IEEE Power & Energy Magazine - November/December 2017 - 44

One efficient way to mitigate these outliers is to provide an early situational awareness for the incoming events
(e.g., large wind ramp-down events and volatile weather). In
this way, operators can anticipate, detect, and react to such
events. Figure 4 shows an example of how such situations
may be dealt with by uncertainty-based reserve predictions.

For that purpose, ERCOT implemented a reliability risk
desk in January 2017. One of its main tasks is to assess the
risk of extreme errors and the associated impact on grid performance. If such large forecast errors could be detrimental to the grid, operators can purchase more reserves, bring
more generation units online, or cancel a scheduled outage
to mitigate the adverse effect.

Better Knowledge by Considering the
Grid Topology and Ensemble Forecasts

(a)

In energy systems with a high share of wind and solar power,
it is crucial that RES production can be curtailed or redispatched by different stakeholders of the energy system. In
systems with established energy markets, as in many European countries, curtailments can be initiated either 1) by
traders (selling RES energy to the market) to maximize their
income, or 2) by distribution system operators (DSOs) or
TSOs to ensure grid stability and security of supply.
In some power systems, only a few very large wind farms
and solar plants are directly connected to the transmission grid
and, hence, visible to the TSO. Many RES units are located in
DSO grids connected to medium voltage levels. Large amounts
of roof-top solar plant located in the low-voltage-level grid
behind the meter can cause issues for TSOs as well as DSOs.
In other power systems, either a larger proportion of RES is
connected to the transmission grid, or the TSO has some degree
of observability regarding the production of DSO-connected
units, enabling it to generate and validate RES forecasts.
A power forecast of the aggregated production of each
DSO area feeding into the TSO grid is a good way to start
and has been established in several European countries. In

3-h Ahead Load
Forecast Error (MW)

6,000
4,000
2,000
0
-6,000 -4,000-2,000
0
2,000 4,000 6,000
-2,000
-4,000
-6,000
-8,000
3-h Ahead Wind Forecast Error (MW)

figure 3. A 3-h ahead load-forecast error versus a 3-h
ahead wind-forecast error in 2016 at ERCOT.
44

ieee power & energy magazine

1,500
1,200
900
600
300
0
-300
-600
-900
-1,200
-1,500
19 Oct.
12:00
16:00
20:00
20 Oct.
04:00
08:00
12:00
16:00
20:00
21 Oct.
04:00
08:00
12:00
16:00
20:00
22 Oct.
04:00
08:00

figure 2. The geographic graphical user interfaces for
(a) the EWeLiNE project and (b) Hawaii's SWIFT project,
showing wind and solar power forecasts over a region of
single transformer stations, as well as forecasts of relevant
meteorological parameters and threshold-exceeding probability values critical for grid security.

(MW)

(b)

Min-p10
p50-p60

p10-p20
p60-p70

p20-p30
p70-p80

p30-p40
p80-p90

p40-p50
p90-Max

figure 4. An example of a dynamic reserve prediction
with percentile bands p10-p90 (gray shading). The black
lines at ±300 MW indicate a static reserve and show the
spill that such static reserve allocation contains. The black
dotted line is the allocated reserve and illustrates the issue
of outliers that, if required to be captured, lead to overallocation of reserves. Nevertheless, the p10/p90 band
indicates the risk of outliers.
november/december 2017

Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - November/December 2017