Parameter identification of a diesel oxidation catalyst model based on segmented adaptive variable forgetting factor recursive least squares

Controlling outlet temperature during active diesel particulate filter regeneration requires accurate online identification of diesel oxidation catalyst parameters. The strong thermal inertia, nonlinear heat-transfer and reaction characteristics, and asymmetric dynamic responses of the diesel oxidation catalyst during heating and cooling make accurate model-parameter identification challenging. Fixed-forgetting-factor algorithm cannot balance rapid transient tracking and stable steady-state estimation, whereas standard variable-forgetting-factor algorithm use a single cycle wide adaptation law, neglecting regime-dependent thermodynamics. In this study, we develop a physics-constrained, control-oriented model of a diesel oxidation catalyst and propose a segmented adaptive variable-forgetting-factor recursive least-squares algorithm guided by catalyst thermodynamics. The outlet temperature derivative defines heating, cooling, and steady regimes; K-means clustering and coefficient-of-variation analysis quantitatively determine their thresholds. Regime-specific forgetting-factor ranges and prediction-error-based online adaptation balance transient tracking and steady regime noise suppression. Evaluation using engine bench data collected over the non-road transient cycle shows that the proposed algorithm achieves a coefficient of determination of 0.9994 and a root mean square error of 1.480 K. Compared with conventional variable and fixed-forgetting-factor algorithms, it reduces the root mean square error by 40.3% and 16.3%, respectively, and exhibits greater robustness to measurement noise. These results demonstrate that thermodynamically informed, regime-specific adaptation improves identification accuracy and transient tracking performance, providing a practical parameter-identification framework for model-based diesel engine aftertreatment control.