*4.4. Modeling Customer Behavior*

In our basic scenario, we modeled EV users via their arrival time *tA* and departure time *tD* at the charging station, plus their average energy demand (specifically, the SoC at arrival *SoCini*). We also assumed a fixed target SoC of 80%. For each user, the specific values for *tA*, *tD*, and *SoCini* were drawn from a normal distribution with mean 9:00 for arrival time, 17:00 for departure time, and 30% for initial state of charge at arrival. The standard deviations were selected as follows: 60 min for *tA* and *tD*and 10% for *SoCini*.

Before the start of the simulation, we first generated a set of *NE* users with the above given statistics, so that user (EV) *u* had an individual value *tA*(*u*), and so on. Based on these data, we calculated the charging demands for each EV and each day by determining arrival time, departure time, and initial SoC, again from a Gaussian distribution. The mean of the distribution was taken from the user characterization for each *u*, e.g., *tA*(*u*), with a variance that was half the inter-user variances (i.e., 30 min for *tA* and *tD* and and 5% for *SoCini*).

This means that there were both variations in the average values between users, as well as daily variations. Note that in addition to charging demands, weather conditions will differ, so that typical CSI values will vary over different days. The process for computing the daily charging demands is depicted in Figure 2. As one can see, it is also part of the user behavior that the vehicle can also be loaded externally to the company and that the user has to wait when there are not enough free charging stations. This also has an effect on *tA* and the CSI calculation.

**Figure 2.** Modeling of the charging process in the MAS.
