML model training to predict nonlinear trends of fuel blends’ physical and ignition properties based on their compositions. − Nevertheless, no ML model is reported for quantitative PIONA assessment in VPU bio-oil products, particularly for online monitoring. Development of a robust ML model to predict the PIONA composition using MBMS would enable integrated and automated process control tools for catalytic vapor phase bio-oil upgrading.…”

Surrogate-Assisted Evolutionary Optimization of CO2-ESGR and Storage

“…Vo Thanh et al 17 simulated 351 data samples with geological and well operating factors as uncertainty parameters and established an ANN model to predict oil recovery and CO 2 sequestration in the residual oil zone. Correa Gonzalez et al 18 employed 243 sample points of gasoline blend fuel ignition for training and testing the machine learning models to predict the fuel ignition characteristics, achieving a satisfactory accuracy. Menon and Krishnasamy 19 used 35 data points to train an ANN for predicting the engine parameters, and approximately 10 data points were used for validation.…”