Predictive energy management of heavy vehicle in uncertain traffic with learning-based optimal control

Abstract

Heavy-duty vehicles (HDVs) contribute to over 30% of CO2 emissions despite representing only 4% of the vehicles on the road in the European Union. To address this imbalance, eco-driving strategies have been developed to improve the energy efficiency of HDVs and reduce their environmental impact. This thesis focuses on a specific scenario in which an ego HDV follows a leading vehicle (another HDV or a passenger car). By analyzing the longitudinal dynamics of the leading vehicle using its speed measurements and road slope information from the onboard digital map, the goal is to accurately estimate the leading vehicle’s future speed trajectory. Combined with a predictive energy management controller onboard the ego vehicle, the system can optimize acceleration and deceleration to minimize braking events and maintain a safe inter-vehicle distance, thereby reducing overall fuel consumption.

To achieve this, a learning-based observer–predictor architecture is proposed. The observer filters noisy speed measurements and fuses them with road slope data to estimate the leading vehicle’s maximum power capability in real time. The predictor then uses this estimate, along with future slope information within the horizon, to generate the future speed trajectory of the leading vehicle. This speed prediction is used by the ego vehicle’s predictive controller to follow the leading vehicle with reduced air drag, enabling energy-efficient and safe following behavior.

The development and validation of the framework were carried out in MATLAB. Performance was evaluated using both simulation and real-world data from recorded driving cycles. The results demonstrate that the proposed structure can predict the leading vehicle’s speed trajectory with good accuracy and has the potential to enable significant improvements in energy efficiency and overall system performance. The main contribution of this thesis is the development of a learning-based observer– predictor structure that can predict the leading vehicle’s future speed trajectory to help the ego vehicle plan its own speed in a more energy-efficient manner.