• No results found

Forecasting power flow with forecast irradiation

ation from the weather prediction as input. Furthermore the difference in power flow forecast accuracy between our method with the forecast irradiation and measured irradiation as input are shown. For this purpose we simulate the same days as in the previous section, but now with the irradiation values of the weather prediction instead of the measured irradiation as input. Note, that the weather prediction is for the same weather station (Ahaus) as the measured weather values.

As described in the literature research (Appendix A), it is in general easier to fore- cast stable weather periods than unstable weather periods. However, in the area of Wettringen, there is quite often an unstable weather situation with a lot of fluctuation. This means it is more difficult to have precise weather predictions since it is among others difficult to forecast the movement of the clouds. Due to these considerations the irradiation input of our method will include a certain inaccuracy already from the beginning on.

In the following two figures (5.3 and 5.4), three lines are shown. The blue line is the real measured power flow at the transformer. The green line is the forecasted power flow with measured irradiation as input to our forecast method. The red line is the forecasted power flow with forecasted irradiation from the weather prediction as input to our forecast method.

Figure 5.3: Forecast results of 08-05-2016.

Figure 5.3 shows the forecast from 08-05-2016 is shown. It was a warm and sunny day without many clouds. Further the shape of the power forecast with forecast irra- diation differs only slightly from the real measured power flow and the forecast power flow with measured irradiation. The power flow forecast with measured irradiation has an average deviation of 0.82 kW whereas the power flow forecast with forecast irradiation has an average deviation of 5.20 kW. So if the weather period is good only a very small inaccuracy is introduced due to the forecast irradiation.

On the other hand, in Figure 5.4, the forecast of 08-07-2016 is given, which was a day in a more unstable weather period. This can also be observed from the accu- racy of the two forecasts. The power flow forecast with measured irradiation has an average deviation of -7.58 kW whereas the power flow forecast with forecast irradia- tion has an average deviation of -20.74 kW. Thus, in this case most of the inaccuracy comes from the inaccuracy of the predicted irradiation.

The above considered factors must be kept in mind when considering the flexibili- ties. The presence of prediction errors implies that there must be a certain safety threshold if this method is used in practical cases.

5.6. FORECASTING POWER FLOW WITH FORECAST IRRADIATION 57

Figure 5.4:Forecast results of 08-07-2016.

5.6.1 Daily pattern

In this section we investigate the slope of the regression function in more detail. For this purpose we calculate the regression function for each hour over the period of a month. Note, that the slope of the regression function indicates the efficiency of the conversion of irradiation in power for the PV installations of Wettringen.

The slope for the month April is shown in Figure 5.5. The exact numbers and the visualisations of April until August can be found in Appendix B.

The figure shows the changes of the slope within the day and through the days of the month. However, for a better analysis, more details are necessary. For this we plot in Figure 5.6 the slope of the regression function for a single day (20-04-2016). Note, that there are time intervals where power is consumed (power is flows from the medium voltage to the low voltage grid) and time intervals when power is produced (power flows from the the low voltage to the medium voltage grid). In the first set of time intervals (from 03:00 until 06:00 and from 17:00 until 19:00) the assumption that the grid is heavily dominated by PV generation is not valid, because there is almost no PV generation.

This means the power flow values are mainly determined by the demand in these intervals, only very small irradiation values occur and that for these irradiation values similar power flow values occur for these irradiation values. For small positive values of irradiation the slope of the regression function is steeper than normal.

Figure 5.5: Slope through April 2016

This behaviour produces the two big peaks (one positive at 05:00 and one nega- tive at 20:00). But times with almost no irradiation are less interesting, because the electricity storage will not be used for peak shaving purposes in these time intervals. The general shapes of the slope for the days are quite similar. A daily pattern how- ever, changes slowly over time. It can be seen that the PV installations are most efficient between 10:00 and 14:00, because in this time period the slope is high and the slope is an indication of the efficiency. Furthermore the pattern shows a quite stable slope, which fluctuates around 0.5.

Because the PV panels are all installed on the roof tops of the farm buildings, they face in many different directions and have different angles to the sun. So there should not be a large peak at midday when the sun is highest, but rather we may expect some smaller peaks during the day. There are two bigger peaks (next to the ones with almost no PV generation) over the day. The first occurs around 09:00 and the second around 13:00 (see Figure 5.6). We assume that these peaks may come from the larger PV installations in the distribution grid of Wettringen. Two of the PV installations in Wettringen (126.9 kW and 154.8 kW) have their own smart meter built in, which sends the measurement data of the generated power to the control centre. These two PV installations are from the same farmer, and therefore we as- sume they stand close to each other. However, the PV installations are installed on two different rooftops. With the data from these PV installations we are able to com- pare the power generation of the two PV installations with the slope values of the regression function, which indicates the efficiency of the PV installations. We expect that the power generation shows power peaks at the same times where the slope is

5.6. FORECASTING POWER FLOW WITH FORECAST IRRADIATION 59 highest. To investigate this, we again considered the data from 20-04-2016. Figure 5.7 shows the power generation profile of two PV installations in the distribution grid of Wettringen. This profile shows the power generation from 20-04-2016.The peak at 09:00 is exact at the same time in both figures (5.6 and 5.7). The slope peak at 13:00 is not exact matching with the power peak, but the power generation at 13:00 is already quite high. We assume that one of the other PV installations that has a peak at 13:00 and in sum with the other PV installation the peak is hence shifted to 13:00.

From the above it can be concluded that not all PV installations face the same direc- tion. However, because also the production of both PV installations in Figure 5.7 has two peaks, we assume that also the individual PV panels on both rooftops face two different directions. Based on the shape of the production curve, we may assume that a small number of panels face east; at 09:00 the sun shined from east (more precise from100.66◦, retrieved from [5]). The majority of panels face southwest, be-

cause at 13:00 the sun shined from southwest (more precise from191.01◦, retrieved

from [5]).

Figure 5.7: Power generation 20-04-2016

5.7 Regression analysis of the slope changes over

Related documents