1504

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 19, NO. 3, JULY 2004

A Power Quality Index Based on Equipment Sensitivity, Cost, and Network Vulnerability
Geun-Joon Lee, Senior Member, IEEE, Mihaela M. Albu, Member, IEEE, and Gerald Thomas Heydt, Fellow, IEEE
AbstractThis paper focuses on voltage sag phenomena and their impact on customer satisfaction. In order to derive a unique power quality of service index, information from both the supply network (according to standards in use) and the customer (defined in terms of load sensitivity and interruption cost) are merged. The well-known CBEMA power acceptability curve is revisited with definitions from the IEEE Standard 1159 superimposed. This depiction suggests a way to assign a cost or index of power quality events. Although individual cost assignments used in the calculation are subject to question, once the assignments are made, the calculation is consistent and gives a useful measure of quality of service. The method is illustrated factoring severity of the events based on the CBEMA curve and IEEE 1159 classified power quality events. The paper addresses the elusive issue of the cost of power quality disruption. Index TermsCBEMA power acceptability curve, cost of power quality, equipment sensitivity, power quality, voltage sag.

I. POWER ACCEPTABILITY CURVES

HE CONSIDERABLE interest in momentary low voltage events (i.e., sags or dips) generally focuses on the depths of the sag and its duration. The voltage sag characteristics depend on the network design and load type, which are deterministic in nature. The actual occurrence of low voltage events, however, is stochastic in nature. Although voltage sags can be analyzed for a given network, sag occurrence can only be predicted from past event data, which unfortunately may not yield a detailed description of the events. Considering the new infrastructures in which power systems operate, a bridge between the two approaches (deterministic and probabilistic) may improve power quality assessment [1]. Most of the power system events are classified according to appropriate standards, e.g., IEEE 1159. A wide range of analytical techniques has been applied to the general topic of voltage sags (deemed by many to be the salient problem of power quality today) including concepts derived from fuzzy logic theory [2]. The customers load can also be modeled, but load diversity and time-dependent operation make this approach impracticable.
Manuscript received February 14, 2003. This work was supported in part by the Electrical Engineering and Science Research Institute of Korea and in part by the Power Systems Engineering Research Center (PSerc) at Arizona State University, Tempe, AZ. G.-J. Lee is with the Chungbuk Provincial University, Chungbuk, Korea (e-mail: gjlee@ctech.ac.kr). M. M. Albu is with the Department of Electrical Engineering, Politehnica University of Bucharest, Bucharest, Romania (e-mail: albu@electro.masuri.pub.ro). G. T. Heydt is with the Department of Electrical Engineering, Arizona State University, Tempe, AZ 85259-5706 USA (e-mail: heydt@asu.edu). Digital Object Identifier 10.1109/TPWRD.2004.829124

Instead, equipment sensitivity to voltage sag can be considered and presented in the form of power acceptability curves (also known as voltage vulnerability or sensitivity curves). One of these curves is the Computer Business Equipment Manufacturers Association (CBEMA) curve. Although CBEMA is now defunct, and the Information Technology Industry Council (ITIC) curve has taken its place, the CBEMA curve is probably the best known of the power acceptability curves. The CBEMA curve is shown in Fig. 1. The vertical scale shows the bus voltage amplitude deviation from rated value, and the horizontal scale shows the duration of a disturbance event as seen at the load point. The vertical scale in Fig. 1 is shown in percent, although this may be rendered in rms values of the voltage. The horizontal scale in Fig. 1 is logarithmic in seconds, but this scale may be shown linearly and/or in cycles. Short disturbances correspond to the left lower corner of the CBEMA plane, and near normal operating conditions are shown close line. The more recent ITIC curve is to the horizontal shown in Fig. 2. In Figs. 1 and 2, the acceptable operating range is shown as the region included between the upper (overvoltage) limb and the lower (undervoltage) limb. Although the overvoltage region of power acceptability curves is important, the high visibility of the effects of momentary low voltage events gives particular interest to voltage sags and this is the focus of attention in this paper. It is believed that the CBEMA curve was originally developed by operating experience with mainframe computers. Kyei, Ayyanar, Heydt, Thallam and Blevins [3] recently noted that if one uses a DC voltage standard to define acceptable and unacceptable operation, one obtains nearly the CBEMA curve. The term DC voltage standard refers to the ultimate criterion for acceptability as whether the DC output voltage of a rectifier load remains above 87% of rated value [3]. If this criterion is applied to a common single-phase bridge rectifier (or a three phase Graetz bridge operating with a balanced supply) with a DC circuit filter having two poles (i.e., an LC filter), one finds that the DC output voltage is, (1) As an example, let the voltage standard be as excursion becomes unacceptable indicated above. Then the in (1). Solution for in terms of at when in this expression gives

0885-8977/04$20.00 2004 IEEE

LEE et al.: A POWER QUALITY INDEX BASED ON EQUIPMENT SENSITIVITY, COST, AND NETWORK VULNERABILITY

1505

Fig. 1. The CBEMA power acceptability curve. Fig. 2. The ITIC power acceptability curve.

The graphical representation of the above formula is similar to in per the undervoltage limb of the CBEMA curve (with unit, in seconds). It is evident that different types of loads and different criteria result in different power acceptability. For example, rotating loads may have a speed standard, which prescribes acceptable operation of the speed drops below a certain value. References [3] and [4] discuss this case. The vulnerability of adjustable speed drives is thoroughly discussed in [5]. Interestingly, when the speed standard is applied, the equipment sensitivity curve that is obtained has a very similar shape to the CBEMA curve. Kyei also discusses the force standard for AC magnetic relays [6]. Again, the equipment sensitivity curve results to be similar to the CBEMA curve. True load diversity is a complex conglomeration of many load types, each with their own load dynamics. The superposition of sensitivity curves for each type yields something like that in Fig. 3. The interior region labeled in the figure is the region of power supply acceptability. Again, experimental studies with actual data reveal that the interior region is very similar to the CBEMA curve depicted in Fig. 1. The foregoing discussion does not address several practicalities such as the point on the wave at which the low voltage event initiates, the nonrectangular shape of the sag variation in time and various three-phase details. The point-on-wave is discussed by Collins, Morgan, Mansoor and Koval [7][9] in experimental terms. From a worst case point of view, the point on wave that produces the greatest disturbance may be used [10], [11]. Three phase complications (e.g., the type of three-phase event that causes sag) can be analyzed in terms of the effect of unbalanced three phase events on the supply voltage to a rectifier load with two pole dynamics [6]. II. IEEE STANDARD 1159 The IEEE Standard 1159 [12] gives the categories and typical characteristics of power system events. Table I shows the terminology of IEEE 1159 and the classification of events by

Fig. 3.

Superposition of several equipment sensitivity curves.

duration of event. The actual standard represents some of these duration times in cycles. Table II shows the typical severity of events represented as voltage magnitudes. European norms require voltage sags (dips) [13] to be entered in a spreadsheet organized as depth of sag versus duration. The European norm uses sag depths [10, 30), [30, 60) and [60, 100) percent; and durations [10, 100 ms), [100, 500 ms), [500, 1000 ms), [1, 3 s), [3, 20 s), and [20, 60 s). Fig. 4 shows the time scale of various voltage disturbance events using the terminology of IEEE 1159. It is possible to integrate the usual CBEMA curve (or alternative CBEMA-like curves, for specific load types) with IEEE 1159 definitions of voltage sag. The same procedure can be applied for a nondifferentiate representation of load sensitivity to voltage sags, as in [10], but, when discrimination among disturbances are available only in terms of IEEE 1159, the method proposed here becomes easier to handle. This is illustrated using the conventional CBEMA curve in Fig. 5. Long duration undervoltages are depicted at the far right in Fig. 5 as sustained events (e.g., steady state). Instantaneous, momentary, and temporary sags are shown below the CBEMA curve in the appropriate time interval. Three IEEE Standards,

1506

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 19, NO. 3, JULY 2004

TABLE I DURATION OF SHORT-TERM VOLTAGE VARIATIONS USING IEEE 1159 (DURATIONS SHOWN IN MILLISECONDS FOR A 60 HZ SYSTEM)

TABLE II TYPICAL VOLTAGE MAGNITUDE FOR SHORT DURATION VARIATIONS OF LOW AND HIGH VOLTAGE EVENTS IN POWER DISTRIBUTION SYSTEMS (IN PER UNIT)

Fig. 4. Categories of voltage disturbance events according to event duration using the terminology of IEEE 1159.

Fig. 5.

IEEE 1159 sag durations integrated with the CBEMA curve.

namely IEEE 1159, IEEE 1250, and IEEE 859 are indicated in Fig. 5 with their approximate range of scope in duration of disturbances [12][15]. III. POWER QUALITY INDICES AND THE COST OF POWER QUALITY Engineers often resort to indices to quantify complex phenomena. In some cases, this strategy is useful in reducing the complexity at little sacrifice of information; in other cases, indices may not capture the desired information, and they may obfuscate critical information [16]. Power quality is often quantified by the use of event-count indices such as the System Average Interruption Frequency Index (SAIFI), System Average Interruption Duration Index (SAIDI), and the System Average RMS (Variation) Frequency Index Voltage Threshold (SARFI). For sustained interruptions the SAIFI index is: SAIFI (Total number of interruptions)/(Total number of points of delivery monitored). For disturbances in general: SAIDI (Total duration of all interruptions)/(Total number of points of delivery monitored). (Summation of the number of cusSARFI V for tomers experiencing rms variation rms for Total number of customers . For the above indices, only temporary interruptions are to be counted [13]. In order to quantify voltage sags, subsets [17] of SARFI V are used, defining indices according to IEEE 1159:

(Summation of the number of customers experiencing rms for instantarms for neous variation Total number of customers . SMARFI V (Summation of the number of cusfor tomers experiencing rms momentary, i.e. 30 cycles to 3 seconds, rms for variation Total number of customers . STARFI V (Summation of the number of cusfor tomers experiencing rms temporary, i.e. 3 to 60 seconds, varirms for ation Total number of customers . Event-count indices have the advantage of ease in instrumentation and convenient comparison. Some electric utility companies set power quality targets on the basis of event-count indices. However, event-count power quality indices can not be translated easily (or accurately) into loss of load data [18]. The most elusive of questions, namely the cost of power quality, is poorly depicted by event-count indices. It seems that tailored CBEMA-like curves, generated for specific load types and specific load dynamics, have a potential of capturing load response and load survival. Cost of events, also, can be estimated albeit the estimates of cost are only as accurate as the data used for the cost of a typical power quality disturbance. At this point, attention turns to a new power quality index regarding voltage sag and load characteristics. The objective is to capture cost and severity information rather than a straight count

SIARFI

LEE et al.: A POWER QUALITY INDEX BASED ON EQUIPMENT SENSITIVITY, COST, AND NETWORK VULNERABILITY

1507

of events as used in the indices above. Further, it is undesirable to require additional instrumentation, and the new index should entail only data processing. IV. VOLTAGE DISTURBANCE EVALUATION Previous indices are useful to represent the interruption frequency and duration. However, it is difficult to enumerate the effects of voltage sags that have the potential of disrupting voltage sensitive loads. This paper suggests a power quality index to quantify the supply voltage quality of each duration class based on the impact of that class of disturbance. A series of measured voltage disturbances can be represented as points on CBEMA curve as in Fig. 5. These points are grouped according to a common disturbance duration class. This philosophy appears to be important in the design of load characteristics and protection devices. If there are groups of disturbance durations and each points of voltage disturbance records, the energy group has impact of each voltage disturbance is quantified as follows: the energy impacts of voltage disturbance is quantified as for all where (the event sequence is classified in is group where the classes are defined in IEEE 1159) and the sag duration under the assumptions made in [10]. The mean , value of each energy disturbance for class-duration is

Fig. 6. Interrupted load area by sag and interruption events at a composite load bus an example.

inverters and rotating machines. Each region in Fig. 6 represents a voltage sag class area in which load interruption might occur. If a voltage disturbance point is in region , for example, electronic devices, converters and motors are dropped. With this information, it is possible to produce the customer average load drop frequency. VI. EXAMPLE OF A POWER QUALITY INDEX CALCULATION It is useful to examine a power quality index based on cost and severity of disturbances. For this purpose, two new indices are studied the average load drop index and the average load drop cost. As an illustration, consider a customer that has the information of total power and load composition window as in Table III. Historical voltage records during a specified period plotted on a CBEMA curve as in Table IV (where the areas are depicted in Fig. 6) are indicated with the number of events falling in the specified areas. Note the special column in Table V corresponding to the area labeled ac in Fig. 6. In Fig. 6 there are voltage sag and interruption classes labeled on the are termed vertical axis: the events falling beneath interruptions, and these are tabulated in a special column in Table IV. These data are archived data obtained from dedicated event recorders. The equivalent load drop index for each time , class is termed the bus average load drop index,

Note that there are groups of disturbance durations. The quanrepresents the average impact of the voltage sag to the tity connected load taking into account the frequency of the sag and and can be interruption at each duration. The energy taken to be a power quality index for each duration and could be used as a basis for power quality conditions in contracts between utilities and customers. However, the impacts of voltage disturbances to loads are different depending on the vulnerability (sensitivity) of those loads to voltage sags. To calculate the load trip effect by the disturbances (i.e., the potential to lose the load because of a low voltage condition), a CBEMA-like curve for the given loads is useful. References [3], [4], [6] tell how to generate equipment sensitivity curves for several load types (i.e., customized sensitivity curves). V. VULNERABILITY OF LOADS TO VOLTAGE SAGS For a bus that has several types of loads, it is possible to classify loads according to their sensitivity to voltage sag and duration. At least in theory, the CBEMA curve of each type of classified load can be drawn. A more practical approach is to classify loads in broad groups assembling the results statistically. With the voltage disturbance data and load sensitivity curves, it is possible to produce load trip frequencies (number of trips/time) for each duration class. The process is illustrated in Fig. 6, which shows a representative example of CBEMA curves classified as three kinds of loads: electronic devices,

where the notation refers to the number of voltage disturbance records in the duration class . The individual load drop index for each class can be expressed as,

(2)

1508

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 19, NO. 3, JULY 2004

TABLE III EXAMPLE LOAD COMPOSITION. TOTAL BUS LOAD P = 100 kW

TABLE V EQUIVALENT LOAD DROP COUNTS IN THE ILLUSTRATIVE EXAMPLE

TABLE IV VOLTAGE SAG

RECORDS

IN

FIG. 6:

TABLE VI AVERAGE COST OF POWER INTERRUPTION

For further simplification, one can assume that the probability of load drop is unity in each unacceptable area, and (2) becomes, (3) represents the load distribution and represents where the number of events falling into the area defined by the time ) and the sensitivity curve of the load class ( type . Note that is the number of load types (e.g., in Table III, ). Applying data from Tables III and IV in (3), the LDI is produced as in Table V. These load drop event counts (loss of load counts) can be utilized as a power quality index, which represents the load drop rate on the customer side. The LDI is useful to quantify the power quality cost at a given bus. Note in Table V that the LDI is effectively an equivalent number of event counts based on a severity weighting of each event. In this way, an equivalent number of load drops is counted. In Table V, is listed as a class index. With the LDI of each IEEE 1159 class, it is possible to estimate the cost of power quality for each customer. A study of interruption costs is presented in [18][20]. Representative costs are listed in Table VI. It is possible to use data appropriate for the region and consistent with the history of the region the data in Table VI are presented only as an illustration. Caution is advised in interpreting cost data because costs of interruptions are highly variable, load and application dependent, and regionally dependent. To the degree that statistical agglomeration relieves some of these variations, the cost data for a particular region and class might be used. Using data from Tables V and VI the average Load Drop Cost (LDC) can be estimated, (4) is an index reflecting the average cost of interruption where in a given IEEE 1159 class. In the cited example, the LDC is calculated as 337.2

VII. EXTENSION TO PRACTICAL POWER SYSTEM CALCULATIONS In power systems, customers are often classified as residential, commercial, small industrial and large industrial customers. Consider a distribution bus which has customer composition data and total bus power given in Table VII. Then a typical load composition chart, as described in Table IV, can be assigned to each customer. With these data and weighted load sensitivity curves, it is possible to compute the load drop power quality indices and power quality costs per bus. Fig. 7 describes the algorithm to obtain the LDI and LDC. Below is the summary of the calculation method: 1. Develop the sensitivity curve for each load type. 2. Construct a weighted CBEMA curve. 3. Subdivide the time axis of the CBEMA curve as Instantaneous, Momentary, Temporary, Sustained failures using the IEEE 1159 definitions. 4. Process the estimation of voltage sag events in the PCC considered for each load into the corresponding CBEMA curve and find the effective region which corresponds to the loss of the load. 5. Estimate the LDI (the percentages of lost loads for each load model or type) and then multiply it by the corresponding interruption cost indices for evaluating LDC. VIII. CONCLUSIONS This study suggests the idea of power quality indices LDI and LDC based on equipment sensitivity curves, voltage estimation, and specific interruption costs. The new indices which can assess the effects of the voltage sag and interruption to a specified bus load as load drop index (LDI) and load drop cost (LDC). These indices can be calculated using customer composition data, load information, the sensitivity to voltage sag events of each load type and historical-derived cost data. The method does not entail new measurements, and it appears easy to implement when information on past events are recorded and classified in terms of IEEE 1159. The advantage over simple event counts is that cost and interruption severity is captured. These indices can

LEE et al.: A POWER QUALITY INDEX BASED ON EQUIPMENT SENSITIVITY, COST, AND NETWORK VULNERABILITY

1509

Fig. 7. Conceptual diagram of power quality index and cost calculation based on CBEMA-like curves and equivalent interruption costs.

TABLE VII CUSTOMER COMPOSITION DATA NOTATION

[4] J. Kyei, G. T. Heydt, R. Ayyanar, J. Blevins, and R. Thallam, Power acceptability curves for rotating machine loads, in Proc. North American Power Symposium, 2001, pp. 120123. [5] M. Bollen and L. Zhang, Analysis of voltage tolerance of ac adjustable-speed drives for three-phase balanced and unbalanced sags, IEEE Trans. Ind. Applicat., vol. 36, pp. 904910, May/June 2000. [6] J. G. Kyei, Analysis and Design of Power Acceptability Curves for Industrial Loads, MSEE, Arizona State Univ., Dept. Electrical Eng., Tempe, AZ, 2001. [7] E. Collins and R. Morgan, A three phase sag generator for testing industrial equipment, IEEE Trans. Power Delivery, vol. 11, pp. 526532, Jan. 1996. [8] D. Koval, Computer performance degradation due to their susceptibility to power supply disturbances, in Proc. IEEE Ind. Applicat. Soc. Annu. Meeting, vol. 2, 1989, pp. 17541760. [9] E. Collins and A. Mansoor, Effects of voltage sags on AC motor drives, in Proc. IEEE Tech. Conf. on the Textile, Fiber and Film Industry, 1997, pp. 916. [10] IEEE Recommended Practice for the Design of Reliable Industrial and Commercial Power Systems, IEEE Std. 493-1997, Dec. 16, 1997. [11] M. Kezunovic and Y. Liao, A new method for classification and characterization of voltage sags, in Elect. Power Syst. Res.. New York: Elsevier, 2001, pp. 2735. [12] IEEE Recommended Practice for Monitoring Electric Power Quality, IEEE Std. 1159-1995, Nov. 1995. [13] A. Robert, Power quality monitoring at the interface between transmission system and users, in Proc. Harmonics and Quality of Power Conf., vol. 2, 2000, pp. 425430. [14] IEEE Guide for Service to Equipment Sensitive to Momentary Voltage Disturbances, IEEE Std. 1250-1995, June 1995. [15] IEEE Standard Terms for Reporting and Analyzing Outage Occurrences and Outage States of Electrical Transmission Facilities, IEEE Std. 8591987, February 1988. [16] G. Heydt and W. Jewell, Pitfalls of power quality indices, IEEE Trans. Power Delivery, vol. 13, pp. 570578, Apr. 1998. [17] D. L. Brooks, R. C. Dugan, M. Waclawiak, and A. Sundaram, Indices for assessing utility distribution system RMS variation performance, IEEE Trans. Power Delivery, vol. 13, pp. 254259, Jan. 1998. [18] K. Koellner, SRP voltage index methods and findings, in Proc. North American Power Symposium, 2002, pp. 239246. [19] G. Tollefson, R. Billinton, G. Wacker, E. Chan, and J. Aweya, A Canadian customer survey to assess power system reliability worth, IEEE Trans. Power Syst., vol. 9, pp. 443450, Feb. 1994. [20] M. J. Sullivan and T. Vardell, Interruption cost, customer satisfaction and expectations for service reliability, IEEE Trans. Power Syst., vol. 11, pp. 989995, May 1996.

give cost information for power system operation for planning and potential rebate policy. The concept of LDI and LDC are offered as power quality indices that capture cost and load vulnerability. The calculation methods shown in the paper are representative of what can be done: no firm conclusion is drawn on the value of the concept without actual implementation. It appears, however, that the approach is in the direction of properly correlating a power quality index with cost and load vulnerability.

Geun-Joon Lee (SM01) received the M.S. and Ph.D. degrees in electrical engineering from Seoul National University, Seoul, Korea, and Ulsan University, Ulsan, Korea. Currently, he is a Professor of Electrical Engineering at the Chungbuk Provincial University, Chungbuk, Korea. He has experience in the Power System Department of the Korea Electric Power Research Institute and the Energy System Research Center at the University of Texas at Arlington. His research interests include power system stability, voltage and reactive power control, FACTS, and power quality.

REFERENCES
[1] R. Allan and R. Billinton, Probabilistic assessment of power systems, Proc. IEEE, vol. 88, pp. 140162, Feb. 2000. [2] J. M. de Carvalho Filho, J. P. G. de Abreu, H. Arango, and J. C. C. Noronha, Analysis of power system performance under voltage sags, Elect. Power Syst. Res., no. 55, pp. 211218, 2000. [3] J. Kyei, R. Ayyanar, G. Heydt, R. Thallam, and J. Blevins, The design of power acceptability curves, IEEE Trans. Power Delivery, vol. 17, pp. 828833, July 2002.

Mihaela M. Albu (M96) received the Ph.D. degree from the Universitatea Politehnica Bucurestiat, Bucharest, Romania. Currently, she is a Professor of Electrical Engineering at the Universitatea Politehnica Bucurestiat. Her research interests include power quality, instrumentation, and remote experimentation. Dr. Albu spent a leave at Arizona State University, Tempe, as a Fulbright Fellow in 2002-03.

1510

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 19, NO. 3, JULY 2004

Gerald Thomas Heydt (M64SM80F91) received the Ph.D. degree in electrical engineering from Purdue University, West Lafayette, IN. Currently, he is the Director of a power engineering center program at Arizona State University, Tempe, where he is a Regents Professor. His industrial experience is with the Commonwealth Edison Company, Chicago, IL, and E. G. & G., Mercury, NV. Dr. Heydt is a member of the National Academy of Engineering.