Environmental Engineering Reference
In-Depth Information
forecast generated with 2 combinations of different power conversion methods each using
300 weather parameters from the 75 member MSEPS ensemble system.
In Table 5 statistics of the aggregated forecasts for the 5 wind farms are displayed.
Table 5. Statistics of wind power forecasts without reserve prediction.
Statistics Parameter
[% rated capacity]
Bias (FC)
-2.10
Mean absolute Error (FC)
11.60
RMSE (FC)
16.80
Correlation (FC)
0.85
In Table 6 all parameters are given in % of installed capacity. The relative cost of
various types of reserve was not taken from market reports, but estimated as a percentage
relative to urgent reserve on the spot market, which was set to 1.0. This gives the following
estimates:
•
Urgent reserve = 1.0%
•
Unused passive reserve = 0.1%
•
Allocated reserve = 0.6%
The “allocated reserve” is the reserve that is allocated according to the forecasts and
assumed to be bought day-ahead. The statistical parameter in the first rows of Table 6
are based on the forecasts including a reserve allocation. This means that the estimated
error is added or subtracted from the forecast according to the used optimisation scheme.
Therefore, these parameters have an index FCR (forecast + reserve), while the statistical
parameters in Table 5 are based on the raw forecasts and indexed FC. These FC forecasts
were optimised on the mean absolute error (MAE) to the observations.
There are furthermore output results of three types of reserve in the table, the required
reserve, the predicted reserve and the unpredicted reserve, respectively. These are named
UpReg for up-regulation or DownReg for down-regulation of the power on the electricity
grid because of incorrect forecasts. Although the difference in price for up-regulation and
down-regulation, where down-regulation is often a factor of 3-5 cheaper than up-regulation,
was not accounted for in these simulations, the cost function is more favourable towards
down-regulation than up-regulation. This can be seen in scenario 2 and 4, which are more
cost efficient and use less up-regulating expensive reserve.
Table 6 also shows the significant difference in effective costs for the first scenario,
the purely static reserve allocation, in comparison to the other 3 scenarios. However, when
comparing the hours covered by the reserve, then it becomes clear that the coverage and the
effective cost are cross-correlated for this type of optimisation. Scenario 2 is for example
equally cost efficient than scenario 5, but covers only 64.5% of the hours, while scenario 4
covers 76% of all hours. Although the security scenario (no. 3) is slightly less cost efficient,
the covered hours are quite significantly higher than for scenario 2 and also scenario 4
(85% versus 76% and 64,5%). Looking at the mean absolute error (MAE) or the root mean
Search WWH ::
Custom Search