Escolar Documentos
Profissional Documentos
Cultura Documentos
I. INTRODUCTION
Smart Grid
Agent Quantity
Generation
Millions
Centralized power plants
(high inertia)
Monitoring and protection
control (e.g., distribution
automation)
Substations, homes and
businesses, generation
facilities
Complex phasor
measurements units (PMUs)
in the transmission layer.
Electric power (energy)
usage as a function of time of
day at smart meters
Sub-cycle (PMUs) to 15
minutes (smart meters)
kbps
Hundreds
Distributed renewables
& storage (low inertia)
Continuous control and
monitoring
Data Function
Agent types
Parameters
Sampling
Intervals
Data Volume
(per Agent)
Stability
Source
grid
Generation, load,
storage facilities
throughout the grid
Complex voltage and
current throughout the
grid at all times
T
G0
L0
micro-grid 0
S0
G1
L1
S1
micro-grid 1
but recognize there are other definitions that may also apply
(e.g., latency, BER, etc.). We therefore propose to assign
access to the communications bandwidth proportionally to
agents power generation/load.
We illustrate this approach by considering a grid with total
capacity of 1 MW that is to be distributed over N =100 nodes
(note: these chosen numbers can be readily scaled). We
consider three scenarios differentiated by the ratio between the
largest (n = 1) and smallest (n = N = 100) agents in terms
of power load/generation. In Scenario A, this ratio, R, is 10;
in Scenario B, 100; in C, 1000. We assume that these loads
are geometrically distributed (Fig. 2) with density given as
f n [ n] = (1 p)
n1
10
(1)
where
p = 1 R
(2)
10
# 1 &
%
(
$ N 1 '
10
70
Scenario A (10)
Scenario B (100)
Scenario C (1000)
60
Scenario A (10)
Scenario B (100)
Scenario C (1000)
10
load (kW)
50
10
20
30
40
load (kW)
50
60
70
40
30
20
10
10
20
30
40
50
60
node number
70
80
90
100
Also assumed for this work is that the data sampling rate
required for the largest agent is 1000 Hz and for the
smallest is 1 Hz and linearly proportional by load in
between. From Fig. 3, we see that the propensity of agents
(<10 kW) will require far less bandwidth than required by the
largest agents. Or conversely, the majority of the bandwidth
will be utilized by only a small percentage of the agents (much
as a scale free network).
In aggregate, based on our
proportioning of the bandwidth, the aggregate data rate is 1.48
Mbps; that is, significantly less than the 100 Mbps required to
provide every agent the highest data rate.
III. AUTOMATON-BASED DESIGN
While allocating the bandwidth to each agent according to
Fig. 3 (e.g, via TDMA) is attractive to ensure that the channel
is utilized efficiently, we also recognize that the network itself
will be dynamic in terms of agent behavior. Not only are
individual agent loads stochastic in nature, but existing nodes
may be removed from the grid and/or new agents be added.
A. MAC Control
In the previous wireless sensor network work, the measure
for QoS was how many nodes sent data during an epoch. For
our grid work, we are interested in ensuring that the channel
adequately accommodates the traffic submitted by the agent.
As the agents in the grid change in their behavior and
number, the automaton design may either over- or under-load
the channel with data therefore some system level control is
required. To illustrate the controllability of a network of N =
100 grid agents, we compare control of our K = 2 design
with a K = 1 design (i.e., an automaton effectively
implementing 1-p CSMA MAC).
In Fig. 5, number of data transmissions per discrete time
(epoch) is plotted vs. the desired control. The K=2 automaton
almost completely satisfies the needs of the control system for
this example. As previously demonstrated [18], this K-state
automaton approach has better bandwidth utilization in
comparison to other random access approaches (e.g., 1-p
CSMA, ALOHA, etc.). It should be noted that the number of
states, K, and their probabilities, Pi, can be refined to improve
the systems response and reduce the variability from epoch to
epoch (e.g., optimal values for Pi for a various K and N was
detailed in [17]). For the grid application space, this control
overhead (reward, punishment) is minimal given the ability to
readily adapt to the number of agents and their variable loads.
Automaton reward/punishment could be determined locally
through a carrier sense method or globally through a broadcast
provided by a channel-monitoring agent.
B. Agent Transmission Performance
Recall that our objective in this work is to ensure that those
agents who have the greatest impact on the grid receive
access to the communications bandwidth. Our results for
Scenario C (where the ratio of power load between the
largest and smallest agents is 1000) are illustrated in Fig. 6.
K=2 automaton
transmissions
60
K=1 automaton
50
40
30
20
10
0
50
100
150
200
250
epoch
300
350
400
450
500
QoS = 60
70
transmission rate(%)
Control
70
QoS = 30
60
50
40
30
QoS = 15
20
10
0
10
20
30
40
power load (kW)
50
60
Fig. 6. Agent activity rate for grid operating under Scenario C. Percentage
of successful transmissions normalized by number of epochs. Largest loads
are seen to participate at the highest rates. Automaton based design enables
virtual priority.
V. CONCLUSION
In this work, we have adapted previous automaton-based
approaches developed for wireless sensor network control to
the problem of MAC for grid communications. Each grid
entity employs a multi-state automaton, the state of which
dictates the load presented to the channel. Each entity (or
agent) is allocated load limits based on their over all impact on
the power activity in the grid. Results show that not only
does the automaton enable the overall network activity to be
well controlled with little overhead but also that agents are
able to communicate with high channel throughput in a
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
VIII. BIOGRAPHIES
Jeff
Frolik
(S85,
M95,
SM11)
received
the
B.S.E.E.
degree
from
the
University
of
South
Alabama
in
1986,
the
M.S.E.E.
degree
from
the
University
of
Southern
California
in
1988
and
the
Ph.D.
degree
in
Electrical
Engineering
Systems
from
the
University
of
Michigan
in
1995.
He
is
currently
an
Associate
Professor
in
the
School
of
Engineering
at
the
University
of
Vermont.
Anthony
Lentine
is
a
Principal
Member
of
the
Technical
Staff
at
Sandia
National
Laboratory,
Albuquerque,
NM.
He
received
his
B.S.E.E.
and
M.S.E.E.
from
the
University
of
Illinois
(Urbana-Champaign)
in
1979
and
1980,
respectively,
and
his
Ph.D.
degree
in
Physics
from
Heriot-Watt
University,
Edinburgh
UK,
in
1993.
Andrew
Seier
(S'11)
is
a
Graduate
Student
at
the
University
of
Vermont
where
he
also
completed
a
B.S.E.E.
in
2011.
He
is
currently
working
toward
a
M.S.
in
Electrical
Engineering
with
a
focus
on
communication
requirements
for
future
power
grids.
Chris
L.
Palombini
(S'06)
recently
completed
his
Ph.D.
from
the
University
of
Vermont.
He
has
worked
at
Sandia
National
Laboratories
on
a
Department
of
Energy
Electrical
Power
Fellowship.
Currently,
he
is
a
Signatures
Engineer
at
Applied
Physical
Sciences,
Lexington,
MA.