Multi-objective optimization of hybrid agricultural powertrain via crowding-adaptive NSGA-II with dynamic population control

The pressures of the global energy crisis and environmental pollution have driven the rapid development of green-powered agricultural machinery. However, dedicated hybrid engines still face technical bottlenecks in emission-fuel consumption co-optimization. In response to the limitations in traditional optimization approaches regarding operational condition selection, this study proposes a hybrid model multi-objective optimization framework based on feature extraction for off-road operating conditions. Firstly, the present study adopts a multi-platform collaborative modelling approach. This is employed in order to develop a high-precision engine model and a BP neural network emission prediction model on the GT-Power and Simulink platforms, respectively. Thereafter, a joint simulation system with dynamic coupling features on the Simulink platform is established through a modular interface. Secondly, high-frequency optimization points are screened based on the off-road operating conditions feature extraction method, and the computational volume is significantly reduced by 94.2 % compared with the full domain optimization search scheme of the complete engine universal characteristic map. Finally, the Crowding-Adaptive NSGA-II with Dynamic Population Control algorithm (CADPC-NSGA-II) with elite mating strategy based on dynamic fitness difference is proposed, which effectively alleviates the problem of genetic diversity decay under small-scale population. Experimental results demonstrate 8.55 % fuel consumption reduction, 5.31 % exhaust smoke opacity decrease, and 17.61 % NOx emissions decline in optimized hybrid agricultural machinery. This study provides practical insights and technical references for the optimization of core components in hybrid agricultural machinery.