Deciding Under Uncertainty: Smarter EV Charging with MPC and Reinforcement Learning
Electric vehicles are no longer just a growing load on the grid. Because their charging can be shifted throughout the time a vehicle remains connected, they have become a valuable source of flexibility for buildings and the wider energy system. The challenge is not creating flexibility, but using it effectively.
Information such as departure times, energy requirements, and electricity prices is not always known in advance. While optimisation techniques can determine efficient charging schedules when these inputs are available, real-world operation often involves significant uncertainty.
As part of BlueBird, Giuseppe Gabriele, PhD researcher at IDLab (UGent - imec) and member of the AI4 Energy research team, researches AI techniques to improve EV charging decision-making under such uncertainty.
When reliable forecasts are available, Model Predictive Control (MPC) provides an effective solution by continuously optimizing charging decisions over a future horizon.
Not all uncertainties, however, can be addressed through forecasting. User behaviour, departure times, market conditions, and system dynamics can evolve in ways that are difficult to predict accurately. In these situations, Reinforcement Learning (RL) offers a powerful alternative. Rather than relying on an explicit model of the future, an RL agent learns through interaction with its environment, using feedback from previous decisions to improve its control strategy over time.
Giuseppe's main research focuses on applying RL to EV charging problems where critical information is unavailable at the time decisions must be made. This includes situations with uncertain vehicle departure times and participation in electricity markets where prices, costs, or revenues may only become known after decisions have already been taken. At the same time, charging decisions must continue to satisfy operational constraints and user requirements. In both cases, the challenge is the same: making the best possible decision today without fully knowing what tomorrow will bring.
Beyond these research challenges, Giuseppe also contributes to practical optimisation use cases within BlueBird. With Elektrizitätswerk Hindelang eG, EV charging flexibility is leveraged to reduce charging costs by adapting charging schedules to day-ahead electricity market prices. At WeSmart, EV charging is coordinated together with battery storage, PV generation, and building electricity consumption to optimise the operation of the overall energy system.
By combining forecasting, optimisation, and machine learning, BlueBird aims to transform uncertainty into intelligent real-time decisions, helping unlock the full value of EV flexibility and supporting the transition towards a more sustainable energy system.
