![](bg4.png)
Since the SESA-Lab is a modular environment,
used by many stakeholders and practitioners of dif-
ferent backgrounds, non-intrusive UQ methods are
deemed the most appropriate. Intrusive methods
would require too much manual effort of the user, and
are not applicable at all when closed-source software
is used.
Additionally to being non-intrusive, UQ meth-
ods must suffice two requirements for the SESA-Lab:
they must be computationally efficient in order to be
applicable in large-scale scenarios, and they must be
applicable for sources of aleatory as well as epistemic
uncertainty. Combining these two requirements un-
derlines the challenging character of UQ in Smart
Grids. Computationally efficient algorithms typically
assume knowledge about the distribution of uncertain
values and are therefore only applicable for aleatory
uncertainty, see e.g. (Lin et al., 2014). Distribution-
ignorant methods, on the other hand, are typically
sampling-based, like Monte Carlo Simulation (MCS),
which leads to high computational costs in complex
systems.
5 STATE OF THE ART
Uncertainty quantification in general is a broad and
active research field. Since it is a collection of widely
applicable methods, it is of interest for every scien-
tific discipline that is associated with measurement or
modeling. Consequently, readily usable tools have
been developed to facilitate the application of UQ
methods. One of the most noteworthy of these tools
is the open-source software DAKOTA (Adams et al.,
2014) that has been developed by Sandia National
Laboratories. It provides not only functionalities for
UQ but also for optimization, parameter estimation
and sensitivity analysis. The UQ capabilities of the
software include different ways to assess initial un-
certainty as well as the most established propagation
methods. However, it is not considered suitable to
manage mosaik’s UQ via DAKOTA by coupling the
platforms. The most important argument against the
coupling is the fact that the analyzed simulation code
has to be started by DAKOTA. This would limit the
independence and thereby the modular character of
mosaik. Furthermore, DAKOTA is not a domain-
specific tool. In the context of Smart Grid research,
it provides a large overhead of unnecessary function-
alities while those functions are not included that have
been specifically developed for the energy domain,
e.g. probabilistic load flow (Borkowska, 1974). Nev-
ertheless, DAKOTA has to be considered as an im-
portant reference and a possible resource. After all,
the open-source character of the software promises to
be helpful for individual implementations of selected
methods.
Although UQ is a field with a long tradition, its
application to power system modeling and especially
Smart Grid modeling has only started to gain atten-
tion in the recent years. It has often been suggested
that large-scale power systems are too complex for
classical sampling-based UQ methods like MCS. In-
stead, new methods are developed and improved, e.g.
the approach by Lin and colleagues (Lin et al., 2014).
They specifically test their collocation method with a
power grid model and demonstrate its computational
efficiency in comparison to MCS.
Hiskens and Alseddiqu present a UQ approach
specifically focused on dynamic, continuous power
system simulations, similar to the ones conducted
by eMEGAsim in the SESA-Lab context (Hiskens
and Alseddiqu, 2006). They point out the compu-
tational efficiency of their trajectory sensitivity ap-
proach, stressing the importance of this feature for
systems as complex as power grids.
The Smart Grid concept increases the complexity
of power systems even more, especially in the con-
text of uncertainty, as suggested by Zio and Aven (Zio
and Aven, 2011). They argue that the large amount
of determining factors yields different forms of un-
certainty due to different states of knowledge. This
is problematic since uncertainty propagation methods
oftentimes rely on knowledge about the uncertainty
sources. Furthermore, they deem it important to con-
sider as much uncertainty sources as possible, but it
can be difficult even for experts to assess input uncer-
tainties for some sources, e.g: what is the probabil-
ity of a fault in a newly developed grid component?
Zio and Aven suggest a general framework for uncer-
tainty assessment in Smart Grids, divided into three
abstract categories, namely “drivers” (observable tar-
gets, e.g. costs), “limiters” (constraints, e.g. limita-
tions in technical deployment), and “effectors” (influ-
encing phenomena, e.g. failures). However, the prac-
tical use of such a framework has not yet been tested.
In the context of mosaik, it is also unclear whether the
framework can be applied to each type of model that
is capable of being integrated through the API.
Li and Zio suggest a more practical approach
for joint assessment of uncertainties from different
sources (Li and Zio, 2012). They combine concepts of
probability and possibility theory in order to account
for different states of knowledge. However, they use
this approach as a first step of MCS that is oftentimes
assumed to be unfit for complex, large-scale systems,
as stated above. It is questionable whether the joint
assessment approach is compatible with more sophis-
UncertaintyQuantificationinSmartGridCo-simulationAcrossHeterogeneousModelDomains
19