The Optimal Selection of Mitigation Methods Against

The Optimal Selection of Mitigation Methods Against Voltage Dips and Interruptions: a Customer-Based Approach Dirk Van Hertem

Marcel Didden

Johan Driesen, Ronnie Belmans

Departement ESAT/ELECTA Katholieke Universiteit Leuven Kasteelpark 10, B-3001 Heverlee, Belgium e-mail: [email protected]

Electrical Power Systems and Metrology Laborelec Rodestraat 125, B-1630 Linkebeek, Belgium e-mail: [email protected]

Departement ESAT/ELECTA Katholieke Universiteit Leuven Kasteelpark 10, B-3001 Heverlee, Belgium

Abstract— Voltage dips and interruptions cause major economic damage. Not only do we have to consider potential loss of production, there is also loss of market, the loss of client thrust, comfort,. . . that make the energy consumers want to implement a certain form of protection for their systems. However, the abundance of available mitigation methods and the, in many cases, unknown interruption cost make the selection of the most cost-effective mitigation method very difficult. This optimal mitigation method cannot be calculated using traditional theoretical economic methods, and is not always obtained by the currently used selection methods in the industry. However, using an alternative selection method, an optimal mitigation method can be found. This paper describes the steps to be taken, to solve the selection problem by means of a practical example. The examined case is a 250 kVA installation with a high sensitivity to voltage dips (pc load), and without a full knowledge of the cost of a process interruption. The proposed method makes a cost-benefit analysis of all possible solutions resulting in an overview that gives the optimal solution for a certain interruption cost. This method makes it possible to clearly interpret the investment costs and to compare completely different mitigation methods, only using data and methods which are available to an industrial customer. As such, this paper is written from a customer point of view. Index Terms— Mitigation, voltage dips and interruptions, techno-economic assessment

I. I NTRODUCTION A. The cost of voltage dips and interruptions Every year, voltage dips and interruptions cause major economic damage. In the US alone the yearly economic damage is estimated to be between $ 104 billion and $ 164 billion [1]. At this moment there is an abundance of mitigation solutions available on the market, some are generally used and available (e.g. static UPS), others are only applicable for special cases (e.g. using a boost converter in the DC-bus of a variable-speed drive). All these methods reduce the effective number of process interruptions to a certain amount, and each of them has a certain cost. The difficulty arises in determining the optimal mitigation method in a techno-economic point of view. The optimal mitigation method is the one which results in the lowest total cost over the lifetime of the investment, but ICHQP Paper # 128

this is depending on many variables. The selection methods currently used by the industry differ in most cases from the economical sound methods that are available in literature [2], [3]. B. Conflict between theory and practice 1) Theory: The optimal selection method for the mitigation of voltage dips and interruptions can be determined using the classic theoretical methods [4]. The NPV (net present value) method (eq. 1) is widely accepted as the most appropriate method to calculate the profitability of investments [5].

N P V = −C0 + where:  C0     C  dev_int     fn · Cdev_int Coperating       r    ntot

n tot X

fn · Cdev_int − Coperating (1 + r)n n=0

(1)

investment cost; cost per process interruption; avoided economic damage, profit; yearly operating costs, including maintenance; capitalisation rate; lifetime of the investment.

However, for the selection of mitigation devices, this formula often cannot be used because of the uncertainty of the parameters, especially the interruption cost [6]. This interruption cost is dependent on many uncontrollable variables [7]: • • • • • • • •

point in time of the event (day – night, winter – summer,. . . ); economic situation; duration of the event; announced or unannounced event; availability of other energy sources; weather conditions; damage caused to the installation; ...

Therefore this method is rarely used in practice. c

Copyright IEEE 2004

2) Practice: The selection of a suitable mitigation method for a given problem in the industry is not based on certain rules, but case depending. The results from surveys [2], [3] show that there are several possible selection criteria used within the industry: • • • •

•

• •

the demand for a minimum availability of the voltage (e.g. six ‘nines’); appliances are divided in groups, and standard solutions for each group are provided; companies tend to be loyal to one brand (reduction in the operator faults); the surveyed companies depended on traditional, conservative choices for their mitigation methods (no DVR, STATCOM,. . . ); mitigation devices for computer protection were in many cases placed in a redundant configuration (N − 1), even without economic calculation; mitigation devices for computer protection were overdimensioned; large installations cannot be fully protected against voltage dips and interruptions. The main objective in this case is to limit damage to the equipment and limit the process interruption duration.

The selection of mitigation methods that are currently used, are mainly based on rules of thumb, and no economic optimization is used.

per year that a customer experiences1 λ ([#/yr]) and the average interruption duration or M T T R (mean time to repair, [min]). The reliability indices of a 15 kV connection in a highly meshed grid (such as can be found in Western Europe) is summarized in table I, depending on the grid connection: directly at a HV/MV substation, or in an open ring structure (see also fig. 1). TABLE I R ELIABILITY

INDICES AT THE CONNECTION POINT IN A HIGHLY MESHED GRID

1

2

2

2

[2]

Case Connection at HV/MV substation Open ring connection, good conditions Open ring connection, medium conditions Open ring connection, bad conditions Case good conditions medium conditions bad conditions

λ [#/yr] 0.1 0.4

M T T R [min] 60 60

0.9

80

1.8

100

#dips [#/yr] 5 14.8 30

HV HV/MV substation MV

II. D ESCRIPTION OF THE CASE A. Installation A typical installation that is considered to need protection against the consequences of voltage dips and interruptions is an ICT center within a large company. These installations have a great susceptibility to voltage dips and interruptions as described in the ITIC characteristic [8]. Also, the interruption cost is perceived as being very high. The rated power of these installations is typically between 40 and 400 kVA. The case here considered is such a center, with a rated power of 250 kVA. The company is connected to the public grid on an open ring structure in a strongly meshed grid (fig. 1). The voltage level at the connection point is 15 kV. The center has to be online 24/7, except for the planned maintenance once a year (during a company-wide holiday). The projected lifetime is 15 year. The capitalization rate is chosen to be 10 %. B. Grid connection 1) Voltage interruptions: The connection to the public grid has an important influence on the number of process interruptions that an installation endures, since it determines the amount of voltage interruptions and their duration. Also the number of voltage dips depends on the connection. The reliability data at the connection point to the public grid can be retrieved from literature [9], from measurements or from contacts with the local grid operator. The reliability indices that are used here are the average number of interruptions

1

2

MV/LV substation

Fig. 1. Simplified scheme of a typical MV distribution grid (protection devices are not shown)

2) Voltage dips: A large part of the process interruptions with sensitive equipment, is caused by voltage dips originating from faults on nearby branches of the HV grid. Estimated values of the yearly number of voltage dips can be obtained in the same way as the outage data. The average number of voltage dips that yearly can be expected at a 15 kV connection in a highly meshed grid (as commenly found in Western Europe) is summarized in table I. In general, the number of 1 Or its reciprocal, the mean time between failure (M T BF ≈ 8760/λ [h] when M T BF  M T T R )

voltage dips is situation dependent, but not directly linked to the number of voltage interruptions or the interruption duration (e.g. a good grid connection concerning voltage dips does not necessarily include a good grid connection for voltage interruptions). 3) Used values: The considered values for the worked out example are λ = 1 /yr, M T T R = 100 min and #dips = 16 /yr. Also the influence of each of these parameters is investigated.

GRID

incoming post 1.2 km cable MV substation

C. On-site reliability 1) Voltage interruptions caused by on-site events: In most cases, the sensitive equipment is not directly coupled to the utility grid, but is, together with other loads, supplied through a grid that is maintained by the customer himself. This local grid is in most cases radial and its influence can be calculated when the necessary reliability data is available. The techniques used are simple reliability calculations (e.g. stochastic reliability or Markov-Chain theory [10]), assuming exponential distribution. Using the Markov-Chain theory, the reliability indices for the series connection (λs and M T T Rs ) and parallel connection (λp and M T T Rp ) of two components (index 1 and 2) can be calculated as follows: 1 µ1 · µ2 · (λ1 + λ2 ) = (λ1 + µ1 ) · (λ2 + µ2 ) M T BFs µ1 · µ2 · (λ1 + λ2 ) 1 µs = = (λ1 + µ1 ) · (λ2 + µ2 ) − µ1 · µ2 M T T Rs λ1 · λ2 · (µ1 + µ2 ) 1 λp = = (λ1 + µ1 ) · (λ2 + µ2 ) M T BFp 1 µp = µ1 + µ2 = M T T Rp λs =

Incoming bus 1.2 km cable MV substation single transformer LV distribution

M T BF [h] 45156 218563 45156 398181 68640

LV distribution

PC load

Fig. 2.

sensitive load

Implemented on-site grid

(2c) (2d)

D. Total number of process interruptions

(2a) (2b)

TABLE II DATA OF THE CONSIDERED GRID ( BASED ON

2 transformers (redundant)

between 40 and 100 ms for faults on the medium voltage level. The depth of the dip is depending on the relative size of the impedances of the source and the fault carrying branch. It can be assumed that the dip depth for faults on the medium voltage level is relatively high (around 50 %). Another source of voltage dips is the startup current of heavy loads, especially induction motors. The influence of these is very site dependent, but can be accounted for if the number of startups per year of the heavy load is known. In this case the contribution of these events is negligible.

Where µ is the repair rate, which is the inverse of the M T T R. Using these four simple formulae, the reliability at the terminals of the sensitive installation can be determined. The needed reliability data of the grid components (such as transformers, switchgear,. . . ) is provided by the manufacturer or can be found in the literature (e.g. [11]). The local grid is schematically represented by fig. 2. In table II the used reliability data of the different parts of the considered network are given (based on [11], [12]). R ELIABILITY

MV grid

[11], [12])

M T T R [min] 626.4 3787.2 626.4 480 76.8

2) Voltage dips by on-site events: Short circuits on other branches (not feeding the sensitive load) within the company grid cause a voltage dip at the terminals of the sensitive load. The dip duration depends on the fault clearing time. This is less then 20 ms for most faults on the 400 V level, and

Not every voltage dip leads to a process interruption: only those who are outside the voltage tolerance characteristic cause damage. Therefore also the voltage dip distribution is an important parameter. However, for an industrial customer, it is not easy to obtain this distribution through measurement, since a large amount of occurred dips is needed to derive this, resulting in long measurement time or a large amount of measurement points. Therefore this data can be obtained from the literature [3] or from contact with the local grid operator [13]. Figure 3 shows an example of a cumulative voltage dip distribution, which is used for the calculations in this paper. The data are obtained from measurement on 17 substations, over a combined period of 29.92 years, resulting in the measurement of 461 voltage dips. As voltage tolerance characteristic the ITIC-curve is used [8], since it is applicable for most IT equipment. When the combination of the voltage dip distribution of fig. 3 and the voltage tolerance according to ITIC is made, it shows that in this case only 31.3 % of the voltage dips, coming from the grid, cause a process interruption. The faults on the local grid are only harmful if they originate from the medium voltage level. It can be assumed that all these dips are harmful for a load with at voltage characteristic such as the ITIC. These on-site voltage dips can be calculated with

cumulative voltage distributiion

1 0.9 0.8 0.7 0.6 0.5 0.4

remaining voltage (%)

0.3 0.2

100

0.1 0 0

80 60 500

40 1000

1500

20 2000

2500

3000

0

dip duration (ms)

Fig. 3. Average cumulative voltage dip distribution of a medium voltage substation in a meshed grid [3].

the reliability data of the grid, and will add about 20 % to the number of harmful voltage dips.

4) 300 kVA static UPS, autonomy 180 min; 5) 2 × 300 kVA static UPS, N − 1 redundant, autonomy 15 min (each); 6) 250 kVA flywheel UPS, autonomy 17.2 s; 7) 215 kVA DySC [14]; 8) 250 kVA Vectek Omnivolt AVC [15]; 9) 346 kVA (500 A) SVR [16]. 1) Fixed costs: The fixed costs (C0 in (1)) mainly consist of the mitigation device cost and the cost to install the device (labor, area, time,. . . ). 2) Operating costs: The operating costs or variable costs are the costs which allow the mitigation device to work. These operating costs consist of heating losses, maintenance costs and additional costs such as the replacement of the batteries at the end of their life, the air conditioner to cool the battery room,. . . Results of surveys [3] show that in most cases companies outsource the maintenance on a contract basis, of which the yearly cost are known in advance. In principle, the operating costs may differ from year to year. IV. N UMBER OF PROCESS INTERRUPTIONS WITH MITIGATION

E. Number of process interruptions without mitigation Using the data and the methods as described in the previous paragraphs, we can calculate the number of interruptions that the sensitive equipment endures when no mitigation method is used. For the given case, this results in 7.01 process interruptions per year. Of these, 1.20 interruptions are caused by voltage interruptions, with an average repair time of 3.95 hours. The other 5.81 are caused by voltage dips. III. M ITIGATION COST The cost of mitigation methods can differ a lot depending on the type of mitigation method. The cost consist of a fixed part and the operating costs A. Proposed mitigation devices The typical selection of mitigation methods starts with a number of offers, comprising of different mitigation principles. These offers give the customer access to the following data for each mitigation device: • apparent power; • efficiency at different loads; • autonomy (if applicable); • lifetime expectancy (also of parts such as batteries); • purchase price; • operating costs; • installation costs; • M T BF and M T T R of the mitigation method; • maintenance costs (in case of a contract); • mitigation capability. The following mitigation devices are proposed as protection for the load: 1) 300 kVA static UPS, autonomy 15 min; 2) 300 kVA static UPS, autonomy 30 min; 3) 300 kVA static UPS, autonomy 90 min;

A. Influence of protection device Introducing mitigation devices reduces the number of voltage dips and interruptions at the terminals of the sensitive device. It is this reduction in process interruptions that determines the actual economic value of the mitigation method, and not, as is often believed, the number of remaining process interruptions. The influence of the mitigation device is depending on the type of mitigation that is used. In case of a UPS system (mitigation methods 1–5), the sensitive device will be interrupted in case there is an event (interruption or dip outside the voltage tolerance characteristic) during the unavailability of the UPS (UPS in bypass) or when the interruption duration exceeds the autonomy of the UPS, during the availability of the UPS (4). The number of voltage interruptions, exceeding the autonomy can be calculated when assuming the exponential distribution of the interruption duration (3). taut

Pr

p(t > taut ) = e− M T T R · P

(3)

where:  taut    MTTR Pr    P

autonomy with rated load; Mean Time To Repair; rated power; consumed power.

M T T RU P S M T T RU P S + M T BFU P S M T BFU P S + p(t > taut ) · (4) M T T RU P S + M T BFU P S where #P Iwith{out} stands for the number of process interruptions with{out} mitigation device. Method 5 (two UPSs placed in N − 1 redundancy) is not able to protect the sensitive devise if #P Iwith =#P Iwithout ·

both the UPSs are unavailable and the grid is unavailable; when one of the UPSs is unavailable and the interruption is longer than the autonomy of the other UPS; • if the grid is unavailable for a time longer then the autonomy of the two UPSs combined. Mitigation methods 7–9 offer no protection against voltage interruptions, they only extend the voltage tolerance characteristic and thereby only protect against voltage dips. The number of process interruptions after mitigation is equal to the number of voltage interruptions added with the number of voltage dips that are outside the extended voltage tolerance characteristic (number of dips without mitigation multiplied with the percentage of dips that are outside the voltage tolerance characteristic). For devices 7, 8 and 9, the relative amount of voltage dips that is not within the extended voltage tolerance characteristic is respectively 5.7 %, 8.3 % and 6.3 % (using the same voltage dip distribution as displayed in fig. 3). •

where #P I is the number of process interruptions.

•

D. Intervals of optimal mitigation method If we apply (6) for varying process interruption costs, using the NPV method (1), and this for each mitigation method, we obtain a plot as in fig. 4. The interruption cost is divided in intervals where one solution is optimal. optimal solutions for the given grid reliability. These ranges in interruption cost can be used in the decision process because rough estimates of the interruption costs are available. This method gives the interruption cost as a result, not as a variable. Fig. 4 shows a typical example for three mitigation devices, the device with the highest NPV is the most economic. The three devices give four possible optimal solutions: no mitigation in zone A, the long dashed method in zone B, the method presented by the continuous line in zone C and the method according to the line with the short dashes for high interruption costs.

B. Total number of faults after mitigation The result of mitigating is a reduction in number of process interruptions. For the nine proposed methods this is given in table III. It can be noticed that the placement of the two UPSs TABLE III R EDUCTION OF THE NUMBER OF PROCESS INTERRUPTIONS AFTER MITIGATION

method 1 2 3 4 5 6 7 8 9

without mitigation #/yr 7.0096 7.0096 7.0096 7.0096 7.0096 7.0096 7.0096 7.0096 7.0096

after mitigation #/yr 1.1125 1.0279 0.7491 0.4661 0.8774 1.2025 1.9120 2.3280 2.0080

reduction

improvement

% 84.1287 85.3358 89.3132 93.3511 87.4836 82.8451 72.7231 66.7884 71.3536

#/yr 5.8971 5.9817 6.2605 6.5435 6.1322 5.8071 5.0976 4.6816 5.0016

in N − 1 redundancy in this case does not lead to very high availability (not the highest). The reason for this is the rather small autonomy (15 minutes per UPS) in comparison to the high average interruption duration (3.9 hours) and the high reliability of a single UPS.

Fig. 4. The calculation of the NPV with variable interruption costs and constant grid reliability leads to intervals of interruption cost, where one optimal solution exists.

The intervals for the given application and grid conditions are given in table IV. Fig. 5. NPV as a function of process interruption cost gives lead to intervals of interruption cost where one solution is optimal

C. Techno-economic optimization The goal of mitigation methods is the reduction in process interruptions, making the interruption costs lower. Of course these protection methods are not free. A techno-economic optimum is reached when the total cost of power quality (interruptions + mitigation) is minimal (5). n o Min CT otal = Cmitigation + Cinterruption (5) This can be reformulated as maximizing the return of the reduction in interruptions reduced with the additional cost of mitigating [17]: n o Max return = #P I · Cone interruption − Cmitigation (6)

TABLE IV I NTERRUPTION COST INTERVALS Optimal investment No investment Investment 8: Investment 7: Investment 6: Investment 1: Investment 2: Investment 3: Investment 4:

Vectek DySC Flywheel UPS Static UPS, 15 min Static UPS, 30 min Static UPS, 90 min Static UPS, 180 min

Interruption cost ($ /int.) 0 → 1, 190 1, 190 → 10, 340 10, 340 → 18, 210 18, 210 → 26, 320 26, 320 → 38, 260 38, 260 → 62, 250 62, 250 → 70, 460 70, 460 → ∞

E. Practical use The usefulness of this approach lies in the fact that the results are easy to interpret, even for non-engineers (e.g. managers). For instance: method 6, the flywheel UPS is useful if the average interruption cost is at least $ 18,210 and not more than $ 26,320, since in that case a static UPS with 15 min autonomy is optimal. The problem statement is shifted from ‘How much do the interruptions cost?’ to ‘How much am I willing to pay, in order to avoid process interruptions?’. The mitigation method which is optimal for the highest interruption costs (method 4) is the device with the lowest number of remaining process interruptions after mitigations (see also table III). V. S ENSITIVITY TO GRID RELIABILITY Since the grid indices are only roughly known, it is interesting to see the influence of these values in the optimization. Fig. 6 shows the influence of λgrid , M T T Rgrid and #dipsgrid , with the other variables at their original value (λgrid = 1/yr, M T T Rgrid = 100 min and #dipsgrid = 16/yr). The thick horizontal line represents the value of the example (and table IV). The numbers in the figure represent which device is optimal in which area. From fig. 6 it can be clearly seen that device 7 and 8 are very sensitive to changing dip frequency, while the other devices are not sensitive to dips, but to the failure rate λ and the M T T R. Influence of lambda (= 8760/MTBF)

lambda (int/yr)

2

3 4

2

1.5

1 1 0.5

8 0

6 is optimal in this area

7 1

2

3

4 5 6 Interruption cost (Euro/int)

8

9

10 4

x 10

Influence of MTTR

200 MTTR (min)

7

150

8

100

1

6

7

2

3

4

50 0

1

2

3

4 5 6 Interruption cost (Euro/int)

7

8

9

10 4

x 10

Influence of #dips #dips (dips/yr)

40 30

6

8

20

2

3

4

7

10 0

1

0

1

Fig. 6.

2

3

4 5 6 Interruption cost (Euro/int)

7

8

9

10 4

x 10

Sensitivity to the different reliability indices

VI. C ONCLUSIONS The difficult process of selecting an appropriate mitigation method is simplified to a method that is easy to interpret, using only data that is available for the customer. Using this method it is also possible to compare totally different mitigation methods.

R EFERENCES [1] PRIMEN, The Cost of Power Disturbances to Industrial & Digital Economy Companies, CEIDS, Ed. Palo Alto, USA: EPRI, 2001. [2] D. Van Hertem, M. Didden, J. Driesen, and R. Belmans, “Optimal mitigation of voltage dips and interruptions,” in Proc. IEEE Young Researchers Symposium in Electrical Power Engineering, Delft, The Netherlands, March 2004. [3] D. Van Hertem, “Techno-economische optimalisatie van maatregelen tegen spanningsdips en -onderbrekingen,” Master’s thesis, Katholieke Universiteit Leuven, June 2003. [4] M. Didden, “Techno-economic analysis of methods to reduce damage due to voltage dips,” Ph.D. dissertation, Katholieke Universiteit Leuven, Dec. 2003. [5] R. Brealey and S. Myers, Principles of corporate finance, 6th ed. McGraw-Hill, 1999, no. ISBN 0-0770-9565-0. [6] H. Khatib, Economic Evaluation of Projects in the Electricity Supply industry, ser. IEE Power & Energy series, A. T. Johns and D. F. Warne, Eds. The institution of Electrical Engineers (IEE), 2003, no. 44, ISBN: 0-86341-304-8. [7] R. E. Brown, Electric power distribution Reliability, ser. Power Engineering, H. L. Willis, A. F. Sleva, and M. Shahidehpour, Eds. Marcel Dekker, Inc., 2002, ISBN: 0-8247-0798-2. [8] Information Technology Industry Council (ITI), “ITI (CBEMA) curve application note,” p. 3, 2000. [Online]. Available: http://www.itic.org/ technical/iticurv.pdf [9] A. Sumper, A. Sudrià, and F. Ferrer, “International reliability analysis in distribution networks,” in International Conference on Renewable Energies and Power Quality, no. ISBN: 84-607-9889-5, Barcelona, Spain, March 2004, paper 4.299.SUMPER, pp. 1–5. [10] M. Bollen, Understanding Power Quality Problems, Voltage sags and interruptions. Piscataway, New Jersey: IEEE Press, 2000. [11] A. Montebaur, Zuverlässigkeitsanalysen industrieller Elektrizitätsversorgungsnetze unter Einschlußder Productionsprozesse. Aachen: Institut fur Elektrische Anlagen und Energiewirtschaft, 1996, no. 42. [12] H.-J. Haubrich, Zuverlässigkeitberechnung von Verteilungsnetzen, Grundlagen – Verfahren – Anwendungen. Aachen: Institut fur Elektrische Anlagen und Energiewirtschaft, 1996, no. 36. [13] Hydro Québec, “Characteristics and target values of the voltage supplied by the Hydro-Québec medium and low voltage systems,” Hydro-Québec, Tech. Rep. 30012-01-02, June 2001. [14] SoftSwitching Technologies. Dynamic sag corrector (DySC) solution. [Online]. Available: http://www.softswitch.com/dysc.htm [15] Vectek Electronics Ltd. Voltage sags & surges. [Online]. Available: http://www.vectek.co.nz/products/voltage_sags_and_surges/index.htm [16] ABB. Series voltage restorer. [Online]. Available: http://www.abb.com/global/abbzh/abbzh251.nsf!OpenDatabase&db= /GLOBAL/SEITP/SEITP322.NSF&v=9AAC910009&e=us&m= 9F2&c=D15A21B3FEC808E8C1256DE1004BABFD [17] D. Van Hertem, M. Didden, J. Driesen, and R. Belmans, “Economic evaluation of mitigation methods against voltage dips and interruptions based on stochastic reliability,” in International Conference on Renewable Energies and Power Quality, no. ISBN: 84-607-9889-5, Barcelona, Spain, March 2004, paper 4.277.HERTEM, pp. 1–8. [18] M. Didden, K. Stockman, R. Belmans, and W. D’haeseleer, “Considerations for choosing the appropiate voltage sag mitigation device,” in Proc. 17th International Conference on Elektricity Distribution, Barcelona, Spain. [19] A. Chowdhury and D. Koval, “Application of customer interruption cost in transmission network reliability planning,” IEEE Transactions on Industry Applications, vol. 37, no. 6, pp. 1590–1596, 2001. [20] APC, “Avoiding costs from oversizing data center and network room infrastructure,” American Power Conversion, 2004. [Online]. Available: ftp://www.apcmedia.com/salestools/SADE-5TNNEP_R4_EN.pdf [21] ——. (2002) Effect of UPS on system availability. American Power Conversion. [Online]. Available: http://www.limbtec.co.uk/documents/ whitepapers/WhitePaper24.pdf [22] W. Grant Ireson and J. Clyde F. Coombs, Eds., Handbook of Reliability Engineering and Management. McGraw-Hill, 1988. [23] C. Warren, R. Ammon, and G. Welch, “A survey of distribution reliability measurement practices in the u.s.” IEEE Transactions on Power Delivery, vol. 14, no. 1, pp. 250–257, 1999.