Análisis de la Oferta

In this section, the best (in terms of linearity) historic time period to be used for the regression is investigated. In Section 3.1.6 it was described that the smaller number of historical data is used to calculate the regression function the smaller the sea- sonal changes are (e.g. the temperature changes). In this section we are looking for a time historic period of days that shows a linearity. In contrast to the previous section, when we considered intervals per day, we now differentiate the number of previous days, which are used as input data for the regression analysis. For this pur- pose we used the result of the last section (2 hours as time interval) and we chose two time intervals (from 8:00 to 10:00 and from 14:00 to 16:00) during the day. The first time interval (from 8:00 to 10:00) is a more unstable time interval, because the sun starts to shine and rises during this time interval. The period (from 14:00 to 16:00) is a more stable time interval, because the sun stands high and there is a lot of power generation. Point clouds are calculated for four different length of time periods: five days, ten days, twenty days and thirty days as historical input data. The point clouds are shown in Figure 4.8 - 4.15

Point clouds with a time period of five days:

Figure 4.8: Time interval from 08:00 to

10:00, 5 days historical input data (RMS: 6.9)

Figure 4.9: Time interval from 14:00 to

16:00, 5 days historical input data (RMS: 7.92)

In the figures with only five days historical data input, the points are not spread over the complete spectrum of possible irradiation values and this may lead to bad pre- dictions outside this spectrum. Rather there are separated points, which give an idea of a linearity. In our opinion there are not enough points to calculate a regres- sion function with a certain accuracy.

Point clouds with a time period of ten days:

Figure 4.10: Time interval from 08:00 to

10:00, 10 days historical in- put data (RMS: 9.23)

Figure 4.11: Time interval from 14:00 to

16:00, 10 days historical in- put data (RMS: 12.11) In the figures with ten days historical data input, the point are spread over the com- plete spectrum. Furthermore, the data points are located in a sort of line and in Fig- ure 4.10 a clear linearity can be obtained. In Figure 4.11 also a linearity is present, but in the end (high irradiation values) there are some outliers.

4.1. POWER FLOW CALCULATION IN A DAY AHEAD APPLICATION 41

Point clouds with a time period of twenty days:

Figure 4.12: Timeinterval from 08:00 to

10:00, 20 days historical in- put data (RMS: 15.5)

Figure 4.13: Time interval from 14:00 to

16:00, 20 days historical in- put data (RMS: 22.6)

In the figures with twenty days historical data input, a sort of point cloud occurs. However, the point cloud does not show the same linearity as the one with ten days input. The data points of this point cloud are more spread and a small influence of the seasonal changes may already be present.

Point clouds with a time period of thirty days:

Figure 4.14: Time Interval from 08:00 to

10:00, 30 days historical in- put data (RMS: 19.52)

Figure 4.15: Time Interval from 14:00 to

16:00, 30 days historical in- put data (RMS: 29.77) In the figures with thirty days historical data input, a point cloud does not occurs. The data points of the point cloud are widely spread over the coordinate system and the influence on seasonal changes can be seen.

To summarise this section, it can be said that the point clouds, which are made with 10 days of historical data input, seem to show the best trade off between linearity and number of data points. That is why we chose to use a 10 day input as time period for all further calculations.

4.2 Calculation of flexibility according to BDEW con-

cept

In this subsection it is explained how the flexibility is calculated. The actual calcu- lation method is based on the theoretical ”traffic light” concept explained in Section 3.5. The method implements the theoretical rules of the ”traffic light” concept with some extension in order to make it work in practice. In general, for each time interval the current phase and the flexibility value have to be calculated. To achieve this, the main step is divided in three substeps. The first step is to calculate the grid states. It i.e. per time interval the phase is determined based on the power flow, which is measured at the transformer. Additionally the auxiliary variables (e.g. SoC and power at the storage inverter) have to be calculated per interval because they are needed to calculate the flexibility rules in the second substep. In the second step, the power and capacity constraints are calculated. The results depend on the phase, in combination with the auxiliary variables. With the help of the power and capacity constraints, the flexibility rules can be calculated and as a result the flexibility sched- ule for the next day can be derived by calculating two diagrams, which visualise the flexibility rules. In the third step the discharge phase is calculated. As described in Section 3.3.1, there are two approaches possible, one with and one without dis- charge phase. This means that the third step depends on the contract with the third party. This step calculates the last possible time interval when the storage has to start discharging to be able to buffer the power peak of the red phase.

The phases that the method distinguishes are:

• green phase • yellow phase • red phase

• orange phase - discharge (optional)

With these four phases, all relevant states of the grid can be distinguished. In the next three subsections, the three steps are explained in more details.

4.2. CALCULATION OF FLEXIBILITY ACCORDING TOBDEWCONCEPT 43