Increasing adoption of EVs in the next few decades is going to present new challenges such as EV charging creating a new and significant demand on the grid. The purpose of this thesis is to create a system that intelligently schedules the charging of EVs while considering the cost of energy and the discomfort of the user. At any given moment, 90% of vehicles are parked and have a huge energy source left unused. EVs could also be used as power sources for vehicle-to-home/vehicle-to-grid (V2H/V2G) to benefit from them during high demand of energy. This way the power plants would see almost a constant demand and usage, in the long run, making them more efficient.
This thesis uses a non-intrusive data-driven technique to create a occu- pancy and EV charging model of the household. Smart meters in each household collect power usage data. From this power usage data we de- termine occupancy and EV charge sessions. The next step is to determine temporal metrics for occupancy and EV charge sessions. The temporal met- rics study the likelihood for occupancy or an EV charge session to occur or to switch from one state to another. Because there are differences between weekday/weekend and seasonal power usage, we have decided to create tem- poral metrics for each time period.
The next step is to create the EV charging algorithm and V2H/V2G algorithms. These algorithms require a flexibility window. This window indicates in which hours the EV can be charged. Which hours of the flexib- ility window are chosen, depends on the type of objective. We have created three objectives: cost minimization, comfort maximization and joint object- ive. The V2H/V2G algorithm is executed when the state of charge (SoC) of the EV is higher than the SoC boundary.
In order to measure the performance of the algorithm, we have created two metrics: relative savings and miss rate. The miss rate measures how an hour was scheduled for EV charging but failed. During the testing of the algorithm, we found that only the objective cost minimization was deemed useful. Each objective uses a flexibility window and we conclude that the user’s preferences are already taken into account during the creation of this window. For the execution of the EV charge scheduling algorithm, a max- imum relative savings can be achieved of 27% and a maximum miss rate of 11.1%. By choosing the SoC boundary value of 60% for V2H, maximum relative savings of 9.9% and a maximum miss rate of 5.2% can be achieved. V2G execution had a negligible effect on the relative savings and miss rate because the pricing dataset did not contain many price surges.