IEEE Power & Energy Magazine - November/December 2019 - 47

By Sue Ellen Haupt, Mayte Garcia Casado,
Michael Davidson, Jan Dobschinski, Pengwei Du,
Matthias Lange, Timothy Miller, Corinna Möhrlen,
Amber Motley, Rui Pestana, and John Zack
weather dependent, while others are associated with different uncertainty factors. Therefore, the
entire system has uncertainties that are most effectively managed with a probabilistic approach.
Because both the atmosphere and the electrical system with significant VERs contain large
uncertainties, a probabilistic approach to forecasting and planning becomes necessary to optimize the weather-dependent systems. Using probabilistic information provides knowledge that
can be used to enhance situational awareness, assess multiple potential grid scenarios, optimize
economic outcomes, leverage smart statistics, and improve plans. Probabilistic forecasts should
not be considered the only means to solve the problem of
predictability of VER output but, rather, as a way forward
to better deal with the many specific tasks in power system operations. These forecasts provide the possibility of
making decisions based on a best guess and preparing for
the less likely (lower probability) scenarios or extremes
that may cause security issues on the grid, such as excessive costs for balancing services or equipment malfunctions or failures. We will describe some of the leading
ways that probabilistic forecasts are being used and show
how they might additionally be leveraged to enhance
power system operation and planning.

©ISTOCKPHOTO.COM/IG0RZH

Building Probabilistic Forecasts

november/december 2019

Four primary methodologies are currently used to generate probabilistic forecasts for the power industry:
✔ type 1: statistical methods of probabilistic forecasts
✔ type 2: statistically based scenarios
✔ type 3: physically based ensemble forecasts
✔ type 4: perturbation-based ensemble forecasts.
These methodologies differ in their output and usage
possibilities, complexity, and computational resource
requirements. Figure 1 summarizes these approaches and
their typical output types.
The first two methodologies statistically generate a
probability density function (pdf) from a deterministic
NWP simulation and the last two by employing different NWP models or model parameterization schemes or
perturbing the initial or boundary conditions. The disadvantage of running multiple ensemble members is the
high computational cost in comparison to statistically
based methodologies.
However, the user of such forecasts should be aware
that each approach conveys a different type of information
about the forecast uncertainty. The statistical postprocessing of a single NWP forecast represents the composite
historical uncertainty (i.e., based on the statistical sample),
whereas an NWP ensemble more explicitly characterizes
ieee power & energy magazine

47
IEEE Power & Energy Magazine - November/December 2019

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