The North American interconnected electric power grid is one of the
largest and most reliable electricity delivery networks in the world.
Since electricity cannot be stored in significant quantities, the power
grid is a nearly instantaneous delivery network. To achieve the quality
of service that we currently enjoy, and to protect this critical infrastructure,
we rely on a sophisticated real-time energy management system (EMS).
The focus of this presentation will be on security assessment tools
that run in an EMS for large-scale power systems. In particular,
a parallel implementation of an online voltage security assessment tool
will be discussed.
The steady-state modeling of large-scale power systems will be presented,
along with continuation techniques for exploring steady-state behavior.
At the core of the computational engine is a nonlinear solver based on
Newton's Method. For current large-scale power system models, the
number of equations is roughly 70,000 and growing. Since the system
must be solved at a series of parameter values during the continuation
trace, we are exploring a parallel implementation of Newton's method on
a distributed-memory message-passing architecture.
The key to good performance is in partitioning the power system network
among available processors. Observations will be presented for improving
the performance of sparse direct linear solvers on a GNU/Linux cluster
with a fast ethernet (100 Mbit/s) or gigabit ethernet interconnect. |