Learning-augmented optimization and control of long-haul mobility propulsion systems.

Area of Science:

• Engineering
• Sustainable Energy
• Artificial Intelligence

Background:

• Decarbonizing the long-haul transportation sector is essential for global climate goals.
• Current propulsion systems face challenges in achieving significant carbon emission reductions.

Purpose of the Study:

• To introduce a novel hierarchical, learning-augmented framework for co-designing an ammonia-hydrogen hybrid electric powertrain.
• To develop a viable carbon-free propulsion solution for long-haul transportation.

Main Methods:

• A deep reinforcement learning (DRL) agent was trained for online supervisory control of power split and hydrogen production.
• A DRL-augmented adaptive non-dominated sorting genetic algorithm (RL-ANSGA) was used for offline multi-objective component sizing.
• High-fidelity co-simulation evaluated candidate designs under a pre-trained DRL energy management policy.

Main Results:

• The optimized ammonia-hydrogen powertrain demonstrated enhanced energy efficiency.
• The proposed system achieved lower well-to-wheel carbon emissions compared to diesel and diesel-hybrid vehicles.
• The framework effectively integrated online control and offline design for powertrain optimization.

Conclusions:

• The ammonia-hydrogen hybrid electric powertrain presents a promising carbon-free alternative for long-haul transport.
• The learning-augmented co-design framework enables efficient optimization of complex hybrid powertrains.
• This approach contributes to sustainable solutions in the transportation sector.