Content uploaded by Inam Ullah Khan
Author content
All content in this area was uploaded by Inam Ullah Khan on Sep 18, 2018
Content may be subject to copyright.
Heuristic Algorithm Based Dynamic Scheduling
Model of Home Appliances in Smart Grid
Inam Ullah Khan
1,2
, Xiandong Ma
1
, C. James Taylor
1
, Nadeem Javaid
3
, Kelum A.A. Gamage
4
1
Engineering Department, Lancaster University, Lancaster LA1 4YW, UK
2
Electrical Engineering Department, COMSATS University Lahore, Pakistan
3
COMSATS University, Islamabad 44000, Pakistan
4
School of Engineering, University of Glasgow, Glasgow G12 8QQ, UK
i.u.khan@lancaster.ac.uk; inamkhan@ciitlahore.edu.pk
Abstract—
Smart grid provides an opportunity for
customers as well as for utility companies to reduce
electricity costs and regulate generation capacity. The
success of scheduling algorithms mainly depends upon
accurate information exchange between main grids
and smart meters. On the other hand, customers are
required to schedule loads, respond to energy demand
signals, participate in energy bidding and actively
monitor energy prices generated by the utility
company. Strengthening communication
infrastructure between the utility company and
consumers can serve the purpose of consumer
satisfaction. We propose a heuristic demand side
management model for scheduling smart home
appliances in an automated manner, to maximise the
satisfaction of the consumers associated with it.
Simulation results confirm that the proposed hybrid
approach has the ability to reduce peak-to-average
ratio of the total energy demand and reduce the total
cost of the energy without compromising user comfort.
Keywords: Demand side management, appliance scheduling,
critical peak pricing, household energy management.
I. INTRODUCTION
Demand side management (DSM) usually refers to those
decisions, which are taken by utility companies at user’s
premises [1]. DSM programs are initiated to use available
energy in a more efficient way without developing new
infrastructure for generation, transmission and
distribution. DSM programs usually encompass demand
response programs, fuel substitution programs, efficient
conservation of energy programs and above all
commercial or residential load management programs [2]-
[4]. Reducing and shifting consumption is one of the main
key design features of the residential load management
program [5]. This can only be achieved if users are
encouraged to build energy efficient buildings and to be
well aware of their energy consumption patterns. A part
from this practical initiative needs to be taken, including
high power appliances shifting from peak hours to off-
peak hours for measurable reduction in peak-to-average
ratio (PAR) in load demand. Load shifting is expected to
be even more important because of high penetration of the
plug-in hybrid electrical vehicles (PHEVs). Usually
PHEVs require 0.2-0.3 kWh charging power for one mile
driving [6]. This significantly enhances new load on the
existing distribution system. Particularly during charging
hours, it doubles average household demand, thus
worsening the existing high PAR. In absence of properly
reinforced system, a high PHEVs penetration can create
unbalanced condition, thus compromising power quality
standards, voltage regulation issues and even prospective
damage to utility and consumer equipment.
Direct load control (DLC) is another useful approach for
residential load management [7]-[9]. By applying DLC
programs, utility company remotely controls energy
consumption and operations of certain household
appliances. For instance, thermal comfort equipment
including heating, ventilating and air conditioning
(HVAC), refrigerators, pumps and light control are well-
known examples of DLC programs. When considering
home automation and residential load control specifically,
users’ comfort is on the top priority and considered as a
hurdle in DLC programs execution [10].
Today dynamic pricing replaces DLC programs features.
In dynamic pricing mechanism, users are motivated to
manage their loads individually on a voluntary basis, e.g.,
shutting and shifting heavy loads from peak hours to off-
peak hours [11]-[13]. Most popular and frequently used
schemes of dynamic pricing includes critical-peak pricing
(CPP), real-time pricing (RTP), inclined block rate (IBR),
time of use pricing (ToUP) and day ahead pricing (DAP)
schemes, etc. With the help of these schemes users are
encouraged to shift appliances from peak hour to off-peak
hours. This helps to achieve a lower PAR and reduces
consumer costs [14].
II. RELATED WORK
Researchers have recently developed and implemented
different state of the art algorithms in smart grid (SG).
These algorithms successfully analyzed commercial,
residential and industrial buildings in terms of their load
consumption profile. Researchers have focused on
optimizing energy controller and scheduler in such a way
that energy cost is brought to an optimum level for utility
companies and customers. Maximum attention is given to
balancing the supply-demand ratio and reducing customer
cost to a minimum level. Multiple factors are considered
while developing these algorithms: appliance rating,
Proceedings of the 24th International Conference on
Automation & Computing, Newcastle University,
Newcastle upon Tyne, UK, 6-7 September 2018
pricing schemes, utility company priorities and consumer
demand in order to get maximum benefit for all
stakeholders.
In reference [9], dynamic price RTP is used for optimally
scheduling smart home appliances. They mainly focus on
reducing unconventional electricity usage, minimizing
cost and maximizing benefits from energy storage.
Electricity cost is reduced to 22.6% and peak price is
reduced to 11.7% as compared to the normal pricing
scheme. However, authors do not pay attention towards
optimizing scheme in their work. While architecting the
DSM model, authors purchase the energy during off-peak
hours and store it in storage bank for peak hours use. Their
main focus is cost minimization and energy storage in
battery bank. Linear programming based optimal
scheduling model is proposed; however, goals remain
unachieved.
Non-deterministic polynomial-time (NP) hardness based
optimal scheduling model is discussed in [12]. The authors
use greedy iterative algorithms to meet the home
scheduling goal. In their work, optimization is achieved
by using linear programming and artificial intelligence
optimization approaches. Lower peak load and lower peak
fluctuation phenomenon is also discussed. Problem
formulation is made not only based on user’s load demand
but also on the generation cost.
In references [11]-[15], authors propose a mixed integer
linear programming based algorithm for scheduling home
appliances in a fascinate way. The real price tariff is used
for scheduling home appliances to reduce cost as well as
peak reduction. In [16], multiple types of users in the
proposed model are evaluated. These users are categorized
as commercial, industrial and residential users. From
simulation results, it can be concluded that the proposed
algorithm contributes significantly to minimization of
PAR and electricity cost.
Genetic Algorithm (GA) based cost minimization
method is used in references [12]-[14]. In these papers,
renewable energy sources (RESs) and battery storage are
integrated into the existing system. RESs are supposed to
charge battery bank for later use, when electricity prices
are high during high demand of energy. For battery
efficiency and life, a controller is developed to monitor the
charging and discharging thresholds associated with the
battery bank. Furthermore, when electricity prices are low,
batteries are supposed to fully charge. Later, when prices
are high then certain high priority appliances are handled
from battery source to save user cost.
In this paper, we propose meta-heuristic optimization
models based on genetic algorithm, grey wolf optimization
(GWO) and a hybrid grey wolf and genetic algorithm
(hybrid G
2
) for scheduling 12 home appliances. Each day
is divided into 96 time slots (each 15 minutes) instead of
one-hour time slot for appliance operation. This is
necessary because in many cases an appliance requires less
than an hour to complete its operation such as the electric
cattle and dish washer. In this way, users have much
freedom and opportunities to reduce cost, PAR and total
energy demand. Finally, simulation results of the
unscheduled, GA schedule, GWO and hybrid
are
presented to show the effectiveness of the proposed
hybrid
model for appliance scheduling in DSM.
III. PROPOSED ARCHITECTURE
In this work, a smart home with multiple smart
appliances is considered. Length of operation time (LoTs)
and power rating (PR) information of all appliances are
already taken from end consumers. The whole system is
divided into three sub-layers, including supply side
management layer (SSML), communication management
layer (CML) and demand side management layer (DSML).
SSML contains all information related to energy
generation. DSML uses energy management controller
(EMC) and appliance scheduler (AS), and schedules
Base Li ne Loads
Regular Loads
Controlla ble Loads
Energy Ma nageme nt
Controller
Demand
Response
Manager
Energy Pr edictor
Load Predi ctor
Utility Control Cent re
Smart
Meter
Fig. 1 The proposed system architecture
smart appliances on the basis of LOTs defined by the end
users. Purpose of load balancer (LB) is used to delay
appliance operation to minimize demand-supply gap and
not to allow consumer demand to exceed the limit.
Through CML, energy forecaster (EF) and demand
response manger (DRM) exchange real-time demand-
supply information with SSML and DSML. Home Area
Network (HAN) conducts effective communication
between EMC through Wi-Fi, Z-wave and Zig-Bee
communication protocols. Furthermore, smart appliances
are further categorized into base line loads, regular loads
and controllable loads depending upon whether their
operation can be interrupted or not when activated. EMC
uses appliance interface (AI) which controls on/off
operation tasks of all smart appliances associated with the
system. It is pertinent to mention here that EMC through
AS, stops all scheduling operation of the appliances if the
interrupt is generated by the consumer to enhance the
comfort.
Three meta-heuristic techniques are adopted in this
paper, including GA, GWO and hybrid
in order to
schedule these smart appliances in home energy
management system (HEMS). Scheduling is performed to
save electricity utilization cost for end users. Knapsack
problem is formulated to establish coordination among
smart appliances at run-time. This gives autonomy to each
consumer for managing appliance operation according to
the comfort.
IV. APPLIANCE CATEGORIZATION
Home appliances are classified into three sub-categories
based on their operational behavior. Interruptible
appliances are those whose operation can be interrupted or
delayed during operation but their operational time is
unchangeable. Similarly, uninterruptible appliances are
those whose operation cannot be delayed or interrupted
once they are in operating mode. However, these
appliances can be shifted to other time slots before their
operations start.
It is important to shift interruptible and uninterruptible
appliances to other time slots to maintain overall energy
consumption up to an allowed level. It is beneficial to use
the interruptible appliances at low peak hours for saving
electricity cost. On contrary, base appliances are those
which can neither be interrupted nor deferred in home
energy management system (HEMS). For example,
refrigerator, air conditioning, lightening and microwave
oven are such devices whose operation pattern remains
unchanged. All appliances along with their length of
operation time (LOT), power rating and category used in
this study are listed in Table 1.
In this section, we will analytically describe power
system, energy cost and load control model for residential
purpose. Based on these descriptions, we will formulate
three design optimization problems in next section.
A. Power System
We consider a smart power system with various load
consumers and a single energy source which may be a
generator or connection to the main grid through a step
down transformer. Furthermore, we assume that each
consumer is equipped with EMC that is capable of
scheduling different appliances (12 in our model) during
different intervals of time (96 intervals in a complete day,
i.e., 15 minutes each). By making use of appropriate
communication protocol, different smart meters are
interconnected, not only with the grid but also with each
other by sharing the updated information.
Let be the set of users, where for each user let
denotes the total load at time slot {1,…….T}, where
T = 96. Daily consumed load by a specific user is
denoted by
[
]. This leads us to
calculate the total load of all users in a single time slot
across the whole day . It is represented as
(1)
Similarly, daily peak load and average load can be
calculated as
(2)
and
(3)
From equations (2) and (3), PAR can be calculated as
below
PAR=
=
(4)
B. Energy Cost Model
For each time slot energy cost for electricity
generation or distribution is represented by
.
Generally, for the same load, cost may differ in different
time slot of a day. It mostly depends upon electrical price
maintained by the utility at generation site. It is worth
mentioning here that cost function being considered in this
paper can represent either the original cost of thermal
generators or artificial cost tariffs maintained by the utility
for proper execution of the load control programs. Actual
energy cost function can be represented in terms of a
quadratic function in equation (5).
(5)
where
at each time slot t
C. Residential Load Control
For an individual user u U, let
denotes the different
set of appliances, including base, interruptible and
uninterruptible appliances in a home. For scheduling
purposes, we initially define a schedule vector for each
appliance a
of individual user, where n is the number
of the appliances.
(6)
where
represents scheduled energy consumption in
one time-slot for appliance by user. We can then
calculate the total load by user.
, t T (7)
In our proposed model, main task of AS is to determine
an optimum time slot in user’s smart meter for the
individual appliance. In this way, user can shape its
daily load profile by making use of equation (7). It is
important to mention here that energy scheduler does not
aim to reduce power consumption of different appliances
rather it shifts to other different time slots for minimization
of PAR and energy cost. Initially, a user needs to initiate
start and end time slot in which a particular appliance is
supposed to complete its task. Let beginning time slot be
represented by
and end time slot is represented
by
and
.
Table 1. Appliance parameters
Appliances Lot
(slots)
Power
rating
(kWh)
Category
Washing machine 20 1.0 Uninterruptible
Clothes dryer
16
1
.
6
Uninterruptible
Electric vehicle 36 2.0 Interruptible
Water pump
32
2
.0
Interruptible
Humidifier
12
0.5
Interruptible
Vacuum cleaner
24
1.5
Interruptible
Water heater
48
2
.0
Interruptible
Dish washer
16
1.2
Interruptible
Refrigerator
96
1.
4
Base
Air conditioner
40
1.5
Base
Light
52
0
.
8
Base
Microwave oven
16
2
.0
Base
For example, an electrical vehicle (EV) having
needs 4 hours to complete its charging cycle for 50
km driving range in a single day. For compiling task, a
user must select a larger time slot because in case of any
interruption, scheduler completes the task by its end time.
For example, the user may select
and
. Mathematically, it is
represented as
(8)
where
represents energy consumption vector of
appliance a during t time slot by u. Also, from equation
(8), it is concluded that appliance a schedules balances
according to daily consumption requirement. Similarly,
total energy consumption by all appliances and by all users
can be summed up.
(9)
Since electronic devices are divided into base, interruptible
and uninterruptible smart appliances, so in case of
uninterruptible appliances strict energy consumption
needs to be adopted. In our case, washing machine (WM)
and clothes dryer (CD) have constraints that once WM task
ends, CD must start its operation immediately. In that case,
for WM and
= 0 for CD. Similarly, a
refrigerator is on all the time, so in that case
for
WM and
= 96. Generally, a scheduler has no active
impact on the operation of the non-interruptible
appliances. For a complete energy consumption profile,
standby power of interruptible appliances needs to be
calculated. It is the power which is consumed by
interruptible appliances when they are in idle mode. we
need to calculate minimum
) and maximum
(
) standby power level for interruptible
appliances. Standby power can be assumed to be such
power which a device is consuming when it is in non
operation mode but ready to start its operation. We can
assume it as:
(10)
We are now ready to calculate different optimal energy
scheduling model by considering equations (1)-(10) in our
proposed hybrid
DSM model.
V. OPTIMIZATION METHODS
Traditional optimization methods like integer linear
programming (ILP), mixed integer programming (MILP)
and mixed integer nonlinear programming (MINLP) are
unable to control large number of appliances.
Furthermore, these methods are computationally
inefficient and hence not suitable for real time
optimization, which is deterministic in nature. Instead, the
meta-heuristic optimization technique can provide a best
solution while considering user defined constraints. We
are applying genetic algorithm (GA), grey wolf
optimization (GWO) method and a hybrid of both
techniques to achieve real time optimal results.
GA is inspired from the genes of living organisms.
Initially, binary coded chromosomes are randomly
initialized. Total number of smart appliances are
represented by the length of chromosomes’ and smart
appliances ON/OFF status is identified through
chromosomes binary coded pattern. Once the initial
population is generated, fitness function of GA is
evaluated which is actually an objective function of this
study. Mutation and crossover are performed to generate
new population. Generated population fitness function is
then compared with the previous one and hence, optimum
results are achieved.
On the other hand, GWO algorithm is based on grey
wolves hunting and leadership hierarchy mechanism.
Alpha, beta, delta and omega are four kinds of wolves in
leadership hierarchy. For performing optimization
hunting, searching, encircling and attacking, prey steps are
implemented. In this way, position of the search agents are
updated in the form of position vector towards prey.
Search agents update its position until it reaches to an
optimal position in n-dimensional search space.
The purpose of proposing the hybrid technique is to
achieve a balance between global search and local search.
GA performs well in terms of exploration mode. Also, it
has good convergence rate to reach to optimal solution.
Initially, GA steps are followed for generating initial
population of chromosomes. These chromosomes actually
represent candidate solution to the problem. Furthermore,
a bit of chromosomes represents the ON/OFF state of the
smart appliances. Fitness function is based on objective
function, taken from GWO. The best population is
regenerated through velocity updating step of GWO.
Firstly, it finds a local best solution and on the basis of this
value, it achieves a global best solution. Through an
optimal stopping rule, the cost minimization problem can
be formulated and the best fit value is thus chosen. Based
on crossover and mutation, a new stream of is generated.
Hence new generation population is created which has
completely different characteristics as compared to the
initial generation.
VI. SIMULATION RESULTS
In this section, we present simulation results and assess
the performances of the proposed algorithms. By making
RTP signal for DSM, PAR reduction, cost minimization
and load balancing are key features to be analyzed. The
cost, load and waiting time for each group is represented
in terms of cents, hours and kWh. Fig. 2 shows the load on
grid for a single home using all three approaches according
to RTP. In RTP tariffs, electricity price changes during
different times of a single day. Particularly prices are
higher in the afternoon, hot summer days and cold winter
days. Fig. 3 clearly demonstrates that during the high price
rate hours, if demand is high, then unscheduled load
creates high peaks as compared to the scheduled load. Due
to this reason, electricity cost of unscheduled load is high.
It also depicts that without affecting the overall load, the
proposed fitness function has a greatest effectiveness on
cost and PAR reduction.
Moreover, load profile during multiple time slots for a
complete day is shown in Fig. 4. It demonstrates that the
proposed hybrid model outperforms the GA and GWO
models in terms of load shifting to off-peak hours; hence
reduction can be many folds in terms of PAR and cost. Fig.
4 illustrates the cost in different time slots during the day,
the consumption pattern by GWO and GA during the peak
price is high as compared to the hybrid
approach. This
affects the overall cost per day for aforementioned
approaches as shown in Fig. 5. It clearly shows that price
using hybrid
is low as compared to GA and GWO.
Using the hybrid
, the proposed approach reduces 20%
cost, which is the best among all three used approaches
Fig. 2 Load profiles
Fig. 3 Energy cost during the time slots
Fig. 4 Cost in different time slots over the day
PAR results are shown in Fig. 6, where the
unscheduled load is very high and for the hybrid
it is
commendable. This shows adeptness of the proposed
approach which is better than GA and GWO. In this case,
about 50% PAR is reduced by hybrid
. While
addressing the cost and PAR, waiting time of different
appliances cannot be overlooked; this is highlighted in Fig.
7. Waiting time has a direct relationship and impact on user
comfort and it is an important parameter for efficiency
measurement in any proposed scheme. It shows that
waiting time for base load appliances for GA and GWO is
higher as compared to the hybrid
.
Fig. 5 Total cost under different approaches
Fig. 6 PAR under different schemes
Fig. 7 Waiting time for the different approaches
During the simulation, we perceived that GA is best for
maximum number of populations. With the increase in
number of population and generation step, difference
among the lowest and highest point becomes negligible.
On the other hand, GWO shows high performance for
small population under hundred intervals. Fig. 5 and Fig.
6 show that GA outperforms GWO in terms of cost
reduction, peak reduction and PAR. The hybrid
shows
positive influence on both approaches by lowering PAR,
cost and peak load values.
VII. CONCLUSIONS
In this paper, we have presented an effective approach
for load management by shifting or balancing home
appliances in an optimum way. The main idea is to
facilitate consumers to reduce electricity cost. From the
simulated results, it is observed that considerable saving in
energy costs can be realized by consumers. To facilitate
consumers, artificial intelligence based optimization
technique is adopted. The results show that through a
carefully designed appliance scheduling model, users can
offer a viable solution to optimal power management
among residential energy users. The proposed approach is
based on a hybrid GA and GWO.
It clearly demonstrates that the hybrid approach
outperforms the GA and GWO. The load is balanced in
such a way that not only load peaks are avoided but also
user comfort is less compromised. It is worth mentioning
that there exists the tradeoff between cost and PAR. Since
cost is minimized at certain time extents the load to off-
peak hours using the proposed model, resulting in
maximizing PAR. Results show the effectiveness of the
proposed hybrid model in terms of cost minimization.
Future work will consider integration and testing RES
along with real-time pricing signal.
ACKNOWLEDGEMENTS
The authors would like to acknowledge the funding
support from Lancaster University, UK and COMSATS
University, Pakistan.
REFRENCES
[1] Wen, Zheng, Daniel O’Neill, and Hamid Maei.
"Optimal demand response using device-based
reinforcement learning." IEEE Transactions on Smart
Grid 6.5 (2015): 2312-2324.
[2] Mohsenian-Rad, Amir-Hamed, et al. "Autonomous
demand-side management based on game-theoretic energy
consumption scheduling for the future smart grid." IEEE
transactions on Smart Grid 1.3 (2010): 320-331.
[3] Shirazi, Elham, and Shahram Jadid. "Optimal
residential appliance scheduling under dynamic pricing
scheme via HEMDAS." Energy and Buildings 93 (2015):
40-49.
[4] Rahim, Sahar, et al. "Exploiting heuristic algorithms to
efficiently utilize energy management controllers with
renewable energy sources." Energy and Buildings 129
(2016): 452-470.
[5] Mesarić, Petra, and Slavko Krajcar. "Home demand
side management integrated with electric vehicles and
renewable energy sources." Energy and Buildings 108
(2015): 1-9.
[6] Deilami, Sara, et al. "Real-time coordination of plug-in
electric vehicle charging in smart grids to minimize power
losses and improve voltage profile." IEEE Transactions on
Smart Grid 2.3 (2011): 456-467.
[7] Azar, Armin Ghasem, and Rune Hylsberg Jacobsen.
"Appliance scheduling optimization for demand
response." International Journal on Advances in Intelligent
Systems 9.1 & 2 (2016): 50-64.
[8] Ramanathan, Badri, and Vijay Vittal. "A framework
for evaluation of advanced direct load control with
minimum disruption." IEEE Transactions on Power
Systems 23.4 (2008): 1681-1688.
[9] Ahmad, Adnan, et al. "An optimized home energy
management system with integrated renewable energy and
storage resources." Energies 10.4 (2017): 549.
[10] Rasheed, Muhammad Babar, et al. "Priority and delay
constrained demand side management in real‐time price
environment with renewable energy source." International
Journal of Energy Research 40.14 (2016): 2002-2021.
[11] Ma, Kai, et al. "Residential power scheduling for
demand response in smart grid." International Journal of
Electrical Power & Energy Systems 78 (2016): 320-325.
[12] Marzband, Mousa, et al. "Real time experimental
implementation of optimum energy management system in
standalone microgrid by using multi-layer ant colony
optimization." International Journal of Electrical Power &
Energy Systems 75 (2016): 265-274.
[13] Lee, Jae Yong, and Seong Gon Choi. "Linear
programming based hourly peak load shaving method at
home area." Advanced Communication Technology
(ICACT), 2014 16th International Conference on. IEEE,
2014.
[14] Nadeem, Zunaira, et al. "Scheduling appliances with
GA, TLBO, FA, OSR and their hybrids using chance
constrained optimization for smart homes." Energies 11.4
(2018): 888.
