Optimising driving velocity profiles is crucial for reducing vehicle fuel consumption and NOx emissions without altering core vehicle components. While many studies have addressed eco-driving, most have focused solely on minimising fuel consumption or have treated NOx emissions separately, resulting in distinct, non-integrated speed profiles, and have often neglected the influence of real-time traffic. To overcome these limitations, this paper introduces a novel Multiobjective Speed Profile Optimisation (MO-SPO) framework for eco-driving that simultaneously minimises fuel consumption, NOx emissions, and travel time while accounting for surrounding traffic. Two solution approaches are developed and compared: a two-phase Model Predictive Control (MPC) method and a newly proposed Deep Reinforcement Learning (DRL) method that directly integrates multiple objectives and real-time traffic constraints into the speed control policy.
Simulation results on a UK highway segment, with vehicle dynamics and engine characteristics derived from GT-SUITE data, demonstrate the benefits of the proposed framework. For instance, at one representative Pareto point, results indicate that the DRL approach achieves up to 10% lower fuel consumption and 16% lower NOx emissions compared to MPC-based methods while reducing travel time by approximately 5%. In addition, the DRL method maintained safer headway distances, offering more robust eco-driving strategies in dynamic traffic environments.
This work is the first to apply multiobjective optimisation to generate integrated speed profiles that consider fuel, NOx, and travel time simultaneously under realistic traffic conditions.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)