Environmental Engineering Reference
In-Depth Information
The average of the ensemble has a score of 5.83%. This is approximately the same
score as the best single ensemble member. The
optimal forecast
gives the same period
5.35% absolute error. In the above example we found that if we add the uncertainty band
and trade the predicted amount of reserve for both up-regulation and down-regulation on
the market, we can reduce the error to 2.72%. This means that we have to trade regulation
for only 2.72% of the capacity instead of 5.35% in the short-term spot market with high
prices.
In this computation, it is assumed that there is no error when the error of the
optimal
forecast
lies within the predicted uncertainty band. This increases the correlation signifi-
cantly, which indicates that a large fraction of the forecast errors within the uncertainty band
are completely unpredictable. We suggest that these errors are handled most efficiently by
being balanced with pre-allocated constant reserve.
To be able to estimate the impact of such a pre-allocation, we considered 5 cases, with
different percentages of pre-allocation and computed the amount of hours, where the pre-
allocation accounts for the forecast error. With these results, it becomes feasible to set up a
cost analysis of the optimal amount of pre-allocation. Our intention here is however merely
to demonstrate how to design an optimised prediction system for a specific problem.
In the following we have therefore constructed 4 cases, where a fixed fraction of the
installed capacity is pre-allocated as reserve with a long term contract and additional ca-
pacity is assumed to be allocated in a market with competition according to the predicted
requirement (see scenario 1 above).
1. Case: Additional long-term contract for reserve of 0.8% of the inst. capacity
2. Case: Additional long-term contract for reserve of 1.6% of the inst. capacity
3. Case: Additional long-term contract for reserve of 3.2% of the inst. capacity
4. Case: Additional long-term contract for reserve of 6.4% of the inst. capacity
5. Case: Additional long-term contract for reserve of 12.8% of the inst. capacity
The following equation has been used to compute the number of hours when the pre-
allocated reserve accounts for the forecast error:
R
=
MAX
(
R
pre
, A
stability
X
+
B
stability
)
(1)
where R
pre
is the pre-allocated reserve from Table 4 and X is the ensemble spread.
The sum of the hours, where the resulting pre-allocated reserve fully covers the forecast
error is shown in Table 4. It can be seen in the table, how well the ensemble spread covers
the forecast error for different levels of pre-allocation. The validation also reveals that only
3 hours out of 2331 hours (97 days) had forecast errors that were not covered by the reserve
given by equation 1 when using a 12.8% pre-allocation.
A pre-allocation of nearly 13% to reach 96% coverage by using this type of reserve is
quite a large amount compared to the mean absolute error (MAE) of the forecast (ยก 6%).
It is therefore recommended to study and evaluate from case to case and with real market
data, if more than 10% of pre-allocation is a reasonable result of the optimisation. However,
Search WWH ::
Custom Search