Date: 04/16/09

Speaker: J. Taylor

Title: A network based approach for designing the all-electric ship

Abstract:
Robustness as a function of network structure and dynamics is a
frequently studied feature of complex networks. The perspective has
been successfully applied in modeling cascading failures in
terrestrial power systems. To date, most work has focused on analysis
and prediction, with little attention given to design. This is largely
because terrestrial power grids are built incrementally over large
time scales, allowing little room for system level design.

It has also been observed that the degree distribution of electrical
power networks is well described by a power law, due to the nature in
which additions are incorporated into existing grids. The integrated
power systems of the navy's all-electric ship is an electrical system
of scale comparable to that of terrestrial power grids; however,
unlike terrestrial power systems, a shipboard power system can be
fully designed prior to construction. It is of strong interest for the
all-electric ship to be robust to broad range of failure scenarios.
Specifically, it is desirable to minimize damage propagation and
prevent cascading failures typically seen in terrestrial power grids.
Cascading failures in terrestrial power grids are modeled using a DC
load flow approximation, and then nodes are failed according to the
capacity of their links to failed neighbor nodes. Because lines are
orders of magnitude shorter in the electric ship, it is acceptable to
use network flows to model the dynamics, and then to propagate
failures as in terrestrial grids.

The current paradigm is simulation-based design, that is to design the
ship via conventional approaches, and then test a large number of
designs through simulation. Detailed simulation is itself expensive,
and, although more efficient than actual construction, severely limits
exploration of the design space. Rather than testing highly realized
designs through expensive simulation, we propose an alternate approach
in which much simpler designs with high statistical robustness are
used as a foundation for detailed design, thus allowing robustness to
be established at an earlier stage in the design process. Towards this
end, we apply random network theory and optimization.