Resourcefully

Introduction

Smart charging is a term widely used when talking about electric vehicle flexibility. Exist a lot of different methods to modify the usual EV charging session in terms of power, time or energy. From a city perspective, controlling the demand of EV will be necessary in a near future in order to avoid congestion in the power grid during peak periods. At the same time, there are different strategies to reduce the aggregated EV peak, like limiting the charging power of every EV, postponing sessions out of peak periods, reducing the total energy charged to the vehicle, etc.

Our smart charging algorithm is composed by two separate steps:

  1. Optimization: a grid flow minimization optimization is first performed to obtain the best-case optimal EV demand (i.e. setpoint) according to the solar production and the demand curve. In the case of EV, the demand can only be consumed later so a forward optimization is applied. The optimization window is set from 6 AM to 6 AM.
  2. Scheduling: all flexible sessions are postponed during the periods when the EV demand is higher than the setpoint. All sessions are ensured to charge all their energy needs even tough they are postponed (up to 6 AM).

Simulation of EV sessions

We develop stochastic EV models from real data sets provided by project partners. The novelty of our EV models is that it is not only a big black box that estimates new EV sessions, but a combination of Gaussian distributions corresponding to generic user profiles. The term user profile refers to a connection pattern or daily behavior. For example, people connecting the EV from 9:00 to 17:00 from Monday to Friday are grouped under the Worktime user profile. Therefore, from a real data set of sessions, first we cluster sessions into user profiles and then we build their Gaussian models separately. The Gaussian models estimate the 4 basic variables of a charging session:

For a project developed together with the municipality of Arnhem, we discovered the following user profiles from the real EV sessions in Arnhem:

If you want to know more about the development of these models, a wider explanation of the methodology is covered in this article.

To simulate new sessions it is necessary to define the EV charging scenario with the following data:

Even though the same EV model could be used for multiple study cases, these parameters must be accurately defined for every different case and scenario. The user profiles of a residential neighbourhood will not be the same than the user profiles of the city center, and the number of sessions per day and the charging powers distributions will also change if we simulate charging sessions in 2022 or in 2030. Let’s see and example of a simulated EV demand during a week:

The grid congestion problem becomes a bit more clear when we plot the aggregated EV demand over the rest of power demand (example of KNSM island demand in Amsterdam): EV users tend to charge during evening peak hours, so EV demand tend to increase the global power demand peak.

Most of connections have flexibility, so, instead of start charging as soon as the vehicle is connected, why not to postpone the charge when the rest of power demand is lower? Let’s do it.

Smart Charging

Optimization: setpoints for the user profiles

Every user profile has a different setpoint according to their possible contribution (flexibility potential) to create an optimal aggregated EV profile. So the optimization does not optimize every user profile independently but taking into account the rest of demand as well. Let’s take the example of Commuters and Worktime profiles.

Worktime

Visit

Shortstay

Dinner

Commuters

Home

Pillow

It is clear that the optimization allows to charge early Commuters demand, but from 18:00 all energy is postponed to late night due to the peak of the rest of power demand. On the other hand, Worktime demand is always postponed to the hours when there is some solar surplus. In the case of not being local generation the demand would be also postponed to valley hours considering the rest of the power demand. However, this is only the setpoint for the EV demand but there it could be a huge difference between the setpoint and the real flexibility potential. If all sessions were connected just during charging time it wouldn’t be possibility to shift the demand so the setpoint couldn’t be accomplished. This flexibility contraint is not considered on the optimization step, but it is on the scheduling process explained below.

Scheduling: postponing sessions

Once we have the optimal setpoint for every user profile then we can start postponing sessions (scheduling), trying to achieve the aggregated setpoint in every time slot. The scheduling algorithm consists in an iteration over all time slots and all sessions starting to every time slot. If the aggregated EV demand is higher than the setpoint, then postpone all sessions that were supposed to start at that time to the following time slot. For example, if a sessions of 11kW was supposed to start charging at 9:00 when we already have an EV demand of 40kW and the setpoint is 42kW, then we postone the session to 9:15 (considering a time reslution of 15 minutes), with the hope that the setpoint will accept this session to charge. Every time that a session is shifted, it loses flexibility potential. So if the session of the example has a connection time of 8 hours and a required charging time of 4 hours, we will not be able to shift it more than 4 hours since then it would not charge all its required energy. Therefore, in some cases we can not simply reach the setpoint and the vehicle must charge anyway.

This process is repeated until there are no more sessions to shift or no more flexibility requirements. Let’s see the algorithm applied on the example demand profiles of this article:

Worktime

Visit

Shortstay

Dinner

Commuters

Home

Pillow

In the plots above we see that flexible EV demand tries to be adapted to the setpoint, but during some hours it is impossible since there are some sessions that were already started (so it is not possible to postpone anymore). In this situation, another smart charging strategy like reducing the charging power or pausing sessions could provide better flexible results. Moreover, we see some cases where the setpoint profile can’t be filled with demand since the power of sessions can not be increased. However, even though the final result is not the optimal one, it is a realistic result and the main point of our approach. Most smart charging simulations just consider the best-case solution from the optimization step, without going deeper to simulate the scheduling in a session level.

Finally, for the aggregated view, we can clearly see that the EV demand has been shifted to the solar production or valley hours, obtaining a flatter total demand profile and increasing self-consumption: