You are on page 1of 5

2016 IEEE Green Technologies Conference

Optimum Packet Service and Arrival Rates in


Advanced Metering Infrastructure Architecture of
Smart Grid
Alireza Ghasempour
Electrical and Computer Engineering Department
Utah State University
Logan, United States of America
Email: a.ghasempour@aggiemail.usu.edu

Abstract—Smart grid as a communication network uses ad- improving the reliability of SG by avoiding line congestion and
vanced metering infrastructure architecture to collect the mea- generation overloads, outages management/alert, upgrading
sured data by smart meters and other devices such as power meter firmware, and interfacing with other systems [3].
measurement units from different parts of the power grid and
analyzes them for billing and sends the collected data to the Advanced metering infrastructure is consists of smart me-
utility center for further analyses and being stored. In this ters, data concentrator, the utility center, and two-way commu-
paper, we propose a three-layer aggregation architecture for nication infrastructure among them. Advanced metering infras-
advanced metering infrastructure in smart grid based on the tructure components are usually located in various networks
M/M/1 queuing system. We derive an expression for packet and different realms (public and private). Electrical appliances
delay and define three objective functions (total cost, payoff, and
cost-reward per unit time) based on two performance metrics and other integrated devices/systems are connected to the
(congestion and load) in terms of service cost rate, waiting cost smart meter via a home area network, building/commercial
rate, packet delay, reward, packet service rate, and packet arrival network, or industrial area network. ZigBee or power line
rate. We obtain the optimum value of packet service and arrival communication can be used for communication among SM
rates of collectors by minimizing or maximizing the proposed and home area network elements. A number of smart meters
objective functions.
Index Terms—Smart Grid; Advanced Metering Infrastructure; communicate to a data concentrator through neighborhood
Collectors; Data Concentrators; Cost; Packet Delay; Packet area network. The data concentrator is connected to the utility
Service Rate; Packet Arrival Rate; load; Congestions; Queuing center using wide area network [4].
Theory; The utility center collects, stores, and analyzes smart me-
ters’ data for billing purposes, interfaces with the suppliers,
I. I NTRODUCTION and issue demand response alarms. The utility center has
Smart Grid (SG) is a communication network which is modules such as geographic information system, meter data
integrated with the power grid to collect and analyze data that management system (MDMS), outage management system,
are acquired from transmission lines, distribution substations, transformer load management, mobile workforce management,
and consumers [1]. Based on these data, smart grid can provide and consumer information system [4].
predictive information to its suppliers and customers on how SMs can typically transmit data in 1 to 60 minute intervals
to best manage power. National Institute of Standards and (mostly every 15 minutes [5] and [6]) and store data for
Technology presented a conceptual domain model for SG [2]. a month. Based on the results of [7], 95 million SMs will
Advanced metering infrastructure (AMI) is one of the most generate 799 petabytes data annually or equivalently each SM
important components in smart gird which creates a two-way will produce 2.397 kbps. By considering future needs (less
communication infrastructure between smart meters and the transfer intervals and sending more data), we assume that the
utility center (UC) to collect consumption data of consumers average data rate of SMs is 5 kbps. This amount of data causes
and send them to the utility center for further analysis and serious challenges for reliability, performance, and scalability
storing. Some functionalities of advanced metering infrastruc- of SG. All collected data must be sent to the UC to be
ture are near real-time power quality (such as voltage and analyzed by MDMS in a real time and if processing of the
current) management, improving energy efficiency, adaptive huge amount of data is beyond the capability of the UC, some
power pricing (to decrease costs, to enhance service delivery to data will be lost and delay will be increased. Thus, reliability
customers, and to update energy prices in real time), demand and performance of the SG will be decreased and resources
side management, self-healing ability to protect SG against of the network cannot be used efficiently.
malicious sabotage and natural disasters, providing commu- To manage this huge amount of data, an improved AMI is
nications between utility companies and smart meters (SMs) crucial. Therefore, based on a M/M/1/∞/FCFS (first come first
to remotely read usage reading, energy saving expected load, serve) queuing system, we propose a three-layer aggregation

978-1-5090-2039-3/16 $31.00 © 2016 IEEE 1


DOI 10.1109/GreenTech.2016.8
advanced metering infrastructure architecture for smart grid
(see Fig. 1). In this architecture, each collector gathers data
from its associated smart meters (the first layer to aggregate
data). A group of collectors send their data to the corre-
sponding data concentrator (the second layer of aggregation).
The data concentrators transmits data to the utility center via
backhaul networks (such as optical fiber networks or wireless
cellular networks). collectors reduce power consumption of
the SMs, because smart meters do not need to transmit
their data directly to the utility center. collectors significantly
increase reliability of the smart grid and decrease collisions
between transmitted data of smart meters. Another advantage
of collectors is enhancing scalability of smart grid by enabling
communications in large networks. In this paper, we derive a
formula for packet delay in M/M/1/∞/FCFS queue system
and define three objective functions, total cost, payoff, and
cost-reward per unit time. By minimizing or maximizing the
proposed metrics, we derive expressions for the optimum value
of packet arrival and service rates.
The rest of the paper is organized as follows: in section II, a
formula is derived for packet delay and two new performance
metrics are defined. Then, the optimum value of packet service
and arrival rates of collectors are found based on these metrics.
Section III presents performance results. The conclusion is
given in section IV.
II. O PTIMUM PACKET SERVICE AND ARRIVAL RATES
The proposed three-layer architecture for the advanced
metering infrastructure is based on this fact that the whole
coverage area of smart grid (which is controlled by the
utility center) is divided to some independent sub-areas (it
means we can optimize design parameters for each sub-area
separately and independently since each sub-area is served by
independent distribution substation). Each sub-area has one
data concentrator, NC number of collectors, and NS number
of smart meters which are deployed uniformly in that sub-area.
Aggregation of data packets from smart meters is done in three
steps or levels. First, each collector stores received packets in
its buffer (queue) and sends the packet at the head of its queue
to the associated data concentrator (it causes a delay). Then, Fig. 1. A proposed three-layer aggregation advanced metering in-
each data concentrator collects data from its corresponding frastructure architecture for smart grid based on the M/M/1/∞/FCFS
queuing system
collectors. Finally, the utility center receives all data packets
from data concentrators.
To optimize the proposed AMI architecture, we must define
and determine the following optimization components: throughput. Cost and reward can include reward to the served
1) Decision parameters and variables packets and cost incurring by waiting packets. The objective
2) Performance metrics function quantifies the implicit trade-off between cost and
3) Cost and reward reward, e.g., if service rate increases, service cost increases,
4) Objective function/s but service time and waiting cost decrease.
Decision parameters in our queuing system can be packet To find the optimum value of packet service and arrival
arrival rate, packet service rate, inter-arrival time and service rates of collectors, we define three objective functions, total
time distributions, and queue discipline. In this paper, we cost, payoff, and cost-reward per unit time, based on two
chose packet arrival and service rates as decision variables. performance metrics, congestion and load, in terms of service
Performance metrics can be expected stationary number of cost rate, waiting cost rate, packet delay, reward, packet service
packets in the system or in the queue, the expected steady rate, and packet arrival rate. Thus, we must first calculate
state waiting time in the system or in the queue, congestion, or packet delay and then define three objective functions.

2
A. Packet Delay and the queuing system will be in the steady state. By
To obtain packet delay, we used NC parallel and inde- considering μ0 = 0 and P−1 (t) = 0, we have:
pendent queues so that each queue has a M/M/1/∞/FCFS Pi (μi + λi ) Pi−1 λi−1
model [8] where the first M indicates that inter-arrival times Pi+1 = − , (3)
μi+1 μi+1
of packets has Exponential distribution, the second M shows
that the probability distribution of packet service time is P 0 λ0
P1 = . (4)
Exponential, ∞ indicates that the queue has infinite buffer, μ1
and FCFS is the queue discipline (see Fig. 2). Equations (3) and (4) can be used to obtain the following
Since packet arrival rate (the number of arrived packets formula for Pi :
in a time interval) is a Poisson random variable and packet  
λj−1
service time has Exponential distribution, we have a birth- Pi = P0 Πij=1 . (5)
death process (see Fig. 2) which is a Markovian process and μj
state changes occurs only between neighboring states, i.e., ∞
Since i=0 Pi = 1, by using (5), we have:
going forward to the next state (birth) or going back to the
 ∞   −1
previous state (death). λi (i = 0, 1, 2, ) is the packet arrival λj−1
rate per unit time (or birth rate) according to Poisson process P0 = 1 + Πij=1 . (6)
i=1
μj
and move from state i to state i + 1 and μi (i = 1, 2, 3, ) is
the service time rate per unit time (or death rate) according to In M/M/1/∞/FCFS queuing model, λi = λ and μi = μ for
Exponential distribution and move from state i to state i − 1. all i, where μ is the average packet service rate.
In the following, we derive equilibrium equation for the By considering the ratio ρ = λ/μ as a utilization factor
probability of having i packets at time t. Assuming that na (t) (ρ < 1, the mean packet arrival rate must be less than the
is a packet arrival process which denotes the total number mean packet service rate of the system), (5) and (6) can be
of packets arrived up to time t (t ≥ 0) and Pi (t) is the rewritten as follows:
probability that we have na (t) = i packets at time t (i.e.,
Pi = (1 − ρ)ρi , (7)
Pi (t) = P r{na (t) = i}). The probability that a packet arrives
between t and t+δt (where δt is an infinitesimal time interval) P0 = 1 − ρ (8)
is equal to λi δt when the queuing system is in state i and the
probability that a packet is served between t and t + δt is To find the packet delay DP (a time that a packet has to
equal to μi δt . To calculate Pi (t + δt ), we must consider all spend in the queuing system to wait in the queue and to be
probabilities that the system might get to state i at time t + δt . served), we can use Little’s formula (DP = NP /λ) where NP
Therefore, we have (for i ≥ 1): is the expected value of the number of packets in the system
 in the stationary state. We have:
Pi (t + δt ) − Pi (t) = δt − Pi (t)(μi + λi ) + Pi+1 (t)μi+1 +
  ∞

Pi−1 (t)λi−1 + δt2 Pi (t)(λi μi+1 + μi λi−1 ) − (μi + λi ) NP = iPi =
ρ
. (9)
  1−ρ
Pi+1 (t)μi+1 + Pi−1 (t)λi−1 ) − δt3 Pi (t)(μi + λi ) λi μi+1 + i=0

μi λi−1 . So, DP can be calculated as follows:
(1)
1
By dividing both side of (1) by δt and taking the limit as DP = . (10)
δt → 0, we will have: μ−λ
B. Optimum packet service rate
dPi (t)
= Pi−1 (t)λi−1 − Pi (t)(μi + λi ) + Pi+1 (t)μi+1 . (2) To optimize the transmission rate of collectors in our
dt
proposed M/M/1/∞/FCFS queuing model, we must find the
After a sufficiently long period of time, the state probabili-
optimum value of packet service rate. Thus, we fix the packet
ties are independent of time (Pi (t) = Pi and dPi (t)/dt = 0)
arrival rate λ and vary the packet service rate μ as an decision
variable. When μ increases, the throughput increases and the
ߣͲ ߣͳ ߣ݅െͳ ߣ݅ congestion between the packets in the system decreases. Thus,
to minimize the congestion, μ must go to infinity. But, in the
real systems, their hardwares have limited capabilities (μ =
0 1 i ∞) and service cost increases by increasing μ. Therefore, there
is a trade-off between increasing service cost and decreasing
the congestion between packet when μ increases.
ߤͳ ߤʹ ߤ݅ ߤ݅൅ͳ In the steady state condition (μ > λ), we consider service
cost and waiting cost to define total cost per unit time as
Fig. 2. State transition diagram of a birth-death process for M/M/1- follows:
/∞/FCFS queuing model CT (μ, λ) = αμ + βDP (μ, λ), (11)

3
where α is the rate of service cost and β is the rate of Equation (20) can be rewritten as follows:
waiting cost. The first term (αμ) in (11) denotes service cost ⎧ 
⎨λopt = 0 0 < μ ≤ βγ
and the second term shows waiting cost. We formulate the   →
optimization problem as follows: ⎩λopt = μ − β μ > β
γ γ
min CT (μ, λ) = min αμ + βDP (μ, λ). (12)  
μ μ β
λopt = max μ − ,0 . (22)
CT (μ, λ) is a convex function because we have: γ

2β Since there is no benefit when λopt = 0, we choose the


CT (μ, λ) = > 0, ∀μ > λ. (13)
(μ − λ)3 values of μ, β, and γ so that μ > βγ .
To find the minimum value of CT (μ, λ) and the optimum D. Optimum packet arrival and service rates
value of μ, we differentiate CT (μ, λ) with respect to μ and Now, we want to obtain the optimum value of packet arrival
set the derivative equal to zero. So, we have: and service rates. So, the packet arrival and service rates are
β decision variables. We combine features of cost and payoff
CT (μ, λ) = α − = 0. (14) models and define cost-reward per unit time as follows:
(μ − λ)2
Equation (14) yields the following expression for the opti- CR (μ, λ) = αμ + βλDP (μ, λ) − γλ, 0 ≤ λ < μ. (23)
mum value of packet service rate (μopt ): where the second term in (23) indicates the cost due to the

number of packets in the queuing system.
β
μopt = λ + . (15) If λ and μ are zero, CR (μ, λ) goes to infinity. Therefore,
α we exclude λ = μ = 0. Also, when α > γ, we have μα > λγ
Also, the minimum value of CT (μopt , λ) will be: (since μ > λ) and CR (μ, λ) will be positive and cannot be
zero. Thus, we assume that α < γ.
CT (μopt , λ) = αλ + 2 αβ. (16) We can formulate the optimization problem as follows:
C. Optimum packet arrival rate min CR (μ, λ) = min αμ + βλDP (μ, λ) − γλ. (24)
μ,λ μ,λ
Now, we want to obtain the optimum value of packet arrival
To find the minimum value of CR (μ, λ) and the optimum
rate that packets be accepted into the queue of our proposed
value of μ and λ, we differentiate CR (μ, λ) with respect to μ
M/M/1/∞/FCFS queuing model. Thus, the packet service rate
and λ and set the derivatives equal to zero. So, we have:
is fixed and the packet arrival rate is a decision parameter.
When λ increases, the load (number of served packets per βλ
α= , (25)
unit time) increases. But the congestion is increases. So, there (μ − λ)2
is trade-off between these two performance metrics. βμ
We define payoff per unit time based on reward of accepting γ= . (26)
(μ − λ)2
packets and waiting cost as follows:
Equations (25) and (26) yield the following expression for
P (μ, λ) = γλ − βDP (μ, λ), (17) the optimum value of packet arrival and service rates (λopt
and μopt ):
where γ is the reward of entering each packet into the queue.
The first term (γλ) in (17) denotes reward. We can formulate βγ αβ
μopt = , λopt = . (27)
the optimization problem as follows: (γ − α)2 (γ − α)2
Also, the minimum value of CR (μopt , λopt ) will be:
max P (μ, λ) = max γλ − βDP (μ, λ). (18)
λ λ
αβ
CR (μopt , λopt ) = . (28)
P (μ, λ) is strictly concave, because we have: γ−α
2β III. P ERFORMANCE RESULTS
P  (μ, λ) = − < 0, 0 ≤ λ < μ. (19)
(μ − λ)3 We implemented the proposed AMI architecture in a square
area of 100 km2 and divide this area to 100 square sub-areas.
Thus, the P (μ, λ) is maximized when Each sub-area has one concentrator (in the center), 10 number
of collectors per km2 , and 100 number of SMs per km2 . Fig.
λopt = 0 P  (μ, 0) ≤ 0
, (20) 3 shows service, waiting, and total costs per unit time versus
0 < λopt < μ and P  (μ, λopt ) = 0 P  (μ, 0) > 0 packet service rate when α = 0.1, β = 0.5, and λ = 9.8.
where As shown in Fig. 3, service cost increases and waiting cost
β decreases by increase of packet service rate. In Fig. 4, the
P  (μ, λ) = γ − . (21) effect of changing service and waiting cost rate are shown.
(μ − λ)2

4
By increasing α or β at fixed packet service rate, total cost 90
α=4, β=2
increases. If the ratio of waiting cost rate and the service cost α=2, β=2
80
rate remains constant, the optimum value of packet service rate α=1, β=2
α=1, β=1
will not change. Fig. 5 and 6 show reward, waiting cost, and 70
payoff perunit time versus packet arrival rate for two cases
(0 < μ ≤ βγ and μ > βγ ). 60

CT(μ,λ)
IV. C ONCLUSION 50

In this paper, we introduce a three-layer aggregation ad- 40


vanced metering infrastructure architecture for smart grid. We
derive an expression for packet delay and define three objective 30
functions (total cost, payoff, and cost-reward per unit time)
based on two performance metrics (congestion and load). 20
We obtain the optimum value of packet service and arrival
10
rates of collectors by minimizing or maximizing the proposed 10 12 14 16 18 20
Packet service rate (μ)
objective functions.
R EFERENCES Fig. 4. Total cost for different value of α and β when λ = 9.9

[1] A. Ghasempour, “Optimum Number of Aggregators based on Power


Consumption, Cost, and Network Lifetime in Advanced Metering Infras- 10
tructure Architecture for Smart Grid Internet of Things,” IEEE Consum.
Commun. and Netw. Conf. (CCNC), pp. 302-303, 2016.
[2] NIST Framework and Roadmap for Smart Grid Interoperability Stan- 0
dards, Release 3.0., May 2014. Reward, Waiting cost, and P(μ , λ)
[3] A. Ghasempour, “Optimized Scalable Decentralized Hybrid Advanced
Metering Infrastructure for Smart Grid,” IEEE Int. Conf. on Smart Grid -10
Commun. (SmartGridComm), pp. 1-6, 2015.
[4] A. Ghasempour and J. H. Gunther, “Finding the Optimal Number of
Aggregators in Machine-to-Machine Advanced Metering Infrastructure -20
Architecture of Smart Grid based on Cost, Delay, and Energy Consump-
tion,” IEEE Consum. Commun. and Netw. Conf., pp. 967-70, 2016.
[5] A. Mahmood, N. Javaid, and S. Razzaq, “A review of wireless commu- -30
nications for smart grid,” Renewable and Sustain. Energy Rev., vol. 41,
pp. 248-260, 2015.
[6] K. Y. Khumchoo and W. Kongprawechnon, “Cluster analysis for pri-
-40
mary feeder identification using metering data,” Int. Conf. of Inf. and Reward
Commun. Technol. for Embedded Systems (IC-ICTES), pp. 1-6, 2015. Payoff
[7] “Turning Big Data Into Power,” AutoGrid (ZPryme Research), 2013. Waiting cost
-50
[8] A. Leon-Garcia, Probability, Statistics, and Random Processes for Elec- 0 5 10 15 20
trical Engineering, 3rd ed., Pearson Prentice Hall, 2008. Packet arrival rate (λ)

Fig. 5. Reward, waiting cost, and payoff (γ = 0.1, β = 50, μ = 21)

2
4
Total cost
Service cost
3.5 Waiting cost 1
Service cost, Waiting cost, and CT(μ , λ)

Reward, Waiting cost, and P(μ , λ)

3
0
2.5

-1
2

1.5 -2

1
-3
Reward
0.5 Payoff
Waiting cost
-4
0 0 5 10 15 20
10 12 14 16 18 20 Packet arrival rate (λ)
Packet service rate (μ)

Fig. 6. Reward, waiting cost, and payoff (γ = 0.1, β = 2, μ = 20.5)


Fig. 3. Service, waiting, and total costs (α = 0.1, β = 0.5, λ = 9.8)

You might also like