Surrogate-Assisted Multi-Objective Optimization of Engine-Level Pneumatic Assistance for Arctic Diesel Generators

Diesel generators remain indispensable for supplying reliable electricity in isolated Arctic microgrids, where harsh environmental conditions, high fuel transportation costs, and stringent emission requirements motivate improvements in engine performance and environmental sustainability. This study proposes and validates a unified, surrogate-assisted tri-objective optimization framework to calibrate the engine operation of a pneumatically assisted CAT 3516 diesel generator under representative Arctic conditions at Raglan Mine, Canada. The optimization focuses on engine-level pneumatic assistance, while compressed air generation, storage, and distribution are outside the scope of the study. The framework integrates high-fidelity Diesel-RK simulations, adaptive Sobol sampling, Gaussian Process surrogate modeling with predictive uncertainty, global sensitivity analysis, deterministic multi-objective optimization, SIMUS-based Pareto ranking, and independent high-fidelity validation. We optimized three engine-control variables (start of injection, compressor pressure ratio, and fuel mass per cycle) simultaneously to maximize brake power while minimizing brake-specific fuel consumption and NOx emissions. The surrogate models achieved coefficients of determination exceeding 0.97 for brake power, brake-specific fuel consumption, and peak cylinder pressure, while the NOx surrogate reached an R2 of 0.82. The selected operating point reduced NOx emissions by 26.9%, increased brake power by 3.7%, and increased brake-specific fuel consumption by only 0.87% relative to the Arctic baseline. The proposed uncertainty-aware framework provides a computationally efficient and reproducible methodology for engine-level calibration of pneumatically assisted diesel generators operating in isolated Arctic hybrid energy systems.