An Integer Linear Programming Based Optimization

for Home Demand-side Management in Smart Grid

Ziming Zhu, Jie Tang, Sangarapillai Lambotharan, Senior Member, IEEE,
Woon Hau Chin, Senior Member, IEEE and Zhong Fan, Senior Member, IEEE

Abstract—We propose a consumption scheduling mechanism distributed energy resource management, load management
for home area load management in smart grid using integer and demand response techniques, as discussed in [4]–[11].
linear programming (ILP) technique. The aim of the proposed The focus of the work in this paper is on power con-
scheduling is to minimise the peak hourly load in order to
achieve an optimal (balanced) daily load schedule. The proposed sumption scheduling known as demand-side management. In
mechanism is able to schedule both the optimal power and [4], a powerful convex optimization technique was proposed
the optimal operation time for power-shiftable appliances and to schedule the power of individual appliances. Both the
time-shiftable appliances respectively according to the power centralised optimization based on linear programming and
consumption patterns of all the individual appliances. Simulation the decentralised approach based on game theory have been
results based on home and neighbourhood area scenarios have
been presented to demonstrate the effectiveness of the proposed proposed. However, the optimization framework in [4] might
technique. not be suitable for all appliances in practice. This is because
some appliances have a fixed power consumption pattern,
Index Terms—Demand-side management, load scheduling, in- which means that once the appliance is turned on, it has to
teger linear programming work according to its own power consumption pattern until
the job is finished. In this case, we could optimize only the
starting time, but the power consumption during the operation
of the appliance is not under the control of the optimizer,
I. I NTRODUCTION hence it is not an optimization parameter. In addition for
consumer’s comfort, we wish to introduce a feature that once
Ever since the first time people were able to light their certain appliance such as washing machine and dish washer
rooms using a switch, the electric power system has been in is scheduled to start, it should be allowed to finish the task
place for more than 100 years. However, the infrastructure without any break.
of the existing system has been little changed by far [1]. In In this paper, inspired by the scheduling mechanism in
the recent years, the energy industry has been facing signif- [4], we propose a demand-side management method using
icant challenges in terms of environmental issues, security integer linear programming technique. The proposed mech-
and utility management. Many believe that a fundamental anism will be able to minimise the peak hourly load and
evolution of the electric power system is needed. The goal satisfy both the user preference and specific requirements of
of the transformation is to make the next generation power all individual appliances. The mechanism can be firstly applied
grid a green, reliable and intelligent system, which is generally to the home energy management unit/gateway embedded in
known as the Smart Grid. the smart meters, which will probably work as a load control
In smart grid, advanced electric technologies will improve unit as well, in the home area network (HAN). Provided
the efficiency of the central bulk generation and transmission the consumption patterns of all the appliances and the user’s
network. Medium/low voltage Distributed Generation (DG) own load plan, the meter will be able to make an optimal
system using various environment-friendly sources will be scheduling for all the connected appliances in the household.
seamlessly integrated. Information and communication tech- The mechanism can also be applied to the distribution stations
nologies (ICT) will be widely applied to the grid to increase (central control node) in the neighbourhood/local areas to
its operation efficiency, reliability and flexibility [2], [3]. achieve centralised load management. In the future, when real-
Eventually, consumers will get the benefit of lower bills while time communications and various smart grid services such as
proactively managing their electricity consumption. flexible demand response technique become widely available,
Smart grid has become an attractive research area. Currently, the mechanism may be further applied to distributed load
the main research directions are on designing workable ICT control for every individual consumer in the grid.
infrastructure and energy management applications such as The rest of the paper is organised as follows. The proposed

978-1-4577-2159-5/12/$31.00 ©2011 IEEE

Authorized licensed use limited to: PUNJAB UNIVERSITY. Downloaded on December 15,2023 at 11:39:56 UTC from IEEE Xplore. Restrictions apply.
 with flexible self-organising network deployments [12], [13]. It
is believed that WSN will be one of the promising technologies
for HAN communications. For communications beyond the
home environment, i.e. among the meters/gateways, control
nodes and operators, the obvious candidates are wireless cellu-
lar technologies and various broadband solutions (e.g. GPRS,
WiFi, or LTE). However, as for non-real-time scheduling
in our system, the information exchange may not require
communications with high data rate and low delay.

Fig. 1. System components. B. Consumption scheduling optimization

The consumption scheduling mechanism can be described
as a linear optimization problem which aims to minimise the
demand-side load management mechanism is described in hourly load, as shown below
Section II. Section III presents system simulation results and
conclusions are drawn in Section IV. min L (1)
L,xa,h ∈R

s.t. xa,h ≤ L, ∀h ∈ H, (2)
II. D EMAND - SIDE CONSUMPTION SCHEDULING a∈A
T
1 xa = la , ∀a ∈ A, (3)
A. System description
xa,h ≥ 0, (4)
Fig. 1 depicts the overall structure of the load management
system. As shown in the figure, the smart meter is the key where 1 = [1, 1, . . . 1]T . We define a as an individual appli-
component in the system. It connects with the user interface to ance in a set of appliances, A. x a = [xa,1 , xa,2 , . . . , xa,24 ]T
collect the user’s own power consumption plan and preference denotes the schedule plan for appliance a and x a,h is the
and display the scheduling information. On the other side, the scheduling variable which represents the power consumption
smart meter connects with all the home appliances, not only of appliance a in the particular hour h ∈ H, H = [1, 2, . . . , 24]
to provide electricity for the appliances but also to determine of the day. The variable L represents the peak hourly load. The
the total requirements and power consumption patterns of all cost function of the above optimization problem is to minimise
individual appliances. Based on all the collected information, the hourly load L subject to the constraints in (2)-(4). The
the meter will globally optimize the hourly consumption and hourly load should be greater than or equal to the sum of
schedule all appliances. For non-shiftable appliances such as the scheduled power for all appliances in that time. For each
TV and fridge which have fixed power requirement and opera- appliance, the total daily supply has to meet the requirement
tion period, the optimization will ensure continuous supply of la . The power consumption x a,h must be non-negative value,
power. The scheduling optimization will be carried out mainly i.e., xa,h ≥ 0. We use a time resolution of one hour for hourly
for the shiftable appliances. For time-shiftable appliances, such scheduling in this paper, hence x a contains 24 elements. This
as washing machine, the smart meter will be able to control resolution can be increased by increasing the length of x a .
the switch and provide sufficient electricity corresponding to We formulate the requirements for all the individual appli-
the power pattern during the scheduled periods. For power- ances and the user preferences as constraints and add them into
shiftable appliances, such as water boiler and electric vehicle the optimization problem (1)-(4). For non-shiftable appliances,
chargers, the smart meter will schedule flexible power and the hourly power requirement is fixed at δ a during its working
ensure the total supply. period from h as to haf where has , haf ∈ H. This constraint
The system can be further extended to multiple users’ can be written as
scenario where many smart meters are connected together
xa,h ≥ δa , ∀h ∈ [has , ha(s+1) , . . . , haf ]. (5)
and they agree to achieve a cooperative scheduling. The
central control node will take the overall responsibility of Note that a similar constraint can also be applied to some
scheduling the whole network and assigning individual meters of the shiftable appliances if the consumer has a specific
their corresponding tasks. requirement in terms of starting and finishing time for the
Effective communication networks are required for the appliance. For example, people usually control the electric
system. Wireless sensor networks (WSN) combines sensing heater using a programmable timer or a thermostat. This kind
and communications together, and provides low-cost and low- of heater can be treated as a non-shiftable appliance.
power information gathering, processing and communication For power-shiftable appliances with a standby power α a

Authorized licensed use limited to: PUNJAB UNIVERSITY. Downloaded on December 15,2023 at 11:39:56 UTC from IEEE Xplore. Restrictions apply.
and a maximum working power β a , together with a possibly constraints as
preferred working period, the consumption requirement can be
min L
written as L,xa,h ∈R,st ∈Z24×1
 
αa ≤ xa,h ≤ βa , ∀h ∈ [has , ha(s+1) , . . . , haf ]. (6) s.t. xa,h + xt,h ≤ L, ∀h ∈ H,
a∈A−T t∈T
Now consider a time-shiftable appliance that consumes 1T xa = la , ∀a ∈ A − T,
power la . If we impose only the constraint (3), the optimization
will ensure supply of a total power of l a over the 24 hours. xa,h ≥ 0,
For example, if a washing machine requires 1.5kWh for its xa,h ≥ δa , ∀h ∈ [has , . . . , haf ],
operation, the constraint in (3) could ensure 1.5kWh power αa ≤ xa,h ≤ βa , ∀h ∈ [has , . . . , haf ],
supply over 24 hours. This total power can be distributed 0 ≤ st ≤ 1, 1T st = 1, ∀t ∈ T,
in any possible ways. For example, the vector x a can have
xt = PTt st , ∀t ∈ T. (10)
0.25 for six hours and zeros for the rest. However, a washing
machine cannot operate according to this power schedule. It Note that at this stage, the scheduling objective for the time-
might require 1kW for the first hour of operation and 0.5kW shiftable appliance is the switch vector s a , not the power x a,h .
for the second hour of operation. This kind of appliances will Therefore, we have to remove the corresponding appliance
need to operate according to their own power consumption from the appliance set A. The set A − T now contains
pattern for correct operation. Therefore, only a constraint in non-shiftable and power-shiftable appliances, whose power
the form of (3) is inadequate to ensure feasible operation of allocations are still needed to be scheduled by x a,h .
certain appliances. Considering this, we model this kind of The optimization problem above is a mixed integer linear
requirement as follows. programming problem which contains both integer variables
and non-integer variables. It can be solved using Branch and
Suppose, an appliance a has a fixed consumption pattern
Bound method [14]. Note that if there are more than one
pa = [pa,1 , . . . , pa,24 ]T , the schedule result x a has to be
optimal solution (with equal cost value) to the problem, the
exactly the same as one of the cyclic shifts of that pattern.
method will provide one of them.
All the possible shifts for the vector p a and hence all the
The above scheduling scheme is suitable for fixed price
possible xa can be put in a matrix form as
  scenario where the unit price of the electricity consumption is
Pa = ca1 ca2 ... ca24 fixed at any time. In this case, consumers have little incentive
⎡ ⎤ to shift their consumption. However, operators who want to
pa,1 pa,24 ... pa,3 pa,2
⎢ pa,2 pa,1 ... pa,4 pa,3 ⎥ reduce the peak load will be able to achieve direct load control
⎢ ⎥
=⎢ . .. . . . .. ⎥ . (7) without hurting the consumers’ preferences by applying this
⎣ .. . . .. . ⎦ scheduling scheme. If the user is on a particular tariff, e.g.
pa,24 pa,23 ... pa,2 pa,1 the UK Eco 7 which provides much lower price for late night
The above constraint cannot be directly formulated using consumption [15], the users may change their preference and
linear programming. Hence, we apply integer linear program- the mechanism will also be able to provide optimal schedule.
ming (ILP) framework to formulate the constraint. The mechanism can also be further applied to distributed
control systems or real-time scheduling with flexible
Firstly, we define the schedule plan as x t = demand response and pricing policies. The main difference
[xt,1 , xt,2 , . . . , xt,24 ]T ≥ 0, and the appliance set is the distributed control system may probably need a
T ⊂ A for the time-shiftable appliance t ∈ T. Then, we perfect communication system to support the exchange of
define a binary integer vector s t = [st,1 , st,2 , . . . , st,24 ]T information.
as the switch control for the time-shiftable appliance. There
is only one non-zero element in the vector s t which is equal
to one. This property is written as follows III. S IMULATION RESULTS

1T st = 1, st ∈ {0, 1}24×1. (8) A set of system simulations has been carried out using Mat-
lab to implement the proposed scheduling mechanism in home
Now the schedule plan can be written as area and neighbourhood area scenarios. In the simulation,
we first define a set of home appliances and their individual
xt = PTt st . (9)
requirements as listed in Table I.
Then, we rewrite the optimization problem in (1) by adding According to the settings, we are able to formulate the total
st as an integer optimization variable, together with other power requirement and the individual consumption patterns to

Authorized licensed use limited to: PUNJAB UNIVERSITY. Downloaded on December 15,2023 at 11:39:56 UTC from IEEE Xplore. Restrictions apply.
 1.5
Name Type User preference and Scheduled load

power requirement
1.Hob Non- Operating period:
and oven shiftable 7pm-8pm(Hob/oven) 1

Power consumption (kwh)

2.heater 9pm-10pm, 3am-5am(heater)
hourly consumption: 1kWh
3.Fridge Non- Operating 24hrs 0.5

and freezer shiftable hourly consumption: 0.12kWh

4.Water boiler Power- Hourly consumption: 0-1.5kWh
shiftable daily requirement: 3kWh
0
5.Electric vehicle Power- Preferred charging period: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (hour)

(15 miles daily shiftable 8pm-8am

driving [16]) charging power: 0.1kW-3kW Fig. 2. Hourly power consumption schedule.
daily requirement: 5kWh
6.Washing Time- Operating 2hrs, once per day 1.5

machine shiftable 1kWh for the 1st hr app1 app2 app3 app4 app5 app6 app7

0.5kWh for the 2nd hr

7.Dish washer Time- power: 0.8kWh for 1 hour

Power consumption (kwh)

1

shiftable daily requirement: 0.8kWh

TABLE I
A PPLIANCES AND POWER CONSUMPTION PATTERNS
0.5

the corresponding constraint formats as in (5), (6), (8) and (9). 0

0 2 4 6 8 10 12 14 16 18 20 22 24 25
For example for appliance 1, which is the hob and oven, we Time (hour)

can use the expression in (5) and make the constraint as

Fig. 3. Optimal schedule for individual appliances.
x1,h ≥ 1, ∀h ∈ [19, 20]. (11)
Similarly for the electric vehicle battery charger we can
write is increased, a more balanced hourly power consumption
0.1 ≤ x5,h ≤ 3, ∀h ∈ [1, 8] ∪ [20, 24]. (12) schedule can be expected.
Fig. 3 depicts the scheduling result for the individual
For the time-shiftable washing machine, we defined the appliances. Clearly, we can see that during 9-10pm, the
power pattern as p 6 = [1, 0.5, 0, · · · , 0]T and the switch vector heater (app2) is consuming fixed 1kWh and the fridge (app3)
s6 to formulate the constraint as is consuming 0.12kWh. The water boiler (app4) has been
totally shut down and only 0.1kWh which is the minimum
x6 = PT6 s6 . (13)
charging power requirement for the electric vehicle (app5)
After incorporating all the constraints together, we can solve is provided. Time-shiftable appliances were not scheduled to
the ILP problem in (10) and achieve the optimal scheduling. operate during the period. Indeed, the washing machine (app6)
Fig. 2 depicts the scheduled hourly load. It can be seen that the and the dish washer (app7) are both scheduled in the day time
maximum hourly load of 1.22kWh appears during the hours between 2pm-5pm where there is lower demand from other
of 3-5, 14 and 19-22. The minimum load of 0.3kWh appears appliances.
during the hours of 9-13 and 17-18. From the appliances We further carry out a simulation for the scenario of a small
consumption requirement, we can see that when the heater is neighbourhood area with multiple households. Assume there
turned on between 9-10pm, and considering the power of the are 4 houses and the total daily load requirement is 43 units.
fridge and the charging of electric vehicle battery, the required Each house has similar appliances as in the previous case
load is 1+0.12+0.1=1.22 (kWh). For any time of the day, the but with different consumer preferences and power patterns
scheduled load is less than or equal to 1.22kWh. Hence we for some of the appliances. All the required information is
claim that the optimal hourly scheduling with a minimum collected from the smart meter of each household. Now a
possible peak is achieved. When the number of appliances centralised scheduling which will meet all the requirements

Authorized licensed use limited to: PUNJAB UNIVERSITY. Downloaded on December 15,2023 at 11:39:56 UTC from IEEE Xplore. Restrictions apply.
 2.5
[3] P. Zhang, F. Li, and N. Bhatt, “Next generation monitoring, analysis and
Scheduled load
control for the future smart control centre,” IEEE transactions on Smart
Grid, vol. 1, no. 2, pp. 186–192, Sep. 2010.
2
[4] A. Mohsenian-Rad, V. Wong, J. Jatskevich, R. Schober, and A. Leon-
Garcia, “Autonomous demand-side management based on game-
Power consumption (kwh)

1.5
theoretic energy consumption scheduling for the future smart grid,” IEEE
transactions on Smart Grid, vol. 1, no. 3, pp. 320–331, Dec. 2010.
[5] C. Wietfeld, C. Muller, J. Schmutzler, S. Fries, A. Heidenreich, and
1 H. J. Hof, “ICT reference architecture design based on requirements
for future energy marketplaces,” in 1st IEEE International Conference
on Smart Grid Communications, pp. 315–320, Gaithersburg, Maryland,
0.5 USA, Oct. 2010.
[6] US Department of Energy, “Benefits of demand response in electricity
markets and recommendations for achieving them,” Report to the US
0
0 2 4 6 8 10 12 14 16 18 20 22 24 25 Congress., Feb. 2006. Available http://eetd.idi.gov.
Time (hour)
[7] T. Sauter and M. Lobashov, “End-to-end communication architecture
for smart grids,” IEEE Transactions on Industrial Electronics, vol. 58,
Fig. 4. Hourly power consumption schedule for multiple households. no. 4, pp. 1218 –1228, Apr. 2011.
[8] B. Jansen, C. Binding, O. Sundstrom, and D. Gantenbein, “Architec-
tureand communication of an electric vehicle virtual power plant,” in
1st IEEE International Conference on Smart Grid Communications,
is achievable. pp. 149–154, Gaithersburg, Maryland, USA, Oct. 2010.
[9] A. Bose, “Smart Transmission Grid Applications and Their Supporting
Fig. 4 depicts the overall scheduled hourly load. As seen,
Infrastructure,” IEEE trans. on Smart Grid, vol. 1, no. 1, pp. 11–19,
the optimized hourly peak load is 2.14kWh which is just Jun. 2010.
around one unit higher than that of the previous case. Besides, [10] J. Medina, N. Muller, and I. Roytelman, “Demand response and distribu-
tion grid operations: Opportunities and challenges,” IEEE Transactions
the overall load allocation is more balanced over the 24 hours.
on Smart Grid, vol. 1, no. 2, pp. 193 –198, Sep. 2010.
The peak to average load ratio is 2.14/(43/24) = 1.19, which [11] K. Hamilton and N.Gulhar, “Taking demand response to the next level,”
means the peak load is just 19% higher than the daily average. IEEE Power and Energy Magazine, vol. 8, no. 3, pp. 60–65, Jun. 2010.
[12] I.F. Akyildiz, et al., “A survey on sensor networks,” IEEE Communica-
It can be observed that the scheduling mechanism is able to
tions Magazine, vol. 40, no. 8, pp. 102–114, Aug. 2002.
reduce the maximum load and improve the performance and [13] V.C. Gugor, et al., “Opportunities and challenges of wireless sensor
the reliability of the power grid due to reducing the peak to networks in smart grid,” IEEE trans. on Industrial Electronics, vol. 57,
no. 10, pp. 3557–3564, Oct. 2010.
average load.
[14] G. L. Nemhauser and L. A. Wolsey, Integer and combinatorial opti-
mization. John Wiley and Sons, New York, 1998.
[15] http://www.energychoices.co.uk/economy 7.html
IV. C ONCLUSION [16] Renault. Fluence Z.E. Specifications, http://www.renault-ze.com/.

We have proposed an integer linear programming

based optimization mechanism for the home demand-side
management in smart grid. The proposed mechanism is VI. B IOGRAPHIES
able to schedule both the optimal power and the optimal Ziming Zhu is a Ph.D. student within the Advanced Signal Processing
operation time for power-shiftable appliances and time- Group (ASPG), Department of Electronic and Electrical Engineering,
shiftable appliances respectively according to user preference Loughborough University, UK. His research interests include resource
management and optimization of wireless networks and smart grid
and the power consumption patterns of individual appliances. applications.
Simulation results demonstrated the effectiveness of the
mechanism. When multiple households participate in the Jie Tang is a Ph.D. student within the ASPG, Loughborough University,
UK. He currently works on optimization of spectrum sharing networks and
scheduling, a more balanced hourly load is achieved. interference alignment.

Prof. Sangarapillai Lambotharan is within the ASPG, Loughborough

V. ACKNOWLEDGEMENTS University, UK. His research interests include wireless relay networks and
cognitive radio networks.
The authors would like to express sincere thanks to Toshiba
Research Europe Limited for funding this work. Dr. Woon Hau Chin is a principal research engineer & team leader at
Toshiba Research Europe Limited. He currently works on signal processing
algorithms, algorithms design, wireless systems, and standardization.

R EFERENCES Dr. Zhong Fan is a chief research fellow at Toshiba Research Europe
Limited. He has substantial experience in wireless networks, IP networks,
[1] E. Santacana, G. Rackliffe, L. Tang, and X. Feng, “Getting smart,” IEEE M2M, IoT, and smart grid communications.
Power & Energy magazine, pp. 41–48, Apr. 2010.
[2] F. Li and et al., “Smart transmission grid: Vision and framework,” IEEE
transactions on Smart Grid, vol. 1, no. 2, pp. 168–177, Sep.2010.

Authorized licensed use limited to: PUNJAB UNIVERSITY. Downloaded on December 15,2023 at 11:39:56 UTC from IEEE Xplore. Restrictions apply.