demand response management for residential-ieee journals pdf

SPECIAL SECTION ON SMART GRIDS: A HUB OF INTERDISCIPLINARY RESEARCH

Received October 1, 2015, accepted October 11, 2015, date of publication November 24, 2015,date of current version December 10, 2015.

Digital Object Identifier 10.1109/ACCESS.2015.2503379

Demand Response Management for ResidentialSmart Grid: From Theory to PracticeWEN-TAI LI1, (Student Member, IEEE), CHAU YUEN1, (Senior Member, IEEE),NAVEED UL HASSAN2, (Senior Member, IEEE), WAYES TUSHAR1, (Member, IEEE),CHAO-KAI WEN3, (Member, IEEE), KRISTIN L. WOOD1, KUN HU4,AND XIANG LIU4, (Member, IEEE)1Singapore University of Technology and Design, Singapore 4873722Electrical Engineering Department, Lahore University of Management Sciences, Lahore 54792, Pakistan3Institute of Communications Engineering, National Sun Yat-sen University, Kaohsiung 804, Taiwan4School of Software and Microelectronics, Peking University, Beijing 100871, China

Corresponding author: W.-T. Li ([email protected])

This work was supported by the Singapore University of Technology and Design through the Energy Innovation Research Program,Singapore, under Grant NRF2012EWT-EIRP002-045.

ABSTRACT In recent years, many studies have investigated the potential of demand response manage-ment (DRM) schemes to manage energy for residential buildings in a smart grid. However, most of theexisting studies mainly focus on the theoretical design of DRM schemes and do not verify the proposedschemes through implementation. Smart grid research is highly interdisciplinary. As such, the establishmentof testbeds to conduct DRM requires various skill sets that might not always be possible to arrange. However,the implementation of a DRM scheme is critical not only to verify the correctness of the design in apractical environment but also to address many important assumptions that are necessary for the actualdeployment of the scheme. Thus, the theoretical aspect of DRM solutions should be discussed and verifiedin a practical environment to ensure that the scheme is suitable for deployment. In this paper, we proposea DRM scheme and construct a residential smart grid testbed to implement the proposed scheme. In theproposed DRM scheme, we suggest two different types of customer engagement plans, namely, greensavvy plan and green aware plan, and design algorithms based on two user inconvenience indices toevaluate DRM for peak load reduction. The testbed verifies the effectiveness and efficiency of the proposedDRM scheme.

INDEX TERMS Smart grid, user inconvenience, peak load reduction, customer engagement plan, demandresponse, energy management service (EMS), implementation.

I. INTRODUCTIONRecently, the demand for electricity has been increasingas a result of economic and industrial developments. Forinstance, a recent annual energy outlook report of the UniteState Energy Information Administration forecasted that theresidential electricity demand will increase by 24% withinthe following several decades [1] because of the substantialincrease in population. A smart grid, with its two-way com-munication and power flow features, is a potential solution tomeet this increasing demand. In particular, demand responsemanagement (DRM) has been identified as one of the keycomponents of the smart grid that can help the power marketincrease the efficient use of energy by remote monitoringand control of electricity load and by setting efficient energyprices with a view to shift the high demand of electricity usersto an off-peak period.

To this end, significant efforts have been exerted in thepast few years to investigate the potential of DRM schemes inmanaging energy for residential buildings in the smart grid.Examples of such studies include recent [2]–[11] in 2015and [12]–[16] in 2014. In [2], an optimization-based homeload control scheme is proposed to manage the operationperiods of responsive electrical appliances. The study alsorecommend several operation periods for non-responsiveloads. Two types of customer engagement plan are designedin [3] to specify the amount of intervention in the customers’load settings to reduce the peak load. Chuan and Ukil [4]establish a mathematical model to represent and modelthe load profile type of residential buildings in Singaporeby describing in detail their energy requirement and con-sumption patterns. Methodologies for managing the temper-ature sensitive parts of residential electricity demand, that is,

VOLUME 3, 20152169-3536 2015 IEEE. Translations and content mining are permitted for academic research only.

Personal use is also permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

2431

W.-T. Li et al.: DRM for Residential Smart Grid

heating, ventilation, refrigeration, and air conditioning (AC)are developed in [5] and [6]. In [7] and [8], the authorsinvestigate energy scheduling techniques with delay-tolerantdemands in residential smart grids. Different techniquesdeveloped in 2015 for residential DRM can be foundin [9]–[11].

Numerous studies have also been conducted in 2014 toaddress the issues of residential DRM. For instance, an energyconsumption scheduling scheme for household appliances isproposed in [12] to reduce the peak-to-average ratio of theelectricity system. In [13], a distributed direct load controlscheme is proposed with a view to match the actual aggre-gated demand with the desired aggregated demand profile.A coordinated DRM scheme is designed in [14] to gener-ate the load profile and minimize the individual cost to thecustomers. In [15] and [16], DRM schemes are designedbased on the time-of-use electricity distribution tariff andcustomer reward plan, respectively. We stress that numerousother studies have discussed similar issues, which can befound in the literature reviewed in [17]–[20].

Most of the existing studies, mainly focus on thetheoretical design of DRM schemes without any verifica-tion through implementation. Smart grid research is highlyinterdisciplinary. As such, the establishment of testbeds toconduct DRM requires various skill sets that might not alwaysbe possible to arrange. However, the implementation of theDRM scheme is critical not only to verify the correctnessof the design in a practical environment but also to addressmany important assumptions that are necessary for actualdeployment of the scheme. For instance, a wireless sensornetwork (WSN) is suitable for various control and mon-itoring applications of the smart grid and has been usedin most smart grid-related studies. Although WSNs canprovide cost-efficient and reliable solutions [21], they areunsuitable for delay-critical applications, which may be chal-lenged when sudden failures occur in the monitored environ-ment [22]. Thus, the suitability of different communicationfacilities should be understood for different smart grid appli-cations [23], which can be addressed once we implement thesystem and verify the designed DRM technique.

Several studies have implemented testbeds to validatepotential applications related to the smart grid [24]–[28].These testbeds have various aims, scales, limitations, andfeatures. For instance, in [24], a microgrid testbed is estab-lished and combined with a cognitive radio network asa means of communication. Two research groups havedeveloped experimental testbeds in laboratory environmentsto validate their research as discussed in [25] and [26].In [27] and [28], the authors demonstrate two microgridtestbeds in campus environments that contain hybrid energysources. However, there has been limited effort in the liter-ature on constructing a testbed for residential DRM studies.Thus, constructing a testbed for residential DRM will verifynot only the theoretical aspect of DRM solution but alsothe schemes that are suitable for deployment in a practicalenvironment.

To this end, this study proposes a control mechanism forDRM in residential buildings and implements the scheme ina smart grid testbed. The main contributions of this study aredescribed as follows:• In the proposed control mechanism, we suggesttwo types of customer engagement plans, namely, GreenSavvy Plan (GSP) and Green Aware Plan (GAP).We also design algorithms based on two user incon-venience indices to evaluate DRM for peak loadreduction.

• For residential DRM, we construct a testbed to imple-ment the proposed scheme and conduct experiments toverify the effectiveness and efficiency of the proposedschemes.

The remainder of the paper is organized as follows:The system model of the proposed scheme is discussedin Section II. The customer engagement plans are pre-sented in Section III. The control algorithms are developedin Section IV. The testbed and experimental results areexplained in Section V. The study is concluded in Section VI.

FIGURE 1. Demonstration of the system model considered in this study.

II. SYSTEM DESCRIPTIONThe system model considered in this study mainly consistsof three layers based on the types of activities conductedfor DRM and is shown in FIGURE 1. The proposed archi-tecture comprises several households, where each house-hold has numerous electrical appliances. These appliancesare divided into essential and flexible load categories. Thepower consumption requirements and scheduling times ofessential loads are fixed and cannot contribute to the DRM.By contrast, flexible loads can be controlled and their powerconsumption and scheduling time can be adjusted accordingto the DRM requirements. Lighting load can be an example offlexible load as we can switch off unnecessary or extra lightsthat would result in power saving. AC load can be anotherexample of flexible as we can switch off or operate the AC ata higher thermostat set point to save power.

2432 VOLUME 3, 2015


We adopt an open Internet-of-things architecture, where acentralized database records the energy usage of all electricalappliances from all households [29]. Therefore, in the pro-posed architecture, each household is connected to a homegateway and several smart appliances that can communicatewith each other. The home gateway is controlled througha home gateway manager in the cloud. Such architectureenables multiple third parties to connect to the system to pro-vide services. For example, a home automation and securitycompany may let households subscribe to its service, where ithelps monitor the home when the owners are away and soundan alarm to the owners when an intrusion is detected byvarious sensor and energy readings in the database. Similarly,an energy management service (EMS) company can partici-pate in an ancillary energy market, which is the focus of thiscase study.

The EMS provider may have a deal with the grid inproviding services to maintain the load below a certainthreshold. Then, the EMS provider recruits house-holds by offering them customer engagement plans thatdescribe incentive and inconvenience. The objective of theEMS provider is to earn some profit while meeting the targetload reduction required by the grid operator by recruiting asufficient number of residents and offering them appropriateincentive. We will discuss this concept and implement atestbed comprising the EMS provider that controls a set ofappliances that represent various households.

III. CUSTOMER ENGAGEMENT PLANSThe EMS provider can offer customer engagement plansto recruit households to participate in DRM to reduce athreshold, as desired by the grid operator. We proposecustomer engagement plans that specify the amount of incon-venience and incentive. In this study, we assume that, forDRM, the EMS provider divides customers into the followingtwo categories:• Green savvy: Users that accept more inconvenience butwith more incentive;

• Green aware: Users that accept less inconvenience butwith less incentive.

For these customer types, we assume that the EMS providerdefines two different customer engagement plans termedas (i) GSP and (ii) GAP. These plans specify the inconve-nience and incentive for green savvy and green aware users.We let 0GSP and 0GAP denote the inconvenience defined inthe two plans, such that 0GSP

≥ 0GAP. Similarly, we let3GSP and 3GAP denote the incentive per unit of inconve-nience offered by the EMS provider in the two plans, suchthat 3GSP

≥ 3GAP.In this study, we define inconvenience in twoways, namely,

1) time cut and 2) power cut. In the time cut method,we define inconvenience in terms of number of minutes(or hours) the user is denied operation of its demandedappliances (regardless of their power consumption) and theincentive is offered as dollars per minute (or hour) of incon-venience. In the power cut method, we define inconvenience

TABLE 1. Sample GSP and GAP plans based on time cut and power cutmethods.

in terms of the number of watt-minutes (watt-hours) thatare cut by the EMS provider for DRM and the incentiveis offered as dollar per Kilowatt-hour (KWh) for the powerbeing cut during the duration. In Table 1, we provide sampleGSP and GAP based on the time cut and power cut methods.1

For example, in the time cut method, GSP users are offered1.5$/h as incentive, which is 1.5 times compared with that ofthe GSP user. As such, GSP users might accept three timesinconvenience at most compared with GAP users and earn13.5$/day at most when the EMS denied its service 3 h/day tothe GAP user, whereas GAP users accept less inconvenience(3 h/day at most) but also less incentive of 1$/h, such thatthe user in this plan can earn 3$/day at most. By contrast,in the power cut method, inconvenience is specified in termsof the total amount of power cut. Similarly, the GSP useris offered an incentive of 4.5$/kWh and accepts three timesinconvenience at most compared with the GAP user(i.e., power of the GSP user is cut 3 kWh/day by the EMSand earn 13.5$ at most). By contrast, for the GAP user, notmore than 1 kWh/day is cut with only 3$/kWh as incentive(giving the user 3$/day at most).

IV. PROBLEM FORMULATION ANDCONTROL ALGORITHMSIn this section, we formulate a generic optimization problem,followed by control algorithms that can be used to imple-ment the customer engagement plans based on inconvenienceindices (time cut or power cut). In this formulation, we asso-ciate weights for each user and for each appliance of eachhousehold. Then, these weights are dynamically adjustedaccording to the user subscription plans (GSP or GAP) andthe criterion (time cut or power cut) being used by theEMS provider.

A. PROBLEM FORMULATIONAs mentioned in Section II, appliances can be divided intoessential and flexible load categories. Therefore, we let Pej (t)denote the total essential load of user j at time t . We alsoassume that all appliances mentioned in the subsequentdescription belong to flexible loads. We let pi,j denote therated power consumption of the i-th appliance of user j, andsi,j = {si,j(1), . . . , si,j(T )} denote the demand status vector ofload i of user j, where the time index of DRM goes from t = 1to t = T and si,j(t) = 1 when appliance i is demanded attime t by user j and zero otherwise. Thus, di,j(t) = si,j(t)pi,j(t)denotes the power consumption of appliance i of user j at

1The exact design of GSP and GAP, that is, the offered values of incon-veniences and incentives respectively can be obtained by the EMS providerusing surveys (such that the values are acceptable to the end users).

VOLUME 3, 2015 2433


time t . Then, the total power demand of user j at time t isexpressed as Pj(t) =

∑i di,j(t) + P

ej (t). Then, we obtain the

total power demand at time t , as follows:

PU (t) =∑j

Pj(t). (1)

However, the grid operator will require the EMS providerto keep the total power consumption of the users below acertain threshold Ppeak, that is, PU (t) ≤ Ppeak(t), because ofeconomic scheduling or stable operation of generators. WhenPU (t) > Ppeak(t), the EMS provider will conduct DRM byswitching off several appliances of the users to maintain thetotal power demand below the given threshold. Thus, theseusers whose appliances are switched off by the EMS providerwill experience inconvenience at time t . As such, theEMS provider should consider not only the engagementplans of users but also the inconvenience of users when theEMS provider conducts DRM.

To this end, we propose a method that associates weightfactors with each user and also with each appliance in eachhousehold. Let ωj(t) be the user inconvenience weight factorthat controls user j’s inconvenience based on his subscriptionplan and let αi,j(t) be the appliance preference factor thatcontrols the ratio of influence in terms of the i-th applianceof user j based on his declaration preference. Subsequently,we formulate an optimization problem whose goal is to min-imize user inconvenience and which is subject to total powerdemand below the required threshold levelPpeak, and the finalresults conform with the subscription plans of users. Thisprocedure is equivalent to determining the optimal demandstatus, such that the sum-weight is minimized at each timeslot. For ease of notation, we drop index (t) from all thesucceeding description because we perform the optimizationproblem at each time slot. Thus, the optimization problem isformulated as

minli,j,∀i,j

∑j

ωj∑i

αi,jli,j

s.t.∑i,j

pi,jsi,jli,j ≥⟨PU − Ppeak

⟩+li,j ∈ {0, 1}, ∀i, j, (2)

where 〈x〉+ = max{x, 0}, and li,j is a binary value. If thei-th appliance of the j-th user is switched off at this time,then li,j = 1; otherwise li,j = 0. When the EMS providerturns off the appliances of certain users at this time slot,it raises the inconvenience of these users. In this case,the afected user inconvenience weight factors increase,thereby reducing the chance of these users being turned off inthe next time slot. Similarly, users can also set their preferencefor the appliances to be turned off. In this case, the appliancepreference factor serves a role, such that the less preferredappliances are turned off first. Using these concepts, the userinconvenience weight factors and the appliance preferencefactors are updated at each time slot.

B. WEIGHT UPDATINGThe user inconvenience can be defined based on manydifferent criteria and presented in many different ways.As mentioned before, in this paper, we define inconvenienceinto two criteria: 1) time cut and 2) power cut. Therefore,we propose a method that designs a factor called user incon-venience weight factor to indicate and quantify the inconve-nience.2 Moreover, we design two user inconvenience weightfactors based on both criteria. Suchweight factors are adaptedto the proposed plans and also consider the fairness for userssubscribing to the same plan. The fairness can be indicatedin terms of both criteria, such as equality of total time cutor equality of total power cut. Moreover, fairness is alsoconsidered between each appliance in the same household.The EMS provider aims to keep fairness for each appliancein the same household, except when users declare their pref-erence for the appliances. Thus, we also design two appliancepreference factors in terms of both criteria and integrate themrespectively according to the following user inconvenienceweight factor:1) Time Cut Method

ωnewj = ωj

(1+ ρj1t ×

∑i

li,j), (3)

αnewi,j =βnewi,j∑i β

newi,j

, (4)

βnewi,j = βi,j(1+ γi,j1t × li,j

), (5)

2) Power Cut Method

ωnewj = ωj

(1+ ρj1t ×

∑i

pi,jli,j), (6)

αnewi,j =βnewi,j∑i β

newi,j

, (7)

βnewi,j = βi,j(1+ γi,j1t × pi,jli,j

), (8)

where 1t is the time slot, ρj is the plan parameter, and γi,j isthe preference parameter. Therefore, the EMS provider canensure fairness for users subscribing to the same plan andeach appliance by updating the user inconvenience weightfactor and appliance preference factor after performingDRM.

C. CONVEX RELAXATIONConsidering tht problem (2) is an integer linear pro-gramming (ILP) and that such problem is generally anNP-hard problem. Although many approaches have beendeveloped to address ILP, they are still impractical becausethey do not guarantee solving the problem in polynomialtime, especially for large systems, such as the proposedsystem which consists of a large number of appliances.To address problem (2), we relax the constraint of li,j ∈ {0, 1}to li,j ∈ [0, 1]. Moreover, the selection li,j must satisfy the

2A similar strategy has proven useful in [30] for the application of electricvehicle charging.

2434 VOLUME 3, 2015


constraints of the real status. For example, if si,j = 1, thenli,j can be 1 or 0. By contrast, if si,j = 0, then li,j canonly be zero. Consequently, the optimization problem can beapproximated as

minli,j,∀i,j

∑j

ωj∑i

αi,jli,j

s.t.∑i,j

pi,jli,j ≥⟨PU − Ppeak

⟩+0 ≤ li,j ≤ si,j, ∀i, j (9)

The optimization problem is a linear programming (LP) andthus a convex optimization problem. To solve the problem,we employ CVX, a package of MATLAB for specifying andsolving convex programs [31].

The relaxed problem is not generally equal to the orig-inal problem because the optimal results of the relaxedproblem, l?i,j, can be fractional. Thus, l?i,j should be mappedinto 0 or 1 by using a threshold to obtain the desired results,which can be performed as

l̂?ji =

{1, if l?ji ≥ τ,

0, if l?ji < τ,(10)

where τ is a given threshold. In our experience, τ = 0.5 cangenerally yield good results. Hence, we adopt such setting(i.e., τ = 0.5) in our experiments.

Algorithm 1 Control Algorithm for DRMinitialize: ωj(0) = 1, αi,j(0) = 1/nj, ∀i, j, t = 1input : the plan parameter ρj, and the preference

parameter γi,j1 Read si,j(t), pi,j(t), Pej (t), P

peak(t), ∀i, j;2 Compute PU (t) by using (1);3 Solve Problem (9) by using CVX;4 Decide l̂?ji ∀i, j by using (10);5 Weightupdating:

Timecutmethod:Update user inconvenience weight factors andappliance preference factors by using (3)–(5).

Powercutmethod:Update user inconvenience weight factors andappliance preference factors by using (6)–(8).

6 t = t + 1;7 Repeat to 1;

The implementation procedure of the proposed con-trol algorithm is summarized in Algorithm 1. In short,in steps 1 and 2 of the algorithm, the EMS providercomputes total the demand power and reads the thresh-old given by the grid. In step 3 and 4, the EMS providerdecides the optimal demand status by solving problem (9).In step 5, the EMS provider updates user inconvenienceweight factors and appliance preference factors based onthe method it uses, that is (3)–(5) or (6)–(8). Finally,

the algorithm proceeds to the next time slot and repeatsfrom step 1.

V. TESTBED AND EXPERIMENTAL RESULTSIn this section, we describe the testbed focusing on the res-idential DRM. We then perform a DRM experiment on ourtestbed to demonstrate the effectiveness and efficiency of theproposed plans and methods.

A. TESTBED SETUPThe overall architecture of the proposed system is shownin FIGURE 1; it mainly consists of of three layers: third partyEMS, cloud server, and households. We have built the testbedshown in FIGURE 2, which consists of the three modules, tomatch the proposed architecture. The details of the implemen-tations are described in the subsequent subsections.

FIGURE 2. Demonstration of testbed module.

1) CLOUD SERVERThe cloud server is where we can store data of households andprovide services for both third party management servicesand households. In our testbed, we implement a cloud server,which consists of a home gateway manager and database, andselect XMPP and RESTful HTTP as the network protocol tobe used in the proposed system. The extensible messagingand presence protocol (XMPP) is an application profile ofthe extensible markup language (XML), which enables thenear-real-time exchange of structured yet extensible databetween multiple network entities [32]. In the proposed sys-tem, the control commands must also be pushed toward theend users for different applications, which require an asyn-chronous communication model [33]. XMPP is capable notonly of sending asynchronous request but also of supporting amassive number of users at the same time through its pub/subprotocol [34]. Hence, XMPP is selected to send control com-mands from the home gateway manager of the cloud to all thegateways of all the households. In addition, RESTful HTTPis lightweight, has a simple HTTP request format, and is

VOLUME 3, 2015 2435


TABLE 2. Description of test-bed equipment.

very easy to implement. Moreover, RESTful HTTP is bestsuited for applications that require periodic communication.Hence, RESTful HTTP is selected for the periodic uploadingof sensor data from the gateway of households to the databaseof the cloud.

FIGURE 3. Household module of testbed.

2) HOUSEHOLDSWe also build a home gateway (HG) and several smart plugsfor each HG to simulate a household, as shown in FIGURE 3.HG is responsible for two tasks: 1) collecting data of appli-ances and uploading to the database of the cloud server and2) receiving control command from the cloud server andsending the control command to the respective devices.HG, which is an IP-based system and possesses a two-waycommunication channel, the HTTP and XMPP protocols, canserve as a translator and communicate with non-IP baseddevices, such as the smart plugs in the system. In our testbed,the Raspberry pi computer (Model-B Rev 1) serves the roleof HG and is connected to the z-wave smart plug througha RaZberry module, as shown in FIGURE 2. Smart plugscan be equipped with different appliances to send the data ofenergy usage to HG and perform the control command fromHG (e.g., switch on/off).

3) THIRD PARTY MANAGEMENT SERVICESFinally, a PC is used as third-party management service sothat EMS is implemented based on MATLAB 2014b. SuchEMS can access data from the database of the cloud serverby RESTful HTTP protocol and perform DRM, which solvesthe optimization problem (9) through the CVX package ofMATLAB. Then, EMS sends the control command based

on the solution results to the home gateway manager of thecloud server. We summarize the above descriptions and theequipment of the testbed in Table 2.

FIGURE 4. Demonstration of (a) base load demand and (b) probability ofremaining turned on of the controllable loads in the considered system.

B. EXPERIMENTAL RESULTSIn our experiment, we use the testbed to simulate a smallresidential community consisting of ten households, witheach household equipped with one HG and two smart plugs.3

Lighting appliances are considered to be flexible loads andequipped with smart plugs to be controllable with two kindsof states: on and off. Each household then contains two light-ing appliances, with the light bulb range between 500 wattsto 200 watts; we always place the light bulb with the higherrating as appliance 1.We assume the total essential load of thehouseholds as the base load. The base loadmodel, as shown inFIGURE 4(a), is assumed to be a random curve based on thereports of National Electricity Market of Singapore (NEMS)and is used to generate the daily base load curve. We generatethe daily usage patterns of the flexible loads based on theirprobability of being turned on, as shown in FIGURE 4(b).4

In FIGURE 5, we show how the proposed systemcan use the system to control the light appliances ineach household to reduce the peak demand in a smartgrid. According to FIGURE 5, a duration of 24 hoursis divided into 8640 time slots, that is, the smart griddetects the total demand every 10 seconds. The greenand blue zones of the figure denote the load profile with

3The system is designed such that the number of households and numberof appliances can be scaled easilywithout the need of changing the algorithm.

4We generate the base load and the daily usage patterns of the flexibleloads byMATLAB.We then send the commands in terms of generated usagepatterns to smart plugs to simulate the users’ behavior.

2436 VOLUME 3, 2015


FIGURE 5. Demonstration of peak load shaving through controlling loadsof the homes using the proposed architecture.

FIGURE 6. Demonstration of result without appliance’s preference(time cut method).

and without demand management, respectively; the grayzone indicates the base load, and the red line indicatesthe maximum allowable peak demand threshold (33 kW).As demonstrated in FIGURE 5, when the total demand isunder the peak limit, the smart grid does not need to doany controlling; hence, no difference is observed between thegreen zone and blue line. However, once the total demandexceeds the peak limit, the system controls the lights andturns some of them off to reduce the demand load promptly asindicated by the blue line from 9:00 to 18:00 hours. Thus, theresult in FIGURE 5. clearly shows that our testbed can effec-tively perform EMS, DRM, and other sevice applications.

To demonstrate the effectiveness and efficiency of the pro-posed plans and method, we assume that households 1 to 4adopt GAP and others agree with the GSP. Therefore, weset ρj = 3, j = 1, . . . , 4 and ρj = 1, j = 6, . . . , 10for both methods. Moreover, we also assume that users donot declare their preferences of appliances. We then set thedefault value for all the appliances, that is, γi,j = 1, ∀i, j.We simulate the usage patterns of one month and showthe result in the FIGURE. 6-9, which contains four sub-figures. The first row from left to right describes thetotal off-time of the household and the total off-time ofthe individual appliances. The second row from left to rightdescribes the total cut power of the household and the totalcut power of the individual appliances.

FIGURE 7. Demonstration of result without appliance’s preference(power cut method).

FIGURE 8. Demonstration of result with appliance’s preference (time cutmethod with less DRM over appliance 2).

FIGURE 9. Demonstration of result with appliance’s preference (powercut method with less DRM over appliance 2).

First, EMS performs DRM based on the time cut method,and FIGURE 6 shows that the total off-time of the GAPhouseholds is lower than those that subscribe to GSP. More-over, the ratio of the off-time, which is around 1 : 3, conformswith the proposed plans. The off-time is almost equal forhouseholds that adpot the same plan. Similarly, the off-time ofeach appliance is almost equal in the household. By contrast,EMS performs DRM based on the power cut method, and

VOLUME 3, 2015 2437


TABLE 3. MSE of total power load w/wo DRM during 9-18 o’clock.

FIGURE 7 shows that the total cut power of the GAPhouseholds is lower than those that subscribe to GSP. More-over, the ratio of the cut power, which is around 1 : 3,conforms with the proposed plans. The amount of cutpower is almost equal for households that adpot the sameplan. Similarly, the amount of cut power of each appli-ance is almost equal in the household. Given that theload of lighting appliance 2 is less than that of light-ing appliance 1, the total off-time of lighting appliance2 is higher than that of lighting appliance 1 to meet thecriteria.

Next, we consider the users’ preferences for appliancesand assume that all users prefer less DRM for appliance 2.We then set γ1,j = 1 and γ2,j = 2, ∀j with bothmethods. FIGURE 8 shows the results of DRM in termsof the time cut method. The total off-time of the house-holds is similar to that in FIGURE 6, but the total off-time of appliance 1 is around twice that of appliance 2.Similarly, FIGURE 9 show the results of DRM in termsof the power cut method. The total of the cut power ofthe households is similar to FIGURE 7, but the total cutpower of appliance 1 is around twice that of appliance 2.FIGURE 8-9 demonstrate that the proposed methods canallow the user preferences.

To demonstrate the performance of the proposed methods,we introduce the mean square error (MSE) as a performancemetric, which is defined as

MSE1 = E{‖PU − Ppeak‖22

}, (11)

MSE2 = E{‖〈PU − Ppeak〉+‖22

}. (12)

In Table 3, MSE1, MSE1 is the result of MSE from9:00 to 18:00 hours of one month, and MSE2 is the resultof MSE, which measures only those situations wheneverthe total power load exceeds the threshold Ppeak from9:00 to 18:00 hours of one month. For the case of the totalload without DRM, MSE1 and MSE2 are equal. This is dueto the total load without DRM, which always exceeds thethreshold Ppeak from 9:00 to 18:00 hours. Moreover, theMSEof the total load without DRM is far more than the resultsfor all the cases of the total load with DRM. This shows theeffectiveness of the proposed DRM scheme. A different alsoexists between the simulation and testbed results because ofthe delay in the testbed, where the smart meter is sampledat intervals of 10 seconds. Thus, a delay occurs when theEMS receives the data, performs optimization, and sends thecontrol. However, no such delay occurs in the simulation.In addition, the performance of the power cut method isslightly better than that of the time cut method.

VI. CONCLUSIONSIn this paper, we propose two types of customer engage-ment plan, namely, GSP and GAP, which describe two userinconvenience indices of participation with DRM and theincentive. Furthermore, we develop an appropriate DRMalgo-rithm with two methods, in terms of user inconvenienceindices, to facilitate EMS performing DRM on peak loadreduction. The objective of the proposed algorithm is todetermine the demand states of all the appliances of allhouseholds subject to the total demand load below a giventhreshold.

We build a testbed to verify the proposed scheme andcompare its performance versus its simulation. The results areuseful to evaluate the gap between actual implementation andtheoretical study, which is an important step to produce a bet-ter design that is more suitable for real-world implementation.In addition, the proposed algorithm can also be extended ina decentralized fashion, such that EMS assigns each user anindividual threshold according to his subscription plan, andusers can individually determine the demand states of theirappliances.

REFERENCES[1] M. Erol-Kantarci and H. T. Mouftah, ‘‘Wireless sensor networks for cost-

efficient residential energy management in the smart grid,’’ IEEE Trans.Smart Grid, vol. 2, no. 2, pp. 314–325, Jun. 2011.

[2] M. Rastegar andM. Fotuhi-Firuzabad, ‘‘Outage management in residentialdemand response programs,’’ IEEE Trans. Smart Grid, vol. 6, no. 3,pp. 1453–1462, May 2015.

[3] N. Ul Hassan, Y. I. Khalid, C. Yuen, and W. Tushar, ‘‘Customer engage-ment plans for peak load reduction in residential smart grids,’’ IEEE Trans.Smart Grid, vol. 6, no. 6, pp. 3029–3041, Nov. 2015.

[4] L. Chuan andA. Ukil, ‘‘Modeling and validation of electrical load profilingin residential buildings in Singapore,’’ IEEE Trans. Power Syst., vol. 30,no. 5, pp. 2800–2809, Sep. 2015.

[5] N. Forouzandehmehr, M. Esmalifalak, H. Mohsenian-Rad, and Z. Han,‘‘Autonomous demand response using stochastic differential games,’’IEEE Trans. Smart Grid, vol. 6, no. 1, pp. 291–300, Jan. 2015.

[6] A. Albert and R. Rajagopal, ‘‘Thermal profiling of residential energy use,’’IEEE Trans. Power Syst., vol. 30, no. 2, pp. 602–611, Mar. 2015.

[7] Y. Liu, C. Yuen, R. Yu, Y. Zhang, and S. Xie, ‘‘Queuing-based energyconsumption management for heterogeneous residential demands in smartgrid,’’ IEEE Trans. Smart Grid, Jun. 2015.

[8] X. Wang, C. Yuen, X. Chen, N. Ul Hassan, and Y. Ouyang, ‘‘Cost-awaredemand scheduling for delay tolerant applications,’’ J. Netw. Comput.Appl., vol. 53, pp. 173–182, Jul. 2015.

[9] M. T. Bina andD. Ahmadi, ‘‘Stochastic modeling for the next day domesticdemand response applications,’’ IEEE Trans. Power Syst., vol. 30, no. 6,pp. 2880–2893, Nov. 2015.

[10] M. Ahmadi, J. M. Rosenberger, W.-J. Lee, and A. Kulvanitchaiyanunt,‘‘Optimizing load control in a collaborative residential microgrid environ-ment,’’ IEEE Trans. Smart Grid, vol. 6, no. 3, pp. 1196–1207, May 2015.

[11] A. Anvari-Moghaddam, H. Monsef, and A. Rahimi-Kian, ‘‘Optimal smarthome energy management considering energy saving and a comfort-able lifestyle,’’ IEEE Trans. Smart Grid, vol. 6, no. 1, pp. 324–332,Jan. 2015.

2438 VOLUME 3, 2015


[12] Y. Liu, C. Yuen, S. Huang, N. Ul Hassan, X. Wang, and S. Xie, ‘‘Peak-to-average ratio constrained demand-side management with consumer’spreference in residential smart grid,’’ IEEE J. Sel. Topics Signal Process.,vol. 8, no. 6, pp. 1084–1097, Dec. 2014.

[13] C. Chen, J. Wang, and S. Kishore, ‘‘A distributed direct load controlapproach for large-scale residential demand response,’’ IEEE Trans. PowerSyst., vol. 29, no. 5, pp. 2219–2228, Sep. 2014.

[14] A. Safdarian, M. Fotuhi-Firuzabad, and M. Lehtonen, ‘‘A distributedalgorithm for managing residential demand response in smart grids,’’IEEE Trans. Ind. Informat., vol. 10, no. 4, pp. 2385–2393,Nov. 2014.

[15] C. Bartusch and K. Alvehag, ‘‘Further exploring the potential of residen-tial demand response programs in electricity distribution,’’ Appl. Energy,vol. 125, pp. 39–59, Jul. 2014.

[16] C. Vivekananthan, Y. Mishra, G. Ledwich, and F. Li, ‘‘Demand responsefor residential appliances via customer reward scheme,’’ IEEE Trans.Smart Grid, vol. 5, no. 2, pp. 809–820, Mar. 2014.

[17] X. Fang, S. Misra, G. Xue, and D. Yang, ‘‘Smart grid—The new andimproved power grid: A survey,’’ IEEE Commun. Surveys Tuts., vol. 14,no. 4, pp. 944–980, Oct. 2012.

[18] P. Siano, ‘‘Demand response and smart grids—A survey,’’ Renew. Sustain.Energy Rev., vol. 30, pp. 461–478, Feb. 2014.

[19] J. Aghaei and M.-I. Alizadeh, ‘‘Demand response in smart electricitygrids equipped with renewable energy sources: A review,’’ Renew. Sustain.Energy Rev., vol. 18, pp. 64–72, Feb. 2013.

[20] J. S. Vardakas, N. Zorba, and C. V. Verikoukis, ‘‘A survey on demandresponse programs in smart grids: Pricing methods and optimizationalgorithms,’’ IEEE Commun. Surveys Tuts., vol. 17, no. 1, pp. 152–178,Mar. 2015.

[21] Y. Zhang, S. He, and J. Chen, ‘‘Data gathering optimization by dynamicsensing and routing in rechargeable sensor networks,’’ IEEE/ACM Trans.Netw., Jun. 2015.

[22] I. Al-Anbagi, M. Erol-Kantarci, and H. T. Mouftah, ‘‘Delay critical smartgrid applications and adaptive QoS provisioning,’’ IEEE Access, vol. 3,no. 1, pp. 1367–1378, Aug. 2015.

[23] M. Erol-Kantarci and H. T. Mouftah, ‘‘Energy-efficient information andcommunication infrastructures in the smart grid: A survey on interac-tions and open issues,’’ IEEE Commun. Surveys Tuts., vol. 17, no. 1,pp. 179–197, Mar. 2015.

[24] R. C. Qiu et al., ‘‘Cognitive radio network for the smart grid: Experimentalsystem architecture, control algorithms, security, and microgrid testbed,’’IEEE Trans. Smart Grid, vol. 2, no. 4, pp. 724–740, Dec. 2011.

[25] V. Salehi, A. Mohamed, A. Mazloomzadeh, and O. A. Mohammed,‘‘Laboratory-based smart power system, part I: Design and systemdevelopment,’’ IEEE Trans. Smart Grid, vol. 3, no. 3, pp. 1394–1404,Sep. 2012.

[26] W.-Z. Song, D. De, S. Tan, S. K. Das, and L. Tong, ‘‘A wireless smartgrid testbed in lab,’’ IEEE Wireless Commun., vol. 19, no. 3, pp. 58–64,Jun. 2012.

[27] C. Wang et al., ‘‘A highly integrated and reconfigurable microgridtestbed with hybrid distributed energy sources,’’ IEEE Trans. Smart Grid,Oct. 2014.

[28] F. Adinolfi, G. M. Burt, P. Crolla, F. D’Agostino, M. Saviozzi, andF. Silvestro, ‘‘Distributed energy resources management in a low-voltagetest facility,’’ IEEE Trans. Ind. Electron., vol. 65, no. 4, pp. 2593–2603,Apr. 2015.

[29] W. Tushar, C. Yuen, K. Li, K. L. Wood, W. Zhang, and X. Liu, ‘‘Design ofcloud-connected IoT system for smart buildings on energy management,’’EAI Trans. Ind. Netw. Intell. Syst., Oct. 2015.

[30] C.-K. Wen, J.-C. Chen, J.-H. Teng, and P. Ting, ‘‘Decentralized plug-inelectric vehicle charging selection algorithm in power systems,’’ IEEETrans. Smart Grid, vol. 3, no. 4, pp. 1779–1789, Dec. 2012.

[31] M. Grant and S. Boyd. (Mar. 2014). CVX: MATLAB Software forDisciplined Convex Programming, Version 2.1. [Online]. Available:http://cvxr.com/cvx

[32] P. Saint-Andre, ‘‘Streaming XML with Jabber/XMPP,’’ IEEE InternetComput., vol. 9, no. 5, pp. 82–89, Sep./Oct. 2005.

[33] H. Flores and S. Srirama, ‘‘Mobile cloud messaging supported byXMPP primitives,’’ in Proc. 4th ACM Workshop Mobile Cloud Comput.Services (MCS), Taipei, Taiwan, Jun. 2013, pp. 17–24.

[34] M. Kirsche and R. Klauck, ‘‘Unify to bridge gaps: Bringing XMPP intothe Internet of Things,’’ in Proc. IEEE Int. Conf. Pervasive Comput.Commun. Workshops (PERCOM Workshops), Lugano, Switzerland,Mar. 2012, pp. 455–458.

WEN-TAI LI (S’12) received the B.S. andM.S. degrees in optoelectronics and communica-tion engineering fromNational Kaohsiung NormalUniversity, Kaohsiung, Taiwan, in 2009 and 2011,respectively. He is currently pursuing the Ph.D.degree with the Institute of Communications Engi-neering, National Sun Yat-sen University and isa research assistant with the Singapore Universityof Technology and Design, SUTD, Singapore. Hisresearch interests lie in power system monitoring,

the smart grid, and wireless communications.

CHAU YUEN (SM’12) received the B.Eng. andPh.D. degrees from Nanyang Technological Uni-versity, Singapore, in 2000 and 2004, respec-tively. He was a Post-Doctoral Fellow with LucentTechnologies Bell Laboratories, Murray Hill,in 2005. He was a Visiting Assistant Professorwith Hong Kong Polytechnic University in 2008.From 2006 to 2010, he was with the Institutefor Infocomm Research, Singapore, as a SeniorResearch Engineer. He has been with the

Singapore University of Technology and Design as an Assistant Professorsince 2010.

He received the IEEE Asia-Pacific Outstanding Young Researcher Awardin 2012. He serves as an Associate Editor of the IEEE TRANSACTIONS ON

VEHICULAR TECHNOLOGY.

NAVEED UL HASSAN (M’08–SM’15) receivedthe B.E. degree in avionics engineering from theCollege of Aeronautical Engineering, Risalpur,Pakistan, in 2002, and the M.S. and Ph.D. degreesin electrical engineering, with a specializationin digital and wireless communications, fromthe Ecole Superieure d’Electricite, Gif-sur-Yvette,France, in 2006 and 2010, respectively. In 2011,he joined as an Assistant Professor with theDepartment of Electrical Engineering, Lahore

University of Management Sciences, Lahore, Pakistan. Since 2012, he hasbeen a Visiting Assistant Professor with the Singapore University of Tech-nology and Design, Singapore. He has several years of research experience.He has authored/co-authored numerous research papers in refereed inter-national journals and conference proceedings. His major research interestsinclude cross-layer design and radio resource optimization in wireless net-works, demand responsemanagement in smart grids, indoor localization, andheterogeneous networks.

WAYES TUSHAR (S’06–M’13) received theB.Sc. degree in electrical and electronic engineer-ing from the Bangladesh University of Engineer-ing and Technology, Bangladesh, in 2007, andthe Ph.D. degree in engineering from AustralianNational University, Australia, in 2013. He was aVisiting Researcher with National ICT Australia,ACT, Australia. He was also a Visiting StudentResearch Collaborator with the School of Engi-neering and Applied Science, Princeton Univer-

sity, NJ, USA, in summer 2011. He is currently a Research Scientist withthe Singapore University of Technology and Design (SUTD)–MIT Interna-tional Design Center, SUTD, Singapore. His research interests include signalprocessing for distributed networks, game theory, and energy managementfor smart grids. He was a recipient of two best paper awards, both as the firstauthor, in the Australian Communications Theory Workshop 2012 and theIEEE International Conference on Communications 2013.

VOLUME 3, 2015 2439


CHAO-KAI WEN (M’04) received thePh.D. degree from the Institute of Communica-tions Engineering, National Tsing Hua University,Hsinchu, Taiwan, in 2004. He was with the Indus-trial Technology Research Institute, Hsinchu, andMediaTek Inc., Hsinchu, from 2004 to 2009. He iscurrently an Associate Professor with the Insti-tute of Communications Engineering, NationalSun Yat-sen University, Kaohsiung, Taiwan.His research interests center around the optimiza-

tion in wireless multimedia networks.

KRISTIN L. WOOD received the B.Sc. degreein engineering science from Colorado State Uni-versity, Fort Collins, CO, USA, in 1985, and theM.Sc. and Ph.D. degrees in mechanical engineer-ing from the California Institute of Technology,Pasadena, CA, USA, in 1986 and 1989, respec-tively. He joined as a Faculty Member with TheUniversity of Texas at Austin in 1989, after com-pleting his Ph.D. work, and established a compu-tational and experimental laboratory for research

on engineering design and manufacturing, in addition to a teaching labora-tory for prototyping, reverse engineering measurements, and testing. From1997 to 1998, he was a Distinguished Visiting Professor with the UnitedStates Air Force Academy (USAFA), where he worked with the USAFAFaculty to create design curricula and research within the EngineeringMechanics/Mechanical EngineeringDepartment. In 2011, hewas a Professorwith the Mechanical Engineering, Design and Manufacturing Division, TheUniversity of Texas at Austin. He was a National Science Foundation YoungInvestigator, the Cullen Trust for Higher Education Endowed Professor inEngineering, a University Distinguished Teaching Professor, and the Direc-tor of the Manufacturing and Design Laboratory and MORPH Laboratory.

KUN HU received the B.S. degree in computer sci-ence and technology from Jilin University, in 2012,and the M.S. degree in software engineering fromPeking University, in 2015. He was a VisitingStudent Researcher with the Singapore Universityof Technology and Design in 2014. His researchinterests lie in Internet of Things, pervasive com-puting, and smart home.

XIANG LIU received the B.Eng. degree in elec-trical engineering from Hunan University, PRCin 1984, the Ms.Eng. degree in pattern recogni-tion and intelligent control from the Institute ofAutomation, the Academy of Sciences of China,in 1987, and the Ph.D. degree in automation fromBordeaux University 1, France, in 1993.

Starting 1994, Dr. Liu joined the Global Soft-ware Group (GSG), Motorola. In 2001, He hadbeen elected as the Member of Motorola Science

Advisory Board Associates (SABA) for his technical contribution. StartingMarch 2003, Dr. Liu joined the School of Software and Microelectron-ics, Peking University, as a Professor and the Chair of the Departmentof Embedded System Engineering. His current research interests includehigh performance embedded computing, ubiquitous computing, and softwareengineering.

2440 VOLUME 3, 2015

demand response management for residential-ieee journals pdf

Documents