Fuel economy-emission trade-off optimization for diesel/natural gas dual-fuel engine using many-objective many-population hybrid genetic algorithm

Abstract

Enhancing the economy-emission balance of diesel/natural gas dual-fuel engines (DNGDFEs) remains a key bottleneck to their wide-scale adoption. To confront this challenge, this study introduced an integrated data-driven optimization framework that unifies high-fidelity physics, machine learning surrogates and a many-objective evolutionary algorithm. A multiphysics coupled simulation model was first developed and rigorously calibrated against experimental data, and an eXtreme Gradient Boosting (XGBoost) model was established via automated batch-simulation framework. On this basis, the many-objective many-population hybrid genetic algorithm (MMHGA) was proposed to optimize both economic and environmental metrics concurrently. The results demonstrate that the XGBoost model achieves excellent predictive accuracy, with R² values of 0.96134, 0.99846 and 0.99835 for brake specific fuel consumption (BSFC), carbon monoxide (CO) and hydrocarbon (HC) emissions, respectively. Across standard benchmark problems, MMHGA consistently surpassed competing optimizers through faster improvement in solution quality, robust convergence and superior Pareto-front uniformity. Among the non-dominated solutions, the fourth-ranked candidate provides the best overall compromise, simultaneously reducing BSFC, CO and HC by 5.16 %, 1.55 % and 2.95 %, respectively. These findings confirm that coupling a machine-learning surrogate with MMHGA offers a computationally efficient and practically viable route for many-objective optimization of DNGDFE operating strategies, offering immediate guidance for low-carbon transport applications.