Environmental Engineering Reference
In-Depth Information
expensive. The investigated scenarios are therefore based on this experience and the next
step in benchmarking and adopting a forecasting system no longer to a low mean absolute
error (MAE), but rather to the conditions under which the wind power integration takes
place and optimise the forecasting system with this knowledge.
7.2.1.
Optimisation Scenarios
1. Scenario: Static Regulation
In this scenario the upper limit for reserve allocation from the error statistic of one
forecast over 1 year has been determined. The actual reserve allocation is limited by
the forecast with the following restriction:
F C
+
R < UB
(2)
and
F C
−
R > LB
(3)
with
F C
−
R < G < F C
+
R
(4)
Here, F C is the forecast, R is the reserve, G is the actual generation, UB is the upper
bound of full generation and LB is the lower bound of no generation. R is chosen
to secure that the generation is always lower than the sum of the forecast and reserve
and higher than the forecast without reserve, which is supposed to historically always
be valid (here: 90%).
2. Scenario: Deterministic Forecast Regulation
In this scenario the reserve R is chosen as a fixed reserve allocation for upward and
downward ramping. We chose +/-11% of installed capacity, which is equivalent to
the mean absolute error (MAE).
3. Scenario: Security Regulation
In this scenario, the reserve R is computed from the difference between minimum
and maximum of the ensemble in each hour of the forecast. There may however be
areas, where it will be necessary to adopt the difference of minimum and maximum
in such a way that single outliers are not increasing the spread unnecessarily.
4. Scenario: Economic Regulation
In this scenario, an optimisation of scenario 3 is used, where the unused reserve
allocation is reduced for economic reasons. As a first approximation the 70% quantile
of the level in Scenario 3 was used. In a future optimisation, the skewness of the price
of positive and negative reserve also would have to be considered.
The data that were used to simulate different scenarios was 1-year (2006) of forecasting
data for two regions in Alberta, the South-West region and the South-Centre region, where 5
wind farms with a combined installed capacity of 251.4 MW were used. The simulation was
carried out with 12-18 hour forecasts issued every 6th hour. The forecasting model system
was run in 22.5 km horizontal resolution and the wind power forecast was a probabilistic
Search WWH ::
Custom Search