IEEE Power & Energy Magazine - November/December 2019 - 50

to obtain reliable wind and PV forecast scenarios with realistic weather-driven correlations among locations. Figure 4 is
an example of a vertical grid load measurement and probabilistic and deterministic forecasts of the wind power share at
this transformer.
Using a calibrated scenario forecast instead of a single
best forecast for all points of common coupling (PCCs)

500

Power (MW)

400
300
200
100
0

-100

7
r.
2

7
29

r.
2

7
26

r.
2

7
Ap
23

r.
2

7
01
r.
2

Ap
17

r.
2

-200

Time
Forecast

Grid Load

Wind Power

figure 4. An example of a load flow measurement at a
transformer between extra-high and high voltage levels.
The blue and green curves show, respectively, deterministic
and scenario-based forecasts of the wind-power share at
this transformer.

120
100
Q at GCP (MVar)

80
60
40
20
0
-20
-40
-60

14
16
Time (h)

Q GCP = 0
Q GCP Normal
Q-Range OPF
Q-Range Forecast
Uncertainty Q-Range

figure 5. The results of a simulation for a use case concerning a reactive power (Q) demand on the part of the
TSO, with Q = 0 Mvar at the PCC. GCP: grid conjunction
point; OPF: optimal power flow. (Source: S. Wende-von
Berg et al.; used with permission.)
50

ieee power & energy magazine

implies that the load flow calculation must be repeated
for each scenario member. Besides the multiple load flow
calculations, grid operators must interpret the characteristics of the resultant load flow for each grid state scenario.
These scenarios can be used to optimize redispatch and
VER curtailment actions based on risk according to the
probability of predefined critical grid states. Scenario forecasts can also be used to calculate system services that a
DSO could provide to a TSO at a single PCC. Figure 5
plots results from simulations using historical data from
a German DSO, for a use case concerning reactive power
demand by the TSO. The system required that no reactive power exchange take place at this PCC. In Figure 5,
the red triangles show the original expected interchange
of reactive power flow at the PCC and the green triangles
the flow after optimization of the operation of the installed
VER (none needed). The green regions represent the possible reactive power flexibility of the VER at each time step.
The pink and yellow areas display a simulated forecast of
reactive power at the beginning of the time series (6:00 a.m.)
and its uncertainty, respectively. In this case, the optimization did a good job of satisfying the criteria of meeting zero
reactive power exchange at the PCC.

Using Probabilistic Information to Fine-Tune
Unit Commitment: A Lesson From the U.S.
Southwest Power Pool
The U.S. Southwest Power Pool (SPP) has incorporated
more than 17 GW of wind resources into its generation fleet
over the past decade. As of early 2019, SPP had a total wind
capability of 21.5 GW. To date, SPP has seen upwards of
16.4 GW of instantaneous wind generation and served an
instantaneous 63.4% of system load with wind energy. These
high levels of variable energy production make accurate and
up-to-date resource forecasting critical for system reliability.
As industry experts and vendors work to improve forecasts
for end-user consumption, SPP is also taking steps to manage real-time, day-ahead, and multiday forecast errors to
better understand the potential impacts of forecast uncertainty. Thus, probabilistic forecasting is likely to provide
value for these time frames.
SPP has developed a process to maintain situational awareness of the impacts of potential forecast errors. Each day,
the staff runs real-time studies that assume various forecast
errors. These multiday reliability studies help the organization ensure that, in the event that forecast errors occur, it will
still have sufficient energy capacity available to serve load.
Currently, SPP runs four daily studies with different levels
of error and resource flexibility: two with an 85th-percentile
forecast errors for both load and variable resource applied
and another two with 99th-percentile errors. An advance
notice interval of either 6 or 20 h differentiates the two studies at each error percentile. This advance notice interval designates how far in advance of a particular target interval a
resource can be called on to be available for that interval.
november/december 2019

IEEE Power & Energy Magazine - November/December 2019

Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - November/December 2019