TY - JOUR
T1 - Integrated energy and thermal management of range-extended electric vehicles using causal global distillation and Koopman-based model predictive control
AU - Gao, Haoming
AU - Sun, Chao
AU - Song, Xupeng
AU - Yang, Guodong
AU - Zhang, Xinyang
AU - Leng, Jianghao
AU - Yang, Dongpo
N1 - Publisher Copyright:
© 2026 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
PY - 2026/12/1
Y1 - 2026/12/1
N2 - At low ambient temperatures, energy management in range-extended electric vehicles (REEVs) must coordinate fuel consumption, battery energy, component warm-up, recoverable engine heat, and cabin heating throughout a trip. Existing real-time controllers struggle to combine trip-level resource allocation with accurate and computationally efficient prediction of coupled electrothermal dynamics because nonlinear interactions, differing electrical and thermal time scales, and limited preview information increase the computational burden of constrained online optimization. This study proposes global-distillation Koopman model predictive control (GDK-MPC). First, a control-oriented electrothermal model integrates range-extender fuel consumption and coolant heat generation, temperature-dependent battery behavior, and heat recovery, exposing energy-thermal interactions under operational constraints. Second, a residual Koopman predictor represents nonlinear electrothermal dynamics in a fixed affine form, improving multistep prediction while preserving the computational efficiency and explicit constraints required for real-time MPC implementation. Third, causal global distillation transfers trip-level resource-allocation priorities from dynamic programming (DP) into adaptive online MPC optimization, allowing the controller to exploit global knowledge without requiring a complete future driving profile. With a 10-s horizon, the Koopman predictor reduced normalized composite prediction error by 23.62-30.92% relative to a fixed linear model, with a prediction time of 0.696-0.729 ms. Across CLTC, UDDS, and NEDC, GDK-MPC reduced equivalent fuel consumption by 0.07-6.70% relative to fixed-weight guided MPC and remained 1.65-2.91% above DP. During CLTC, battery and coolant temperatures remained within approximately −10 to −9.2 °C and −10–90 °C, respectively. Testing on a held-out real-world cycle supported transfer. HIL tests demonstrated closed-loop performance on actual automotive controller hardware.
AB - At low ambient temperatures, energy management in range-extended electric vehicles (REEVs) must coordinate fuel consumption, battery energy, component warm-up, recoverable engine heat, and cabin heating throughout a trip. Existing real-time controllers struggle to combine trip-level resource allocation with accurate and computationally efficient prediction of coupled electrothermal dynamics because nonlinear interactions, differing electrical and thermal time scales, and limited preview information increase the computational burden of constrained online optimization. This study proposes global-distillation Koopman model predictive control (GDK-MPC). First, a control-oriented electrothermal model integrates range-extender fuel consumption and coolant heat generation, temperature-dependent battery behavior, and heat recovery, exposing energy-thermal interactions under operational constraints. Second, a residual Koopman predictor represents nonlinear electrothermal dynamics in a fixed affine form, improving multistep prediction while preserving the computational efficiency and explicit constraints required for real-time MPC implementation. Third, causal global distillation transfers trip-level resource-allocation priorities from dynamic programming (DP) into adaptive online MPC optimization, allowing the controller to exploit global knowledge without requiring a complete future driving profile. With a 10-s horizon, the Koopman predictor reduced normalized composite prediction error by 23.62-30.92% relative to a fixed linear model, with a prediction time of 0.696-0.729 ms. Across CLTC, UDDS, and NEDC, GDK-MPC reduced equivalent fuel consumption by 0.07-6.70% relative to fixed-weight guided MPC and remained 1.65-2.91% above DP. During CLTC, battery and coolant temperatures remained within approximately −10 to −9.2 °C and −10–90 °C, respectively. Testing on a held-out real-world cycle supported transfer. HIL tests demonstrated closed-loop performance on actual automotive controller hardware.
KW - Dynamic programming
KW - Energy management
KW - Hybrid electric vehicles
KW - Model predictive control
UR - https://www.scopus.com/pages/publications/105054469280
U2 - 10.1016/j.energy.2026.142939
DO - 10.1016/j.energy.2026.142939
M3 - Article
AN - SCOPUS:105054469280
SN - 0360-5442
VL - 367
JO - Energy
JF - Energy
M1 - 142939
ER -