Research on injection parameter optimization of a dual-fuel engine based on NPSO-BPNN and NSGA-III

In gasoline/diesel dual-fuel engines, injection parameters exhibit highly nonlinear and strongly coupled effects on combustion and emission performances, which are further modulated by variations in engine load and speed. This study develops a data-driven multi-objective optimization methodology aimed at simultaneously optimizing fuel economy and emissions through precise regulation of dual-fuel injection parameters. First, a load-speed stratified Latin hypercube sampling method is proposed to conduct engine experiments. Subsequently, a nonlinearly improved particle swarm optimization (NPSO) algorithm is synergistically integrated with a backpropagation neural network (BPNN) to construct a high-fidelity surrogate model. Engine load and speed serve as operating-condition inputs, while fuel injection parameters act as control inputs; the optimization metrics include brake specific fuel consumption (BSFC), nitrogen oxides (NOx), carbon monoxide (CO), and ringing intensity (RI), where RI is constrained for operational safety. Model validation demonstrates exceptional predictive accuracy and generalization capability, with coefficients of determination exceeding 0.96 for all performance indicators. Subsequently, the non-dominated sorting genetic algorithm III (NSGA-III) is employed to perform Pareto optimization. Experimental validation against the baseline case confirms that the optimized strategy simultaneously improves fuel economy and emissions while maintaining RI within acceptable safety thresholds. Finally, the optimal fuel injection strategy of the dual-fuel engine under various operating conditions are explored.