IEEE Power & Energy Magazine - November/December 2019 - 49

How End Users Use
Probabilistic Forecasts
Nearly all grid system operators with high shares of VERs
use weather-dependent generation forecasts. Typically, they
use only a best, single (i.e., deterministic) forecast. The use
of probabilistic forecasts by grid operators with heavy renewable penetrations is growing, leading to enhanced grid management decision making. Such forecasting is also providing
market uncertainty products to enable automatic commitment
of resources to assist with renewable uncertainty. The following sections provide examples of the current use of forecast
uncertainty information in grid management processes.

Using Probabilistic Information as Input for
Grid Security Calculations: A German Example
Several German transmission system operators (TSOs) and
distribution system operators (DSOs) have recognized the
november/december 2019

importance of including VER predictions in load flow calculations to detect future grid congestion and identify countermeasures. In the German power system, most renewable
generators are connected at the 110-kV or lower voltage
level. This leads to an interest in the power flow through
substations that connect different grid levels, referred to
as vertical grid load. The exchange of data between grid
operators, especially between TSOs and DSOs, is traditionally tedious. Several recent initiatives to exchange data
regarding grid topology, VER units connected to specific
grid points, and actual output promise to optimize gridrelated forecasts. Vendors in Germany now provide forecasts of the vertical grid load to grid operators for a number
of substations. Recent experience indicates that integrating
generation, load, and price forecasts leads to more accurate projections of the vertical grid load. In addition, confidence bands show the uncertainty of the vertical grid load
forecast, as illustrated in Figure 3.
German grid operators currently calculate congestion
forecasts for a specific voltage level based on deterministic generation and load predictions. The resultant load flow
forecasts and grid security measures do not incorporate
any uncertainty information about volatile wind and photovoltaic (PV) generation. German TSOs plan to use probabilistic forecasts, i.e., scenario forecasts, as an input for grid
congestion security calculations (e.g., congestion forecasts)
to consider uncertainty in grid management processes.
Several methods are used for calculating forecast scenarios
of the power production of spatially distributed wind plants
and PV stations based on probabilistic forecasts. Using methods such as ensemble copula-coupling approaches (method 2
in Figure 1), one can calibrate the scenario in time and space

Power (MW)

the former contain more information. There are three fundamental attributes that describe the performance of probabilistic forecasts: 1) reliability, 2) sharpness, and 3) resolution.
The reliability of a probabilistic forecast is the degree
to which the frequency of the actual outcome matches the
forecasted probability. For example, do events with a 60%
probability occur in 60% of the cases? If the reliability is
not high, the forecast presents a distorted view of the uncertainty and requires calibration.
The sharpness attribute measures the spread of a pdf.
The sharpest probability forecast is one in which one value
of a forecast variable has a probability of 100%, and all other
values have a probability of 0%, which is essentially a deterministic forecast. A sharper forecast (i.e., closer to a deterministic forecast) is desirable but only if reliability is high. If
a set of forecasted pdfs typically displays less spread than the
climatological frequencies, then the forecast is often considered to have meaningful sharpness.
The resolution attribute refers to the ability to distinguish
among cases with different pdfs. A forecast method that produces the same probability distribution for all cases (e.g.,
every forecast cycle) shows zero resolution. An example of a
zero-resolution forecast is the climatological frequency distribution, which does not distinguish among forecast cases.
However, such a forecast may be highly reliable (as in the
case of a climatological distribution). A probabilistic forecast
with high resolution enables the user to more effectively distinguish cases with low forecast uncertainty from those with
high uncertainty, which may be the most valuable aspect of
probabilistic forecasts for many applications.
There are metrics to measure each of these attributes,
with reliability measured most often. Although a high degree
of reliability is necessary for a probabilistic forecast to have
value, it is seldom sufficient. In practice, the sharpness and
resolution distinguish among alternative probabilistic forecast solutions. Forecast providers often use composite scores
that are implicitly sensitive to all three attributes.

5
0
-5

-10
21 Mar. 2019 22 Mar. 2019

23 Mar. 2019

Vertical Grid Load Forecast
Vertical Grid Load Forecast Spread

figure 3. A forecast of vertical grid load (solid black line)
for one substation between TSO and DSO grids with upper
and lower bands indicating uncertainty. Negative values
refer to transport to the lower-voltage grid; positive values
represent transport to the higher-voltage grid.
ieee power & energy magazine

IEEE Power & Energy Magazine - November/December 2019

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