Professional Documents
Culture Documents
IMECE2013
November 15-21, 2013, San Diego, California, USA
IMECE2013-62970
METHODOLOGY
Lewis et al. combine both Type I and Type II robust design where !! is the adjacency matrix, !! represents the actual
principles and apply them to complex systems, in an effort to !"#$%!
length between all pairs of nodes, and !! includes the cost
address uncertainty from both internal and external
environment [25]. The goal of Lewis formulation was to meet for the unit length between all pairs of nodes. Intuitively, the
performance requirements, while minimizing the variation most robust solution would have the highest connectivity as
about the mean. Figure 2 outlines this relationship, displaying well as the highest cost.
how the optimized solution may exist at the boundary of an
objective, where variability is greatest [9, 25]. The objective Power Grid Structure
value of the robust solution is slightly higher, although with To accurately represent a power grid, both generation
less performance variation. Power grid designs can benefit nodes and demand nodes are assigned to represent power
from applying this method, as uncertainties from both sources generation and demand sources respectively, and are user-
are present, including external noise factors (e.g., natural defined values. Generation values are based on power plant
disasters) and internal noise variables (e.g., expected demand). energy production, and demand values are the total power
demand required to reliably service a given area. It is assumed
Connectivity relationships are based on previous research Where and !"#$ are the adjacency matrix and the number
by Kirk, where Dijkstras algorithm is used to determine the of disconnected components of the network, respectively.
shortest path distances between generation and demand nodes
which are used to calculate the loads on the connection lines CASE STUDY AND RESULTS
[27]. The average demand that is satisfied after a typical failure
has happened is referred to as Expected Demand (! ). Failure IEEE 14 Test Bus System
events are administered randomly by the removal of a single The IEEE 14 Test Bus system (Fig. 3) was used to validate
arc, and resultant demand being satisfied after cascading failure the methodology presented [26]. This network consists of two
is calculated as (! ). Arc removal is performed 10 times within power generation stations, and 12 additional demand
a given power grid network design, and Expected Demand is connections. Since cascading failure is being evaluated in
calculated based on the average of resultant demand for each terms of transmission line loading, the physical topology of
failure scenario, given total network connectivity (Eq. 6). lines was considered (Fig. 4). Based on IEEE 14 transmission
line lengths calculated by the Power Systems Engineering
!!!!" Research Center, a geographical power grid map was
! = (6)
!" constructed [27]. This is an important system attribute as line
lengths directly drive connectivity costs. In order to accurately
Arcs can also be removed according to the distribution of represent system demand, nominal demand node power
failure events anticipated based upon historic data using a requirement values were used from the IEEE 14 system.
Monte Carlo simulation approach.
Optimization Algorithm
Cascading Failure A genetic algorithm (GA) was used within the MatLab
In this model, a failure event is initiated by the random Optimization Toolbox to solve objectives [28]. Since Expected
removal of an arc, forcing the original load on that arc to be Demand Variability is part of the objective function, values
redistributed to the next available shortest path in the system. were normalized so the GA could evaluate solutions on the
This is an iterative process where !"#$ () is the initial arc same scale. Values were calculated for Cost, Expected
load at a given time t, and its value is based on the demand Demand, and Expected Demand Variability from the original
node values associated with it. For an arc failure to occur, line IEEE 14 transmission line configuration. These nominal values
load must exceed line capacity. Ten arcs were selected were included in the fitness function for each objective (Eq. 7).
randomly for removal, although future work will include actual
transmission line failure rates. After a failure, the arc is
removed from the network and load is recalculated for all
remaining arcs. If load exceeds capacity in another arc, this
one is also removed from the network and the redistribution
process is repeated. The cascading failure will eventually end
!"#!
= + !"## (8)
!"""!"!
13' 94.7"
14'
54.4"
11'
34.44"
54.1"
12' 73.6"
69.6"
52.3"
6'
5'
2.86"
11.8" 4'
15.2" 10'
12.0" 11.6"
1(Gen)' 9'
23.0"
2(Gen)' 13.46"
3'
4.02"
30.0"
8' 48.0"
7'
Optimization Results design, since the number of node connections was the same.
The simulation was run for approximately 600 iterations. Expected Demand is greater than one, signifying a greater
From these results a plot of Pareto optimal solutions is ability of the robust solution to meet demand requirements after
displayed in Figure 5. In this plot, design values are a cascading scenario. Expected Demand Variability is less than
normalized with respect to the performance of the original 0.1, displaying a significantly lower variance. These results are
IEEE 14 network configuration. Tradeoffs between each expected as the IEEE 14 network was physically constructed
objective solution are explored within this design space, and an based on both population demand and geography, and was
optimal value calculated. In the Robust Design solution (Fig. unlikely optimized for performance. The physical topology of
5), the normalized Cost is slightly less than one, determining the robust network, compared to the original IEEE 14 is shown
the robust solution was less expensive than the original IEEE in Figure 6 and 7.
14 network. The cost was reduced, primarily due to the
efficiency of transmission line topology used in the robust
Network Topology
In terms of network topology, the robust network (Fig. 6)
consists of 19 transmission lines, versus 17 in the optimal
solution (Fig. 8). The network is fully connected, with no
disconnected demand nodes or sub-networks. The generation
Figure 7. IEEE 14 ORIGIONAL CONNECTION
nodes (node 1 and 2) of the robust solution each have 3 and 5
TOPOLOGY (19 LINES)
degrees of connectivity respectively, versus 3 and 4 degrees
ACKNOWLEDGMENTS
The authors would like to acknowledge the support of the
University of Alabama in Huntsville Center for System Studies
and NASA Marshall under grant SUB2012-052
REFERENCES
Figure 8. OPTIMAL NETWORK CONNECTION [1] Fairley, P., 2004, "The Unruly Power Grid," IEEE
TOPOLOGY (17 LINES) Spectrum, pp. 23-27.
[2] U.S.-Canada Power System Outage Task Force, 2003,
"Interim Report: August 14th Blackout in the United States and
CONCLUSIONS AND FUTURE WORK Canada," U.S. Secretary of Energy Minister of Natural
This paper presents a methodology for high-level power Resources Canada.
grid design by incorporating robustness into network topology. [3] Hines, P., Apt, J., and SaroshTalukdar, "Large
Simulating this system mathematically provides the opportunity blackouts in North America: Historical trends and policy
to iteratively test various network connection strategies against implications," Energy Policy, 37, pp. 5249-5259.
an initiating failure event, and address subsequent cascading [4] Faza, A. Z., Sedigh, S., and McMillin, B. M.,
failure effects. Failure resistance in systems can be improved "Reliability Analysis for the Advanced Electric Power Grid:
by examining tradeoffs between robustness and performance in From Cyber Control and Communication to Physical
early design. In this research robustness is represented as the Manifestations of Failure," Proc. International Conference on
variability in performance metrics as a result of external effects Computer Safety, Reliability, and Security, pp. 257 - 269.
on the system, such as transmission line failure. [5] North, M., Conzelmann, G., Koritarov, V., Charles
The IEEE 14 Test Bus case study demonstrates the Macal, Thimmapuram, P., and Veselka, T., 2002, "E-
application of the method presented, and compares objective Laboratories: Agent-Based Modeling of Electricity Markets,"
values between the original network, the optimized network, American Power Conference, Argonne National Laboratory,
and the robust network. This proof-of-concept simulation Chicago, IL.
highlights the importance of topology configurations when [6] Carreras, B. A., Lynch, V. E., Dobson, I., and
designing to mitigate cascading failures. It also validates the Newman, D. E., "Dynamics, Criticality and Self-organization in
use of a mathematical model as a tool in early power grid a Model for Blackouts in Power Transmission Systems," Proc.
design. International Conference on System Sciences, IEEE.
A primary benefit of the method presented is its ability to [7] Pottonen, L., and Oyj, F., "A Method for Analysing
be scaled to more complex power grid systems. Larger the Effect of Substation Failures on Power System Reliability,"
networks can be evaluated, based on user inputs for desired Proc. 15th Power Systems Computation Conference, PSCC.
system topology, demand requirements, and availability of [8] Lininger, A., McMillin, B., Crow, M., and
generation sources. This is limited however, by the Chowdhury, B., 2007, "Use of Max-Flow on FACTS devices,"
computational requirements of the evolutionary algorithm. One 39th North American Power Symposium, pp. 288-294.
direction for future work is the division of computational [9] Chen, W., 2012, "Design Under Uncertainty," Design
workload using multi-core processing capabilities to create Under Uncertainty, Northwestern University, Evanston.
large and distributed network simulations. While this may be [10] Hines, P., Apt, J., and Talukdar, S., 2009, "Large
considered a brute force method, the availability of remote blackouts in North America: Historical trends and policy
computing facilities makes it a viable strategy. implications," Energy Policy, 37(12), pp. 5249-5259.
As an alternative strategy to increase simulation efficiency, [11] Electrical Power Research Institute, 2000, "Rx for
social network analysis could be applied. In this method, Stress: Power Delivery Reliability Initiative," Electrical Power
smaller demand populations would be optimized independently Research Institute, Palo Alto.
of each other, and connected using a second optimization [12] Long Island Power Authority, 2009, "Project K014:
objective which optimizes the topology of the total system. Probabilistic Risk Assessment (PRA) User's Group," Long
This method would allow the examination of performance Island Power Authority.