Multi-objective prediction and optimization of alcohol–gasoline SI engines using a hybrid gradient boosting and evolutionary algorithm framework

Abstract

Alcohol-blended fuels in spark-ignition (SI) engines offer an efficient way to enhance engine efficiency and reduce emissions. This study integrates experimental engine testing with a three-level data-driven model; Gradient Boosting to perform multi-output predictions, and Response Surface Methodology (RSM) and NSGA-III to determine operating points that offer a balanced outcome between performance and emissions in 1-propanol-gasoline blends. The multi-channel tests of an SI engine of one cylinder (1700–3800 rpm; two load settings) produced eight simultaneous responses: torque, brake power, BSFC, BTE, CO, CO₂, HC, and NOₓ. The trained model was very accurate on most of the targets (e.g., BTE test R²: 0.95; BP test R²: 0.99). Both RSM and NSGA-III converged near 18% propanol and 2900–2950 rpm, with comparable Pareto-optimal performance. The 18% propanol mixture gave significant boosts as compared to pure gasoline at the same operating conditions: brake power was up by 31%, torque by 29%, CO and HC emissions were both reduced by 37%, and BSFC had risen by 19%. In conjunction with GB-RSM-NSGA-III methods, there exists a clear, repeatable approach to optimizing multiple objectives in alcohol blends for SI engines to promote clean-burning modes to fulfill SDG Goal 7 and SDG Goal 13.

Keywords: Biofuels, Engine characteristics, Gradient boosting, machine learning, NSGA-III, Optimization, RSM