Optimization
Methodologies for Large-Scale Power Systems back

Motivation:

The objective of this project is to push
the state-of-the-art in optimization
algorithms and software to enable the use
of high-performance computing in power
grid planning and operations. This will
enable decision-making under uncertainty
and the incorporation of high-fidelity
power system models.

Research:

Sponsored by the Department of Energy
(DOE) Office of Electricity, we reached
and demonstrated a high level of
technical maturity of our stochastic
optimization solver, PIPS.
We have used PIPS to solve challenging
problems arising in power systems. The
solver is now capable of solving a stochastic
unit commitment problem over the entire
Illinois transmission system with a
prediction horizon of 24 hours, 32,000
scenarios, a full physical DC
network model, and up to billion
variables in less than one hour on a
variety of high-performance computing
platforms such as BG/P, XE6, XK7 and XC30.
This is a drastic reduction in
computational time from previous versions.
This enables us, for the first time, to
consider operational solutions with full
physical network models (we do not need to
make assumptions on uncongested lines as
is done in practice) and to capture spatio-temporal
wind power uncertainty in
deployable times. The tests took
place in the BlueGene/P system and in Oak
Ridges' Titan system. The solution of
these problems required significant effort
in extending PIPS to integrate the
cutting-edge linear algebra solver PARDISO
with Schur assembling capabilities
capable of exploiting hypersparsity. In a
related activity, we also implemented and
tested network partitioning strategies
in interior point solvers (see Figure
1) and have been capable of reducing
computing times by a factor of 10 in the
Illinois system using Argonne's Fusion
cluster. This is an important step in
addressing ISO-sized transmission systems.

We are also currently developing new
models to capture reliability constraints
and to capture emerging phenomena from
the interconnection of natural gas and
power grid systems. Finally,
we are exploring the use of the the advanced
scripting languageSWIFT
to perform computationally intensive
simulations using heterogeneous computing
systems (see Figure 2).

C. G. Petra, O. Schenk, M.
Anitescu. Real-time Stochastic
Optimization of Complex Energy
Systems on High Performance
Computers. Submitted, 2013
[pdf]

M. Lubin, J. A. J. Hall, C. G.
Petra, and M. Anitescu. Parallel
distributed-memory simplex for
large-scale stochastic LP
problems. Computational
Optimization and Applications, 2013
[pdf]

Lubin,
M.; Petra, C.
G.; Anitescu, M., and Zavala,
V.M. Scalable
Stochastic Optimization of Complex
Energy Systems.
Supercomputing, 2011. [pdf]

Cioaca,
A.; Zavala, V.M. and
Constantinescu, E.M. Adjoint
Sensitivity Analysis for Numerical
Weather Prediction: Applications
to Power Grid Optimization.International
Workshop on High Performance
Computing, Networking and
Analytics for the Power Grid,
2011. [pdf]

Robbins,
B.; and Zavala, V.M. Convergence
Analysis of a Parallel Newton Scheme
for Dynamic Power Grid Simulations.International
Workshop on HPC, Networking
and Analytics for the Power
Grid, 2011. [pdf]

Zavala, V. M.; Botterud, A.;
Constantinescu, E. M. and Wang,
J. Computational and Economic
Limitations of Dispatch
Operations in the Next-Generation
Power Grid.
IEEE Conference on Innovative
Technologies for and Efficient and
Reliable Power Supply,
2010. [pdf]

Figure 1. Partitioning of Illinois Network using
Metis

Figure 2. Time Evolution of Locational
Marginal Prices Generated with Swift