IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 27, NO. 4 ...apic/papers/Power_Quality/2013-Assessing...

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 27, NO. 4, OCTOBER 2012 1937

Assessing the Collective Harmonic Impact of ModernResidential Loads—Part I: Methodology

Diogo Salles, Student Member, IEEE, Chen Jiang, Student Member, IEEE, Wilsun Xu, Fellow, IEEE,Walmir Freitas, Member, IEEE, and Hooman Erfanian Mazin, Student Member, IEEE

Abstract—The proliferation of power-electronic-based residen-tial loads has resulted in significant harmonic distortion in thevoltages and currents of residential distribution systems. There isan urgent need for techniques that can determine the collectiveharmonic impact of these modern residential loads. These tech-niques can be used, for example, to predict the harmonic effectsof mass adoption of compact fluorescent lights. In response to theneed, this paper proposes a bottom-up, probabilistic harmonicassessment technique for residential feeders. The method modelsthe random harmonic injections of residential loads by simulatingtheir random operating states. This is performed by determiningthe switching-on probability of a residential load based on theload research results. The result is a randomly varying harmonicequivalent circuit representing a residential house. By combiningmultiple residential houses supplied with a service transformer,a probabilistic model for service transformers is also derived.Measurement results have confirmed the validity of the proposedtechnique. The proposed model is ideally suited for studying theconsequences of consumer behavior or regulatory policy changes.

Index Terms—Harmonic analysis, residential loads, statisticalanalysis, time-varying harmonics.

I. INTRODUCTION

T HE proliferation of power-electronic-based modern resi-dential loads has resulted in significant harmonic distor-

tions in the voltages and currents of residential power distri-bution systems. These new harmonic sources have comparablesizes and are distributed all over a network. Although they pro-duce insignificant amount harmonic currents individually, thecollective effect of a large number of such loads can be substan-tial [1]–[3]. At present, there is an urgent need for techniquesthat can determine the collective harmonic impact of modernresidential loads. Such techniques can be used, for example, topredict the harmonic effects of mass adoption of compact fluo-rescent lights (CFLs) and to quantify the effectiveness of certain

Manuscript received August 26, 2011; revised March 24, 2012; acceptedMay 19, 2012. Date of publication August 14, 2012; date of current ver-sion September 19, 2012. This work was supported in part by the NationalSciences and Engineering Research Council of Canada, in part by the Al-berta Power Industry Consortium, and in part by FAPESP, Brazil. Paper no.TPWRD-00718-2011.

D. Salles and W. Freitas are with the Department of Electrical EnergySystems, University of Campinas, Campinas 13083-852, Brazil (e-mail:[email protected]; [email protected]).

C. Jiang, W. Xu, and H. E. Mazin are with the Department of Electrical andComputer Engineering, University of Alberta, Edmonton, AB T6G 2V4 Canada(e-mail: [email protected]; [email protected]; [email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TPWRD.2012.2207132

harmonic control measures. One such measure is to adopt theIEC device level limits in North America [4].

The main challenge to developing the above-mentioned tech-niques is how to model the random nature of the harmonic cur-rents produced by residential loads. Over the past many years,some researchers have investigated the summation of randomharmonic phasors [5], [6], the algorithms of stochastic harmonicpower flows [7], [8], and methods to predict the mean valuesof harmonic indices [9]. All of these works have greatly con-tributed to our understanding on the modeling and analysis ofsystems with randomly varying harmonic loads. Unfortunately,the available techniques are still not in a shape to fulfill the needsof predicting harmonic distortions caused by consumer behavioror regulatory policy changes.

The objective of this paper is to present a systematic, versa-tile technique based on Monte Carlo simulation to study the har-monic impact of residential loads. The main idea is originatedfrom the following observation: the random harmonic genera-tion of residential loads is almost exclusively due to the randomon/off states of the loads. For example, a CFL can be in an ON

or OFF state randomly at any given time. But once it is turnedon, its harmonic currents essentially follow a known, determin-istic spectrum. Therefore, the key to develop the aforementionedharmonic assessment technique is to model the random ON/OFF

state change events of residential loads properly. Once the statesof all residential loads are known, the problem becomes a de-terministic harmonic power flow problem. Fortunately, a bodyof knowledge on residential load behaviors has been developedfor load research purposes [10]–[13]. These techniques can beadapted to solve the problem of predicting the operating statesof residential loads. The result is a bottom-up-based harmonicanalysis technique ideally suited for studying the consequencesof consumer behavior or regulatory policy changes.

The aforementioned idea has been applied to create a simula-tion technique for predicting the harmonic impacts in secondarydistribution systems. Such a system typically consists of 5–20single-detached houses in North America. An aggregate modelfor service transformers is also derived with the technique. Themodel can help to determine the harmonic conditions in primarydistribution systems. Field measurement results taken from adozen of service transformers in Canada have validated the pro-posed modeling technique.

The paper is organized as follows. Section II describesthe electrical models of typical residential loads. Section IIIpresents a procedure to determine the ON/OFF state of a res-idential load based on the behaviors of inhabitants and theusage characteristics of the residential load. Build on the above

0885-8977/$31.00 © 2012 IEEE

1938 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 27, NO. 4, OCTOBER 2012

results, the harmonic model of a single house is developed inSection IV. Section V derives a service transformer model.Verification results are also presented. Section VI summarizesthe conclusions.

II. MODELING OF RESIDENTIAL LOADS

There are two aspects to model a residential load. One is itselectrical model and the other is its operating state model. Thissection is focused on the electrical model.

Residential loads can be divided into linear and nonlinearloads. In this paper, linear loads are modeled as constant powerloads at the fundamental frequency (60 Hz) and as impedanceat harmonic frequencies (h) [14]. The model parameters can beobtained from the measurements or derived from the load’s elec-tric characteristics. The nonlinear residential loads are modeledas constant power loads at the fundamental frequency (60 Hz)and as current sources at harmonic frequencies [14]. The mag-nitude and phase of the source is calculated from the harmoniccurrent spectrum of the residential load using the well-knownprocedure recommended in [14] as follows.

1) The harmonic-producing load is treated as a constantpower load at the fundamental frequency, and the funda-mental frequency power flow of the system is solved.

2) The current injected from the load to the system is thencalculated and is denoted as .

3) The magnitude and phase angle of the harmonic currentsource representing the load are determined as follows:

--

(1)

- - (2)

where the subscript “spectrum” stands for the typical harmoniccurrent spectrum of the residential load. In this study, harmonicspectra of more than 20 types of residential loads (which addsup to about 100 individual loads) have been measured.

The measurement activities have identified a need to deter-mine which residential loads are linear or nonlinear. In a dis-torted supply voltage scenario, even a resistive load could havea distorted current waveform, appearing as a nonlinear source.In this paper, the correlation of the instantaneous current andvoltage waveforms ( – plot) are used to separate linear loadsfrom nonlinear loads. The method is illustrated in Fig. 1 for tworesidential loads. If the load is linear, the - plot is shaped ei-ther as a straight line (resistive load) or as a ring (reactive load).The voltage versus current plots for nonlinear residential loadsare neither a straight line nor a ring, for example, the microwaveoven in Fig. 1.

When adopting the current source model for nonlinear loads,the issue of whether to include the attenuation and diversity ef-fects has been examined. Attenuation refers to the reduction ofthe magnitude of harmonic currents produced by a residentialload when its supply voltage is distorted [15]. Based on the pub-lished research works [8], [14], [15] and extensive lab tests con-ducted by the authors [16], we concluded that the attenuationeffect becomes significant only when the voltage distortion isquite high (for example, THD above 10%). We have not ob-served such a high-voltage distortion level in the field nor found

Fig. 1. Correlation between instantaneous voltage and current of a load. (a)� –� plot of the dryer (linear). (b) � –� plot of microwave (nonlinear).

Fig. 2. Current waveform measured from CFLs of different vendors.

from simulation results. So the attenuation effect is omitted fromthe model proposed at present. This will lead to slightly higherharmonic levels in the system. If the need to include the effectarises, iterative harmonic power-flow algorithms can be used[14], [17].

The diversity effect refers to the differences of harmonic cur-rent phase angles associated with a residential load [14], [15]. Itis common knowledge that a residential load always producesharmonic currents at the same phase angle with respect to itssupply voltage (if given in the same operating condition). Thisis why typical spectrum can be used to model a harmonic-pro-ducing load. Fig. 2 shows the waveforms of multiple CFLs fromdifferent vendors. It can be seen that the waveforms are similar,meaning the harmonic angle (and magnitude) differences aresmall. As a result, the random variation of phase angles fromtheir mean values is not included in the electrical model. Thevariation due to different brands can be included in the MonteCarlo sampling explained later. It is important to note that thevariation of phase angles caused by the variation of the phase an-gles of supply voltages is included through (2). The harmoniccancellation effect caused by voltage phase differences at dif-ferent points of a feeder because of the branch impedances isalso modeled in the simulation through (2) and the multiphasenetwork model.

III. MODELING OF RESIDENTIAL HOUSES

This section develops a probabilistic bottom-up model thatprovides the time distribution of the ON/OFF state of linear andnonlinear residential loads during the course of the day. It isuseful to note that each load may switch ON/OFF randomly. Butthe switching activities of all types of loads follow some statis-tical distributions. These distributions are used to determine theON/OFF states of the loads. From this perspective, the loads donot operate in random completely.

SALLES et al.: ASSESSING COLLECTIVE HARMONIC IMPACT OF MODERN RESIDENTIAL LOADS—PART I 1939

Fig. 3. Time-of-use probability profile for cooking activity.

Fig. 4. Time- of-use probability profile for laundry activity.

A. Residential Load Profile Modeling

Considerable research efforts have been made in the past onthe subject of developing detailed residential electrical load pro-files (i.e., reconstructing the expected daily electrical loads of ahousehold based on residential load sets, occupancy patterns,and statistical data). Based on the published works [10]–[13],the basic idea of home load profile modeling is to address threefactors:

1) When a residential load will be turned on. This can besimulated using the Daily Time of Use Probability Profilesof the residential loads.

2) How long a residential load will stay on (i.e., working cycleduration). The duration can be determined from measureddata and from understanding the purpose of the residentialloads involved.

3) How to include the impact of the habits and population ofa household.

A load profile is created by combining these factors. Eachfactor is presented in detail in the following subsections.

1) Daily Time of Use Probability Profiles: In [10], electricload profiles are constructed from individual residential loads topredict the tendency of the occupants to switch on a residentialload at any given time. Since then, many researchers [11]–[13]have investigated how to quantify the probability of the speci-fied activity being undertaken as a function of time of day, whichis called the Time of Use Probability Profiles. It represents theprobability of a household performing a specific activity duringa 24-h period. In this paper, each activity profile data is mainly

TABLE IUSAGE PATTERN FOR MAJOR RESIDENTIAL LOADS

TABLE IIAVERAGE USAGE PATTERN FOR OTHER RESIDENTIAL LOADS

TABLE IIIOCCUPANCY PATTERN FOR A TYPICAL HOUSEHOLD

collected from [12] and [18], which uses the survey data pro-vided by [19]. For the sake of space, the probability profile of afew activities is discussed and illustrated as follows.

Most of the kitchen loads are related to cooking activities.They share the same probability of switching on at a giventime of the day. As shown in Fig. 3, the curves are applied topredict the occupants’ cooking-related actions and to establishthe probability of a load switch-on event, such as using mi-crowave, blender or griddle, etc. The probability profile is givenwith 1-min resolution. The higher the value in the profile, thehigher the probability that the load switches on. For example, amicrowave switch-on event would be far more likely to occurat 17:00 than at 4:00. The profile on weekends is flatter thanthe profile for weekdays because people tend to cook at morerandom times on weekends.

In addition, people are more likely to use a toaster and waffleiron in the morning, and stove at noon and in the evening. Spe-cific profiles for these residential loads are also considered inthe proposed method.

Daily time of use of washer machines is determined by thelaundry activity profile shown in Fig. 4. The time-of-use pro-file for dryers is generally the same shape as the washer profile,and is offset from the washer profile in time [13]. People usu-ally turn on dryers between 10 and 30 min following the endof the washer cycle. The time-of-use probability profile associ-ated with other activities is also included in the methodology,


including TVs, computers, lighting, house cleaning, and occa-sional switch-on events (garage door and furnace).

2) Residential Load Duration Characteristics: For majorresidential loads, the cycle duration can be established ac-cording to the measurement data from the Canadian Center forHousing Technology (CCHT) [20]. At the CCHT, a simulatedoccupancy system triggers daily residential load ON/OFF statechange events in a real single detached home. The averagecycles per year is derived from standard residential load testmethods of the Canadian Standards Association [21], [22].Details of some major residential loads and working cycledurations are presented in Table I.

Information regarding other residential loads was extractedfrom buyers’ guides [23], [24] and from field measurements.Several other residential loads were surveyed and a partial listof the expected hours that each residential load remains onper month is presented in Table II. The data in the table belowonly include the total working hours per month. The number ofswitch-on events per day is determined as follows: the averageworking hours for an electric kettle is 15 per month per Table II.If the average working cycle duration of a kettle is 3 min, one canobtain 15 60 30 3 10 switch-on events per day.

3) Size and Occupancy Pattern of the Household: The size ofhousehold has a significant impact on daily electricity demand.In order to include the impact of different household sizes, ahousehold size factor is introduced. When modeling a residen-tial house with a specified number of occupants, is equal tothe ratio between and the average number of people per house-hold (assumed to be 2.5 in this paper based on [25]). The valueof average hours per month for each residential load providedin Table II cannot be used directly for different household sizessince, for example, a house with more people will lead to an in-crease in load usage. In order to take this into account, the usagetimes provided in Table II will be multiplied by the householdsize factor .

An occupancy pattern (i.e., when occupants of a residence areat home, and using residential loads) affects the ON/OFF stateof the loads. The common factors influencing the occupancypattern are as follows [12]: (a) the time of the first person gettingup in the morning and the last person to go to sleep and (b) theperiod of inactive house occupancy during working hours. Dueto a lack of information about the house occupancy pattern, thefive most typical scenarios of the household occupancy patternin Canada are proposed. Table III lists these possible scenarios.

4) Probabilistic Model of Residential Load Switching-On:Based on the usage pattern of residential loads discussed in theprevious sections and the method of [12], a procedure to deter-mine load switch-on events at a given time of a day (which is alsocalled a simulation step) has been established. The procedure is aform of Monte Carlo simulation and implemented in a computerprogram. At each simulation step, a list of residential loads thatare on is generated, which forms the load profile at that step. Thesimulation procedure is shown Fig. 5 and explained as follows.Step 1) The time-of-use probability profile is selected ac-

cording to the chosen load and whether it is aweekend or not; the occupancy pattern is also se-lected according to the working type and whether itis a weekend or not.

Step 2) When the simulation starts, the probability of loadactivation (Pr) can be read from its activity profile.

Step 3) The number of residential load switch-on events (m)are modified to consider household size

, where is the household size factor.Step 4) The probability ( ) of a residential load to switch

on at the present instant in time ( ) is equal to theprevious probability Pr multiplied by the modifiednumber of load switch-ons in the simulation timeperiod and a calibration scalar ( ),

. A discussion of how the calibration scalaris derived is presented below.

Step 5) The calculated probability is compared to a nor-mally distributed random number ( ) between 0 and1. If is larger than , go to Step 6); otherwise, goto Step 7).

Step 6) The residential load is switched on and the presentsimulation time step is updated to , where

is the load working cycle duration. After that, goback to Step 2).

Step 7) The load remains off and the present simulation timestep is updated to . is the simulation res-olution (e.g., 1 min). After that, go back to Step 2).

The calibration scalar (c) is introduced to reflect the influenceof the household occupancy pattern [12]. This paper uses the oc-cupancy function, as follows, to represent the occupancy pattern

when house is actively occupied(e.g., morning, evening)when house is inactively occupied(e.g., daytime, midnight)

where “inactively occupied” refers to the scenario where no-body is at home or awake.

If , the calibration scalar c is made 0 for mostresidential loads, which makes the probability of certain loadswitch-on to be zero when nobody is at home or awake.

If , the calibration scalar c is introduced in orderto make the mean probability of an activity taking place, whenmultiplied by the calibration scalar, equal to the mean proba-bility of a load switch-on event. As shown in Fig. 6, before cal-ibration, we have

After introducing the occupancy function and calibrationscalar c

(n) = 0(n)= 1.


Fig. 5. Procedure to determine switch-on events of residential loads.

Some residential loads, such as fridge, freezer, and furnace,etc. are independent of household occupancy, so the calibrationscalar is always made equal to 1.

The simulation of microwave usage is used as an example toillustrate the aforementioned procedure Fig. 7. The simulatedhousehold is set to have an average size and is full-time worktype. The day of interest is weekday. The simulation resolutionis 1 min, which means 1440 steps in one day. As can beseen from Fig. 7, the time of the first person waking up is 05:40,the inactive occupancy period during work is 7:28–16:58, andthe time of the last person going to sleep is 22:36 for this ran-

domly generated instance. If the simulation runs to the first step(00:01), the value of occupancy function is 0, and calibrationscalar 0, the resulting probability of microwave switch-on( ) at this time step is 0, which means the microwave has nochance to switch on. However, if the time step is equal to 1200(20:00), the value of occupancy function is 1, and the calibra-tion scalar is


Fig. 6. Time-of-use probability profile calibration with occupancy function.

Fig. 7. Time-of-use probability profile for a microwave.

The probability of microwave use at this step becomes0.0552%. According to the residential load usage characteristic,the average working cycle duration and the average workinghours per month of a microwave are equal to 4 min and 10 h(Table II). So the number of switch-on events per day of themicrowave is equal to . Hence,

0.0552%0.64% which means the probability of microwave switch-on atthe current time step is 0.64%. When the simulation finishesthe 1440 steps (i.e., covering the 24-h period, results like thoseshown in Fig. 7 will be obtained. For this day of simulation, themicrowave is used five times, two usages are at around 17:00,


Fig. 8. Simulated usage time for different residential loads.

and 2 at around 20:00. The usage duration for one time is 4–5min, and total usage time is 21 min/day.

The same procedure is conducted for each installed residen-tial load at each simulation step. An example use pattern of someloads as simulated by the procedure is shown in Fig. 8. One canobserve that the PC is used between 17:00 and 24:00 and thewasher and dryer are not used throughout the day. Televisionsand PCs are used for a relatively long period throughout the day,while cooking loads are used for much shorter periods but mul-tiple times.

At the bottom of Fig. 8, the refrigerator and freezer are acti-vated periodically throughout the entire day and not associatedwith household occupancy pattern.

IV. SINGLE-HOUSE HARMONIC LOAD MODEL

Once the residential load usage time information is derived, aload with ON state will be represented with its electrical modeland be connected to the electric circuit of a house. Note thatthere could be different electrical models of the same load ifone wants to consider different brands or consumer trends. Aparticular model is obtained by drawing randomly from a data-base of residential loads.

In North America, residential customers are usually suppliedthrough three-wire single-phase distribution transformers. Thesecondary of these transformers has a neutral and two hot phasescarrying 120 V with respect to the neutral. In a residence, resi-dential loads are connected in an essentially indeterminate waywith respect to the circuits, and this arrangement makes it diffi-cult to electrically model a residence. Since the objective of theproposed technique is to predict the harmonic impact on powersystems, an equivalent circuit can be developed for a house.Based on the theoretical analysis presented in [26], the modelof Fig. 9 has been established.

In each simulation step, the house impedance andcurrent source are established randomly based on theprocedure described earlier. Impedance representsthe linear loads in a house, and current source representsthe nonlinear loads. An example simulation output is shownin Fig. 10, for only the fundamental, third, and fifth harmoniccomponents.

The results of seven days simulation, which contains 5 week-days and 2 weekend days, are shown in Table IV. The table lists

Fig. 9. Equivalent circuit model to represent a residential house.

Fig. 10. Simulation output � �� of a house during one day. (a) Funda-mental component. (b) Third harmonic component. (c) Fifth harmonic compo-nent.

the mean value and standard deviation of the total house funda-mental, 3rd, and 5th harmonic currents for each day.


TABLE IVSIMULATION RESULTS OF A RESIDENTIAL HOUSE ��

TABLE VMEASUREMENTS RESULTS OF A RESIDENTIAL HOUSE ��

In order to validate the simulation results, field measurementswere conducted for typical residential houses and the results areshown in Table V. The measured sample house has some occu-pants that do not need to go to work, so “no working”-type oc-cupancy pattern is chosen for simulation. Comparing Table IVagainst Table V, the mean values of the fundamental and eachharmonic current matched quite well.

V. SERVICE TRANSFORMER HARMONIC LOAD MODEL

Residential houses are supplied through single-phase ser-vice transformers connecting the primary to the secondarysystem. The secondary is a 120/240-V three-wire service.Each distribution transformer normally supplies 10–20 houses.The loads are modeled collectively as one load connected tothe secondary side of the service transformer (the “servicetransformer model”). This model is needed for studying theharmonic impact on primary distribution systems.

The steps for constructing such a model are as follows.1) Generate harmonic models of 10–20 houses according to

the proposed approach presented in Section IV.2) Connect the house models as shown in Fig. 11(a).3) Build the equivalent circuit as shown in Fig. 11(b).The aforementioned service transformer model has been ver-

ified by comparing its results against those from field measure-ments. Field measurements of harmonic currents in a sampleservice transformer are shown in Fig. 12 (only the 1st and 3rdharmonic components are illustrated). These data were collectedfrom ten different transformers serving residential loads in Ed-monton, AB, Canada, in Winter 2008. There is a total of 55 daysmeasurement data. The current variation of each transformer is

Fig. 11. Equivalent service transformer circuit model. (a) Service transformercircuit model. (b) Equivalent model.

Fig. 12. Example of the transformer current output during one weekday ob-tained from real field measurements. (a) Fundamental component. (b) Third har-monic component.

TABLE VIPERCENTAGE OF VARIANCE OF THE FIRST PRINCIPAL

COMPONENT OF THE TRANSFORMER CURRENT

unique, so it is not easy to compare the results with those ob-tained from simulation which also exhibit random characteris-tics. The approach implemented in this paper is to extract andcompare the principal components of the current profiles.

The principal component analysis (PCA) is a mathematicalprocedure that transforms a number of possibly correlatedvariables of the original data into a smaller number of uncor-related variables called principal components. Mathematically,PCA is a linear transformation that converts the data to anew coordinate system so that the greatest variance by anyprojection of the data comes to lie on the first coordinate, thesecond greatest variance on the second coordinate, and so on.This way, one can choose not to use all of the components andstill capture the most important part of the data. More detailscan be found in [27].

Table VI lists the variance given by the first principal com-ponents for the magnitudes of measured transformers. Data are


Fig. 13. Comparison of the first principal components (weekdays).

grouped into weekday and weekend. As can be seen, for thefundamental and third harmonic current, first principal compo-nents can represent almost 60% of the original data; however,for higher harmonics, these percentages drop to 20%–30%. Thereason is that higher harmonics are more difficult to predict anddo not seem to show similar trends.

Based on the aforementioned analysis, two methods are pro-posed for the transformer model verification as follows.

1) For fundamental and third harmonic current, the verifica-tion method is to extract the first principal componentsfrom the measurement data and the simulation results, re-spectively. The components are then correlated to verifytheir consistency.

2) For higher order harmonic currents, the verification is tocompare the normalized probability distribution of themeasured and calculated data.

Fig. 13 shows the daily variation of the first components ofthe fundamental and third harmonic currents from both mea-surement and simulation on weekdays. The first component ofthe simulated fundamental current fits quite well with that ofthe measured one. The correlation factor is 0.94. There is anacceptable difference between the first components of the thirdharmonic current from simulation and measurements. The cor-relation factor is 0.7.

The remaining part of the fundamental and 3rd harmoniccomponents contains almost 40% of the original data. Acomparison is made in the form of statistical distributions inFig. 14. The distributions exhibit similar characteristics. Forhigher order harmonics, their random variation is too strongfor the principal components to yield meaningful results. Sothe components are not used for verification. Instead, normal-ized harmonic distributions are used for comparison. Fig. 15shows the results for a weekday. Table VII shows the standarddeviation of the harmonics for both measurements (Meas.)and simulation (Sim.) results. The results show some formof consistency for weekdays and weekends. Based on thisverification analysis, it can be concluded that the proposedprobabilistic residential house model is accurate and can beincluded in the modeling of distribution systems for harmonicimpact assessment.

Fig. 14. Probability distribution of measured and simulated residue part for aweekday.

Fig. 15. Probability distribution function (PDF) curves of higher harmonics.

TABLE VIISTANDARD DEVIATION OF HIGHER ORDER HARMONIC CURRENTS

VI. CONCLUSION

A probabilistic method to determine the harmonic impact ofresidential loads and houses has been presented in this paper.The method models the random harmonic generations of resi-dential loads by simulating the random operating states of theloads. This is done through determining the switching-on prob-ability of a residential load based on the load research results.


The result is a randomly varying harmonic equivalent circuitrepresenting a residential house. By combining multiple resi-dential houses served by a service transformer, a model for ser-vice transformers is also derived. PCA and statistical analysison the simulated and measurement results have confirmed thevalidity of the proposed modeling approach.

One of the attractive characteristics of the proposed method isits bottom-up approach. As a result, one can simulate the effectof market trends and policy changes. For example, the harmonicimpact of CFLs can be studied by adjusting the composition oflighting fixtures in the residential load database. An exampleapplication of the proposed method, assessing the harmonic im-pact of residential loads on secondary distribution systems, isdescribed in a companion paper.

REFERENCES

[1] A. E. Emanuel, J. Janczak, D. J. Pileggi, E. M. Gulachenski, C. E. Root,M. Breen, and T. J. Gentile, “Voltage distortion in distribution feederswith nonlinear loads,” IEEE Trans. Power Del., vol. 9, no. 1, pp. 79–87,Jan. 1994.

[2] N. R. Watson, T. L. Scott, and S. Hirsch, “Implications for distribu-tion networks of high penetration of compact fluorescent lamps,” IEEETrans. Power Del., vol. 24, no. 3, pp. 1521–1528, Jul. 2009.

[3] R. Dwyer et al., “Evaluation of harmonic impacts from compact fluo-rescent lights on distribution systems,” IEEE Trans. Power Syst., vol.10, no. 4, pp. 1772–1779, Nov. 1995.

[4] Electromagnetic Compatibility (EMC)—Part 3–2: Limits—Limits forHarmonic Current Emissions (Equipment Input Current � �� A perphase), IEC Int. Standard 61000–3-2, 2005.

[5] “Probabilistic aspects task force of the harmonics working group, time-varying harmonics: Part II— harmonic summation and propagation,”IEEE Trans. Power Del., vol. 17, no. 1, pp. 279–285, Jan. 2002.

[6] M. Lehtonen, “A general solution to the harmonics summationproblem,” Eur. Trans. Elect. Power Eng., vol. 3, no. 4, pp. 293–297,Jul./Aug. 1993.

[7] A. Cavallini and G. C. Montanari, “A deterministic/stochastic frame-work for power system harmonics modeling,” IEEE Trans. Power Syst.,vol. 12, no. 1, pp. 407–415, Feb. 1997.

[8] S. R. Kaprielian, A. E. Emanuel, R. V. Dwyer, and H. Mehta, “Pre-diction voltage distortion in a system with multiple random harmonicsources,” IEEE Trans. Power Del., vol. 9, no. 3, pp. 1632–1638, Jul.1994.

[9] G. Zhang and W. Xu, “Estimating harmonic distortion levels for sys-tems with random-varying distributed harmonic-producing loads,” Inst.Eng. Technol. Gen., Transm. Distrib., vol. 2, no. 6, pp. 847–855, Nov.2008.

[10] C. Walker and J. Pokoski, “Residential load shape modelling based oncustomer behavior,” IEEE Trans. Power App. Syst., vol. PAS-104, no.7, pp. 1703–1711, Jul. 1985.

[11] A. Capasso, W. Grattieri, R. Lamedica, and A. Prudenzi, “A bottom-upapproach to residential load modeling,” IEEE Trans. Power Syst., vol.9, no. 2, pp. 957–964, May 1994.

[12] I. Richardson, M. Thomson, D. Infield, and C. Clifford, “Domesticelectricity use: A high-resolution energy demand model,” EnergyBuildings, vol. 42, no. 10, pp. 1878–1887, 2010.

[13] M. Armstrong, M. Swinton, H. Ribberink, I. Beausoleil-Morrison, andJ. Millette, “Synthetically derived profiles for representing occupantdriven electric loads in canadian housing,” Building Perform. Simul.,vol. 2, pp. 15–30, 2009.

[14] “Task force on harmonics modeling and simulation, modeling andsimulation of the propagation of harmonics in electric power net-works–part I: concepts, models, and simulation techniques,” IEEETrans. Power Del., vol. 11, no. 1, pp. 452–465, Jan. 1996.

[15] A. Mansoor, W. M. Grady, A. H. Chowdhury, and M. J. Samotyj, “Aninvestigation of harmonics attenuation and diversity among distributedsingle-phase power electronic loads,” IEEE Trans. Power Del., vol. 10,no. 1, pp. 467–473, Jan. 1995.

[16] A. B. Nassif and W. Xu, “Characterizing the harmonic attenuation ef-fect of compact fluorescent lamps,” IEEE Trans. Power Del., vol. 24,no. 3, pp. 1748–1749, Jul. 2009.

[17] B. C. Smith, N. R. Watson, A. R. Wood, and J. Arrillaga, “A newtonsolution for the harmonic phasor analysis of AC/DC converters,” IEEETrans. Power Del., vol. 11, no. 2, pp. 965–97, Apr. 1996.

[18] R. Hendron, Building America Research Benchmark Definition,Golden, CO, Tech. Rep. NREL/TP-550-40968, 2007.

[19] IPSOS-RSL and Office for National Statistics, “United Kingdom TimeUse Survey, 2000” (computer file), 3rd ed. Colchester, Essex, U.K.,U.K. Data Archive (distributor), 2003.

[20] M. Bell, M. Swinton, E. Entchev, J. Gusdorf, W. Kalbfleisch, R.Marchand, and F. Szadkowski, “Development of micro combined heatand power technology assessment capability at the Canadian centre forhousing technology,” Ottawa, ON, Canada, Tech. Rep. B-6010, 2003.

[21] CSA, Energy Consumption Test Methods for Household Dishwashers,CSA Standard CAN/CSA-C373–92, 1992.

[22] CSA, Energy Performance, Water Consumption and Capacity ofAutomatic Household Clothes Washers, CSA Standard CAN/CSA-C360–98, 1998.

[23] “Natural Resources Canada, Photovoltaic Systems—A Buyer’sGuide,” 2002.

[24] “Natural Resources Canada, Micro-Hydropower Systems—A Buyer’sGuide,” 2004.

[25] Statistics Canada, Dwelling Characteristics and Household Equipment.Tech. rep. 62F0041XDB, 2010. [Online]. Available: http://www5.statcan.gc.cabsolc/olc-cel/olc-cel?catno=62F0041X & lang=eng

[26] H. E. Mazin, E. E. Nino, W. Xu, and J. Yong, “A study on the harmoniccontributions of residential loads,” IEEE Trans. Power Del., vol. 26, no.3, pp. 1592–1599, Jul. 2011.

[27] I. T. Jolliffe, Principal Component Analysis. Berlin, Germany:Springer-Verlag, 1986, vol. I.

Diogo Salles (S’04) received the B.Sc. and M.Sc. degrees in electrical engi-neering from the University of Campinas, Campinas, Brazil, in 2006 and 2008,respectively, where he is currently pursuing the Ph.D. degree in electrical engi-neering.

From 2010 to 2011, he was a Visiting Doctoral Scholar at the University ofAlberta, Edmonton, AB, Canada. His research interests focus on power qualityand analysis of distribution systems.

Chen Jiang (S’09) received the B.Eng. degree in electric engineering and au-tomation from the Huazhong University of Science and Technology (HUST),Wuhan, China, in 2008, and is currently pursuing the M.Sc degree in electricalengineering at the University of Alberta, Edmonton, AB, Canada.

His main research interests are power quality and smart-grid technology ofdistribution systems.

Wilsun Xu (M’90–SM’95–F’05) received the Ph.D. degree in electrical engi-neering from the University of British Columbia, Vancouver, BC, Canada, in1989.

Currently, he is a Professor and an NSERC/iCORE Industrial Research Chairin Power Quality at the University of Alberta, Edmonton, AB, Canada. His cur-rent research interests are power quality, harmonics, and information extractionfrom power disturbances.

Walmir Freitas (M’02) received the Ph.D. degree in electrical engineering fromthe University of Campinas, Campinas, Brazil, in 2001.

Currently he is an Associate Professor at the University of Campinas. Hisareas of research interest are the analysis of distribution systems and distributedgeneration.

Hooman Erfanian Mazin (S’08) was born in Tehran, Iran, in 1981. He receivedthe B.Sc. and M.Sc. degrees in electrical engineering from the Amirkabir Uni-versity of Technology, Tehran, Iran, in 2004 and 2006, respectively, and is cur-rently pursuing the Ph.D. degree in electrical engineering at the University ofAlberta, Edmonton, AB, Canada.

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 27, NO. 4 ...apic/papers/Power_Quality/2013-Assessing...

Documents

Transcript of IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 27, NO. 4 ...apic/papers/Power_Quality/2013-Assessing...