Thermal management strategy for improving DPF active regeneration performance via multi-objective optimization based on a random forest surrogate model

Abstract

Active regeneration of diesel particulate filters (DPFs) is essential for particulate emission control but involves competing effects among soot oxidation, heat removal, flow resistance, and thermal safety. A one dimensional transient Corning EX 80 DPF model was validated with a mean absolute percentage error (MAPE) of 2.24% for pressure drop and a relative error of 3.09% for peak temperature. Four random forest surrogate models trained on 1993 cases achieved R 2 values above 0.98 and MAPE values below 4%. The surrogates were coupled with the non-dominated sorting genetic algorithm III (NSGA-III) and technique for order preference by similarity to ideal solution (TOPSIS), and selected solutions were verified using the original physical model. Transient analysis showed that the temperature rise under low flow conditions resulted mainly from reduced convective heat removal rather than faster soot oxidation. Exhaust mass flow rate primarily affected regeneration time and pressure drop, whereas initial soot loading and inlet O 2 mass fraction had stronger effects on peak substrate temperature. Comparison of the full design space with the Pareto region showed that variable importance could change within preferred operating conditions. Fixed and variable soot loading optimization cases were also compared to distinguish regeneration execution from trigger selection. Optimization identified approximately 89 g of soot loading as a suitable regeneration trigger. Compared with the baseline, the selected low oxygen strategy reduced axial midpoint peak substrate temperature by 74.88 K, while increasing 95% regeneration time and theoretical gas sensible heat requirement by 16.2% and 23.4%, respectively.