<p>Self-switching between the hybrid electric vehicle (HEV) internal combustion engine and electric motor is a difficult task for drivers. To maximize fuel consumption and drive range in the vehicles, optimize the Energy Management in Hybrid Electric Vehicles. It is decision-making the old way with intricate trade-offs that are hard to optimize in real time online. This paper introduces Smart Energy Management for Maximizing Performance in Hybrid Electric Vehicles that uses Deep Reinforcement Learning (DRL) to remove human intervention in making decisions between the electric motor and the internal combustion engine. By simulating hybrid vehicle physical dynamics and using the Deep Deterministic Policy Gradient (DDPG) algorithm, the system is trained to make instant decisions optimizing fuel consumption without triggering a loss in vehicle performance and battery life. The given structure can support fluctuating traffic and climatic conditions, i.e., problems of energy wastage, sudden fluctuation, and unstable power supply. The intelligent agent dynamically allocates engine power and electrical loading based on true vehicle speed, acceleration, and battery state-of-charge, hence optimal use of energy. Comparison with representative driving cycles depicts dramatic progresses in fuel efficiency and responsiveness of the system compared to traditional rule-based methods. DDPG shows consistent performance across diverse driving patterns indicating robust adaptability. DQL exhibits higher variance in fuel consumption (from 3.11 to 4.41 L/100Km), suggesting it is more sensitive to driving cycle complexity and less stable. Demonstrated an average <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(9\%\)</EquationSource> </InlineEquation> improvement over DQL in fuel efficiency. This study introduces a novel driving strategy that balances resource consumption, power responsiveness, and battery longevity. Utilizing an end-to-end reinforcement learning framework integrated with a high-fidelity vehicle dynamics model, the approach employs DDPG for continuous control. It delivers a <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(12.4\times\)</EquationSource> </InlineEquation> improvement in fuel efficiency compared to rule-based techniques, marking a significant advancement beyond conventional DRL methods. The proposed system maintains the state of charge (SOC) within an optimal <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(5\%\)</EquationSource> </InlineEquation> range, with performance deviating less than <InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(2\%\)</EquationSource> </InlineEquation> from ideal values across diverse and dynamic driving scenarios.</p>

Dynamic energy allocation models to improve performance for hybrid vehicle efficiency

期刊论文
发表日期：

2026年3月30日

查看全文

Varsha Kushwah,
Umesh Manikanta Dhulipalla,
Venkata Nikhil Krishna,
Kushagra Vyas,
Sreeram Kethavath

摘要

Self-switching between the hybrid electric vehicle (HEV) internal combustion engine and electric motor is a difficult task for drivers. To maximize fuel consumption and drive range in the vehicles, optimize the Energy Management in Hybrid Electric Vehicles. It is decision-making the old way with intricate trade-offs that are hard to optimize in real time online. This paper introduces Smart Energy Management for Maximizing Performance in Hybrid Electric Vehicles that uses Deep Reinforcement Learning (DRL) to remove human intervention in making decisions between the electric motor and the internal combustion engine. By simulating hybrid vehicle physical dynamics and using the Deep Deterministic Policy Gradient (DDPG) algorithm, the system is trained to make instant decisions optimizing fuel consumption without triggering a loss in vehicle performance and battery life. The given structure can support fluctuating traffic and climatic conditions, i.e., problems of energy wastage, sudden fluctuation, and unstable power supply. The intelligent agent dynamically allocates engine power and electrical loading based on true vehicle speed, acceleration, and battery state-of-charge, hence optimal use of energy. Comparison with representative driving cycles depicts dramatic progresses in fuel efficiency and responsiveness of the system compared to traditional rule-based methods. DDPG shows consistent performance across diverse driving patterns indicating robust adaptability. DQL exhibits higher variance in fuel consumption (from 3.11 to 4.41 L/100Km), suggesting it is more sensitive to driving cycle complexity and less stable. Demonstrated an average \(9\%\) improvement over DQL in fuel efficiency. This study introduces a novel driving strategy that balances resource consumption, power responsiveness, and battery longevity. Utilizing an end-to-end reinforcement learning framework integrated with a high-fidelity vehicle dynamics model, the approach employs DDPG for continuous control. It delivers a \(12.4\times\) improvement in fuel efficiency compared to rule-based techniques, marking a significant advancement beyond conventional DRL methods. The proposed system maintains the state of charge (SOC) within an optimal \(5\%\) range, with performance deviating less than \(2\%\) from ideal values across diverse and dynamic driving scenarios.