In recent years, the rapid growth of the electric vehicle market has slowed because of chasm phenomenon. As a result, Hebrid Electric Vehicles (HEVs) are receiving renewed attention as a practical alternative. HEVs can be categorized into series, para...
In recent years, the rapid growth of the electric vehicle market has slowed because of chasm phenomenon. As a result, Hebrid Electric Vehicles (HEVs) are receiving renewed attention as a practical alternative. HEVs can be categorized into series, parallel, power-split configurations, among which the parallel architecture employs both an internal combustion engine and an electric motor as power sources. Depending on the motor position, parallel hybrids are further classified into configuration ranging from P0 to P4. This study adopts the P2 layout – one of the most widely commercialized architectures – in which the electric motor is positioned between the engine and the transmission. In the P2 configuration, the engine and motor can be operate independently, and therefore, vehicle performance and fuel efficiency are highly dependent on the applied control strategy. Consequently, establishing an effective Energy Management Strategy (EMS) is a ctrical facotr in determining the overall efficiency of such systems.
This study applies Model-Based Reinforcement Learning (MBRL) to design an optimal power-split control strategy and performs a comparative analysis with a conventional rule-based strategy. MBRL offers advantages over Model-Free Reinforcement Learning, including higher sample efficiency, faster computation, reduced uncertainty during training, and stronger real-time applicability—features that are particularly beneficial in complex and unpredictable real-world driving environments. In contrast, rule-based control provides simplicity and consistent performance regardless of driving conditions but shows limitations in capturing diverse operational scenarios. To examine the characteristics, limitations, and fuel-economy differences between the two approaches, both control strategies were evaluated under identical simulation conditions.
A forward-looking HEV simulation environment was developed in MATLAB/Simulink based on a detailed vehicle model. Using standard driving cycles, which are UDDS and HWFET, the performance of MBRL-based control and the conventional rule-based strategy was compared. The proposed approach incorporates the Dyna-Q algorithm, and extensive cost-function tuning was conducted to optimize fuel economy. The results demonstrate that the MBRL-based strategy achieves superior fuel-efficiency performance compared to the rule-based control method.