This page visualizes an algorithm for online electrical vehicle charging, as presented in Section 3 of
Gerards, M.E.T.; Hurink, J.L. Robust Peak-Shaving for a Neighborhood with Electric Vehicles. Energies 2016, 9, 594 (Open Access).
Problem statement
Electrical vehicles impose a high load on the electricity grid.
Although nowadays most neighborhoods have few electrical vehicles and typically no problems occur, a future scenario where many vehicles are charged simultaniously may result in an overloaded grid.
In a recent field test such a future scenario was studied by charging a few vehicles and using some electrical ovens in a Dutch neighborhood, which resulted in power quality issues and a black out on a part of the local network (Dutch media coverage).
Our research from Section 3 makes it possible to shape the house profile, and can be used as a subroutine in demand-side management approaches (such as the approach presented in the paper, or profile steering) to prevent overloading or meeting another objective that requires reshaping the house profile.
Our solution
To ease the discussion, only the case where a flat profile is desired will be discussed here (for details, see the paper).
In the offline case, the solution can be characterized using a fill level.
Charging the electrical vehicle according to the optimal profile fills the total power consumption to this level.
Our approach is inspired by model predictive control, online optimization and rolling horizon planning: everytime a time interval begins, we use a prediction of the power for this interval together with a prediction of the fill level.
By making the decision only at the moment when it is needed, we are able to create more accurate predictions to be used for the upcoming interval, and errors from previous time intervals are compensated this way.
The main result from Section 3 is that we bound the objective value (deviation from the desired profile) in terms of the predicted fill level.
In this section we observe that when this prediction is too high, the negative impact on the solution quality is low.
When the prediction is too low, the electrical vehicle (EV) does not charge fast enough and should charge at a relatively high power slightly before the deadline (in this demo: at 7 kW).
Our algorithm requires only a fill level and a single power prediction as input for each interval, is easy to implement and runs in constant time.
Demonstration of the algorithm
The interactive demonstration below demonstrates the algorithm from the paper.
The top graph corresponds to the fill level, which is the level to fill to with respect to the total power in the house minus the desired power.
For details about the fill level and its interpretation, we refer to the paper.
The bottom graph shows the loads in the house, and shows the desired profile separately.
Using , a single iteration of the algorithm is demonstrated, while repeat multiple steps automatically.
To quick forward to the completion of the algorithm, press .
When the algorithm is not active (e.g., after pressing to restart), the house baseload, car and charging interval (using the slider below the graph) can be configured.
When changing the parameters, the fill level prediction is updated to a reasonable prediction (1.05 * (to be compensated energy + required charge) / (length of the charging interval)).
A new fill level prediction can be chosen by clicking in the graph, even when the simulation is active.
This demonstration allows selection of a predefined target profile.
For details on this we refer to the paper.
, a single iteration of the algorithm is demonstrated, while 
repeat multiple steps automatically.
To quick forward to the completion of the algorithm, press 
.
When the algorithm is not active (e.g., after pressing 
to restart), the house baseload, car and charging interval (using the slider below the graph) can be configured.
When changing the parameters, the fill level prediction is updated to a reasonable prediction (1.05 * (to be compensated energy + required charge) / (length of the charging interval)).
A new fill level prediction can be chosen by clicking in the graph, even when the simulation is active.
This demonstration allows selection of a predefined target profile.
For details on this we refer to the paper.