Road vehicle energy demand predictions under uncertain operating conditions

Licentiate thesis, 2026

The push for electric transport has brought new challenges to the vehicle industry, many of which stem from the limited range and long charging time. Consequently, vehicle users who regularly utilize the entire battery during their transport missions may have to endure slow charging speeds, queues, or even immobilization due to battery depletion. Even among some, the mere fear of battery depletion is enough to deter the pursuit of purchasing an electric vehicle. However, with intelligent vehicle functions like range estimators and route-planning algorithms becoming more accurate, users are enabled to make informed choices to address some of these problems. While the literature on routing algorithms is extensive, the focus has merely been on defining the optimization problem and algorithm, often using simple energy consumption models. In contrast, research in range estimation relies on rather complicated energy consumption models, which are often derived from vehicle data. These models do, unfortunately, have poor transferability between different drivers, environmental conditions, and vehicles. A great effort has thus been undertaken to model these effects in isolation, for instance, the study of rolling resistance and air drag. Building on models like those, numerous complex complete vehicle simulation models have been developed with excellent accuracy in controlled environments, but at the cost of being too computationally expensive for in-vehicle use. Additionally, these models seldom quantify uncertainty, a crucial parameter for preventing battery depletion.

To this day, the uncertainty of a range estimate is most commonly inferred from data, sensitivity analyses, or empirical model parameters. Methods relying on data or sensitivity analyses generally impose a constant uncertainty, owing to the estimation methods adopted. In contrast, using a model-based approach, for instance, derived from empirical model parameters, has the advantage of capturing dynamic characteristics that vary between transport missions. Notably, these parameters may not necessarily convey any physical meaning, but instead exist solely as internal elements of a black-box model. In contrast, by adopting a physical model-based approach, variations in energy demand can be derived from exogenous parameters like those obtained from weather, traffic, mission, and road information. This approach aligns precisely with that adopted in this thesis. Specifically, three models are formulated, all aimed at estimating energy demand uncertainty in the presence of reference‑speed perturbations. In two of the models, the reference speed is considered a measurement, permitting the use of the Luenberger observer framework. This novel approach enables the estimator to acknowledge uncertainties from parameters that exert an indirect effect on the energy consumption, typically those affecting the vehicle motion. However, even for these models, merely parameter and model-induced uncertainties were considered, with the major dynamics being known. To assess and quantify the additional uncertainty introduced by certain parameters being undefined, the same transport mission was repeatedly simulated operating under different Gross Combined Weights (GCW). The findings demonstrate that precise knowledge of the GCW is essential for reliable energy demand predictions.