Energy
Energy Transition: Optimising Electric Vehicle Charging
Date:
Changed on 20/12/2024
France is aiming for 33% renewable energies by 2030, an ambitious goal that requires efficient management of fluctuations in power generation from solar and wind sources, for example. In this context, EDF’s R&D Centre and the DISCO team at the Saclay Inria Centre have joined forces to develop innovative optimisation solutions that will allow EV charging facilities to adapt to the needs of the power grid.
The French National Strategy for Ecological Transition aims to reach 33% of renewable energies by 2030. By 2028, installed capacity is set to increase from 15 to 45 GW for photovoltaic sources, and from 24 to 35 GW for wind energy. But the power produced by solar panels and wind turbines, whether they are in Finistère, Lozère, Lubéron or Moselle, constantly fluctuates with varying weather conditions.
At the same time, forecasts estimate that in 2028 there will be 4.8 million electric vehicles on the roads of France, with drivers arriving in their thousands to recharge them at suitably equipped car parks.
The operator EDF faces a considerable challenge: how to handle this instability and channel electricity to the right place at the right time, without the need for it to be transmitted all the way across the grid? This is where the collaboration dating from 2020, when the DISCO project team from the Saclay Inria Centre (working jointly for CentraleSupélec and CNRS) teamed up with the OSIRIS Department (Energy Market Optimisation, Simulation, Risks & Statistics) from the EDF R&D Centre in Palaiseau, comes in.
Smart Charging, for Enhanced Energy Flexibility
To successfully manage this instability, Inria and the EDF R&D Centre are looking at the smart charging approach. Smart charging involves the use of algorithms to optimise the times of day and the durations of electric appliance charging in relation to other criteria such as user behaviour, power availability and associated costs. The idea is to keep environmental and economic impacts to a minimum, while maintaining fluidity in the service to users.
Rather than an EV starting to charge as soon as the user requests it, the operation can begin a few minutes or even a few hours later, as soon as the overall demand for power decreases, reducing pressure on the grid.
However, for EDF to reap the full benefit of this flexibility, guarantees have to be provided and certain constraints observed, which involves some complex mathematical calculations.
Modelling Collective Decisions
Image
Verbatim
From a mathematical perspective, the huge number of vehicles makes the optimisation equation much more difficult to solve. Our approach focuses mainly on the proportion of vehicles being charged in a given time slot and on adapted behaviour patterns.
Auteur
Poste
Inria researcher and member of the DISCO team
“This application context”, Laurent Pfeiffer continues, “echoes a mathematical theory that has been developed considerably in recent years, especially in France: the mean-field game theory.” This theory is used to model situations involving a large number of agents. Each agent adopts a behaviour, but interactions between agents create mutual dependence, the decisions of each influencing those of the others.
To improve programming possibilities, increase execution speed and ensure confidentiality, the DISCO researchers have also contributed to the choice of the Frank-Wolfe algorithm as a tool to break down the tasks, among other functions. This means the work can be divided up and resolutions reached simultaneously, enhancing both the speed and performance of the process. This breakdown will be applied at e-vehicle car parks. “Each iteration of the algorithm is based on separate calculations: one type of calculation for vehicles of model X, another for model Y, and so on.”
Upscaling the Prototype
When this collaborative work began, the research focused on a very abstract expression of the problem, containing a set of mathematical variables rather than representations of the vehicles. In a second phase, the researchers developed a simplified, but more concrete model. “Our variables related to the charge level of each battery over time, we had vehicles of different types and we knew when they were static, consuming power or feeding it back into the grid, etc.” With 2,000 cars being charged, this research prototype demonstrated its upscale potential, an area where commercial software fails even before the batteries are half charged.
“With EDF, the industrial context is an advantage for our work,” concludes Laurent Pfeiffer. It is a source of inspiration that gives us food for thought and drives us to tackle real-life problems.”
As for large-scale roll-out, the French transmission system operator, RTE, estimates that, on the basis of 15.6 million electric vehicles being on the road in 2035, smart charging “can help France to save around 900 million euros per year”.
The Inria-EDF Challenge
This work is being followed up under a broader framework. Indeed, DISCO is one of ten teams involved in the Inria-EDF Challenge entitled ‘Managing Tomorrow’s Power Systems’. Launched in 2023, this cross-disciplinary research initiative aims to design a set of innovative management tools for EDF to implement in its transition to renewable energies.
DISCO Project Team and Control of Complex Systems
The DISCO team has just been granted a four-year extension, to work on the control of interconnected complex systems. It is led by Catherine Bonnet, a Director of Research who specialises in automated systems. “Historically, we’ve always been interested in what we call closed-loop control.” In other words, a system in which the controller compares what is generated to what is required. It therefore reacts to events as and when they occur. So it adapts the command to meet the objective, even in the event of communication delays, thanks to the team’s progress in this area.
Inversely, in the optimisation issues studied by Laurent Pfeiffer with EDF, the latter relies on an open-loop control. Everything is predetermined. When an unforeseen event occurs, we cannot intervene. “What we want to do now is combine these two methods based on the expertise of other researchers in the team in terms of closed-loop control of dynamic systems.”
Building on their expertise, the DISCO team focus their research on the reality on the ground, but through a strongly mathematical approach. “We take an application problem. We formalise it. We solve it. This way, we provide solutions for categories of problems that can then be useful in various fields.”
DISCO is made up of 9 permanent members - Frédéric Bonnans, Catherine Bonnet, Islam Boussaada, Ziad Kobeissi, Guilherme Mazanti, Frédéric Mazenc, Silviu-Iulian Niculescu, Laurent Pfeiffer, Giorgio Valmorbida, and an associate member, Ali Zemouche