1 Direct and Indirect Measurement of Residential and Commercial CIC: Preliminary findings from South African Surveys R. Herman, Senior Member, IEEE & C.T. Gaunt dispensation after 1994 and a National Electrification Abstract—Interruptions in the supply of electricity to customers Programme meant that for a period up to 2000, new are inevitable but it is important to measure, monitor and minimize connections were made at the rate of about 450’000 per the costs incurred. In this paper it is shown that customer annum. Extensive load research resulted in more appropriate interruption costs (CIC) depend on a variety of factors. They include load models being developed for these ‘electrification’ the type of country and locality in which the interruptions occur. Actual costs depend on the types of appliances that are interrupted projects. A probabilistic design method was also developed and the sectors within which they are associated. Not all the methods for the low voltage networks within HL3 (Hierarchical Level of evaluation developed in a first world setting are appropriate 3). However, insufficient attention was given to HL1 and everywhere. This paper describes pilot studies undertaken in South HL2 issues, which by nature require much longer lead times. Africa to determine the characteristics of CICs in the residential and This created a situation which might be similar to aging, de- commercial sectors. The results are discussed and analyzed. regulated networks in developed countries. The problem has reached critical proportions during the last two years. Supply Index Terms— Customer Interruption Costs, reliability evaluation. failure (SAIFI and SAIDI) figures have increased significantly and the supply authorities regularly implement curtailment. I. INTRODUCTION This type of supply scenario is well known in other parts of Africa where the development has not kept pace with the A pilot study was undertaken to determine the impact of the interruption of supply to customers so that it could be factored into design, operational and regulatory decisions. To demand. The role of the electricity distribution utility is to meet the demand of its customers at a level of quantity, price, quality add value to the impact measurements, an attempt was made at and reliability. The price and quantity parameters are estimating the Customer Interruption Costs (CIC) in financial routinely regulated, but standards for the Quality of Supply terms. It may not always be possible to determine this (QOS) and reliability are required. These need to be accurately since it involves the responses from the customers. monitored and should be included in the design of new In our study we investigate two approaches – (a) the direct networks. However, the familiar SAIDI and SAIFI indices and (b) the indirect methods. may not be adequate for this purpose because, while they give The context within which this work was conducted has average annual outage duration and frequency figures, they do both similarities with and differences to other parts of the not reflect the financial impact on the customer. The world. South Africa is a developing country with large motivation for this study is to develop appropriate methods to industrial and commercial consumers (mines, SASOL and determine the CIC for the various sectors in South Africa. smelting, and several large cities) and a residential population When this has been done, it will have a twofold application: of about 45 million. Until about two decades ago (a) to develop a ‘reliability value’ that the National Regulator electrification had not reached a significant portion of (NERSA) may use to asses a utility’s compliance and (b) to residential customers. Concern about the oil crisis and other incorporate CIC within planning and design decision-making political drivers during the 1970s initiated the construction of generation and transmission infrastructure to meet the need II. INTERRUPTIONS AND THEIR RELATIONSHIPS more than adequately. In global terms the South African supply structure was robust at all three hierarchical levels of A. Interruption types reliability, similar to developed countries. A new political For the purposes of this paper interruptions are typified by the following descriptions. This work was supported in part by THRIP funding. • Momentary interruptions: These are characterized by R Herman and C.T. Gaunt are with the University of Cape Town, South Africa, (e-mails: firstname.lastname@example.org and email@example.com) flicker, dips, sags caused by switching actions such as 2 auto-reclosers and the switching of large loads. They Energy-based appliances such as water heaters, freezers, usually manifest themselves at HL2 and HL3. pumps, compressors etc., may have a low to high impact • Sporadic interruptions: These are mainly found in depending on whether there is a storage component for the developed countries with robust delivery systems. In energy. For example, hot-water cylinders have low impact these situations the interruptions are often caused by while lifts could have a high impact. Traffic signals are a severe weather conditions and occur mainly at HL2 mixture of lighting energy and control device. They are and HL3. Most CIC surveys in developed countries obviously vulnerable to interruptions. measure sporadic interruptions. In many applications a connected load consists of ‘series’ • Chronic interruptions: These interruptions that occur and ‘parallel’ combinations of intelligence-based and energy- more regularly are generally due to load shedding based devices, collectively forming processes. Even necessitated by a shortfall in generation (HL1) or momentary interruptions can cause a process to fail, resulting severe ageing, lack of maintenance or overloading of in: the delivery systems (HL2 and 3). This type of • Loss of production interruption is typical of developing countries where • Spoilage the load growth outstrips the supply and reinforcement • Expensive restarting procedures requires a substantial lead time. This type may also C. Interruption cost to customer and utility occur in developed countries when a particular fuel source such as hydro suffers from a drought condition The end products of the supply utility and of the connected (HL0). In reliability terms the SAIDI values are usually customers are quite different and thus the interruption costs to significantly higher in developing than in developed the two entities are also quite diverse. Utilities lose revenue countries. when they fail to supply customer load and in reliability terms this is measured as ENS (energy not served). The unit of B. Interruption/appliance relationships measurement is simply $/kWh. Surveys are conducted to Supply failures essentially affect the use of a variety of relate these figures to various customer sectors. The cost of connected electric appliances, thereby incurring a cost to the losing customers is also analyzed as LOLC (loss of load cost) customer. This cost depends on the type of appliance. In this and is dependent on duration (SAIDI) . Further costs to the paper we identify these as (a) intelligence-based, (b) energy- utility are incurred in restoring supply and, possibly, when the based and (c) process-based. Schematically they are presented regulator enforces penalties or customers litigate for failure to in figure 1. meet contractual obligations. The end products of connected customers have widely differing values depending on their applications, ranging from Control (1 household comforts to sophisticated commercial products or Tills Security services. Reference to the discussions in the previous section Intelligence-based Computers suggests that the value of an end product may not be directly Appliance Communication related to the amount of electrical energy used. In many cases Display continuity of supply rather than capacity is the most important (2 issue. To use a blanket unit of $/kWh for all sectors is Motors: Energy-based pumps, lifts, machinery therefore not a good measure of CIC. In emerging economies Appliance Other energy conversion: small to medium sized enterprises are particularly vulnerable heat, cooling, lighting to the chronic type of interruption. High SAIFI figures can Supply chemical threaten the viability of the business by eroding commercial (3 confidence. Process-based Series/Parallel D. Economic and socio-political impact (South African Appliance groups combinations of (1) & (2) experience) The costs of interruptions are not limited to direct financial Fig. 1. Supply interruption on various appliance types impact. Production and commerce contribute to the Gross Domestic Product of a province or country and constraints on The first type of appliance generally supplies a small output caused by electricity interruptions reduce the GDP. amount of power to an intelligent device such as tills, While the impact of sporadic interruptions would be computers, fax machines, control devices etc. In these cases negligibly small, the effects of chronic interruptions can be quantity of energy is not as important as availability. In significant. For example, the extended interruptions practice such devices may be protected against temporary experienced in parts of South Africa during 2006 and 2007 supply failures by using un-interruptible power supplies initiated substantial expenditure on UPS and stand-by (UPS). In some cases, where interruptions are not frequent generator systems in the expectation of continuing problems. and the extra protection is not used, even a small interruption Since individual back-up is inherently less economical than a can incur significant loss to the customer. This could take the reliable utility supply, the overall economic value is negative form of system failure, loss of data, loss of control, loss of even if the individual installations are viable as insurance for sales etc. many customers. Further, the perceptions of expected future 3 interruptions affect business confidence and willingness to C. Event-chasing invest in new industrial and commercial facilities and, during .This type of survey is, in essence, a sub-set of the direct the same period, there were several reports of potential foreign approach and is conducted shortly after a recorded investors deciding to locate in other countries with more interruption event. It is particularly useful as a means of reliable electricity supplies. estimating outage costs associated with a specific feeder Chronic interruptions can also have social and political where the customer types are known . Most utilities record impact through damage to a utility’s brand image or if it is outages on a regular basis and so interruptions are time- perceived, for example, that rural communities are being stamped and their durations noted given second-class supplies. It is immediately evident that both direct and indirect approaches have merits and disadvantages. In situations III. DISCUSSION OF CIC MEASUREMENT APPROACHES where there are more frequent interruptions of varying duration (high CAIFI and CAIFI indices), as would be A. Direct and indirect expected in less developed regions, the direct approach can be Measuring CIC is not a simple as it relies on voluntary feasible. In fact, both approaches may be applied customer responses collected in the form of surveys. The simultaneously yielding a composite approach from which the Cigré report Methods to consider customer interruption costs scaling of the CIC measurement may be done more in power system analysis  reviews the methods followed in comprehensively. 13 different countries. It is immediately evident that different countries use different methods, and have adapted their CIC D. Interruption scenarios surveys to suit their own conditions and constraints. Despite Each combination of event and context is an interruption the vast number of referenced papers and reports (148) in the scenario. The various combinations are illustrated in Table 1. Cigré document, the conclusion avoids identifying the merits Customers who have experienced actual events can be asked of specific methodologies. In essence two philosophies about the cost of the events as experienced (W). This is emerge: considered as a direct approach since it is based on actual • Direct approach interruption events. The other scenarios (X, Y and Z) relate to • Indirect approach the indirect approach where the effects of the same event The direct approach attempts to measure the impacts of would have been in a different (hypothetical or conceptual) actual interruption events subsequent to their occurrence. The context (X) that did not apply at the time. Customers can also indirect approach is based on hypothetical scenarios suggested be asked to estimate the costs of a hypothetical event that to the customers and relies on the customers’ perceived happened in their own real context (Y). Finally, customers responses to such scenarios. can estimate the costs of conceptual events in hypothetical contexts (Z). TABLE 1 INTERRUPTION SCENARIOS The intuitive, scientific way to measure the cost of an interruption would be: Context • identify the interruption (occurrence and duration) Actual Conceptual • identify the customer’s characteristics Event Actual W X • determine the financial impact to the customer (value Conceptual Y Z of inconvenience and incurred mitigation costs) • repeat the measurement a significant number of times IV. DATA COLLECTION METHODS to include different customer interruption frequencies (CAIFI) and durations (CAIDI) A. Data collection for scenario W • determine the statistical properties of the measurement However, most of the authors of papers on CIC Scenario W collects costs of actual events in actual context measurement in the domestic sector of developed countries from customers. Two options arise: appear to favor the indirect approach, suggesting that the • Customers are asked to respond with information about “actual experience” approach is not feasible . recent faults identified by the customers. • The questionnaire is completed for an interruption B. Communities with high and low SAIFI and SAIDI identified by the survey managers. Two variants of this The direct approach is less feasible on robust systems with option, one immediately after the event and another after a low SAIFI and SAIDI because there are insufficient outages delay of about three weeks, can test the effects of the in a given survey interval. This is the case in the developed respondents’ emotion and memory. This option requires countries where CIC has been extensively surveyed for more close liaison with the operations controllers for the utility’s than two decades, providing the basis of most of the published network. It also allows closer linking to the network research literature. The alternative is to suggest hypothetical performance indices. interruption scenarios to the customers and request them to Questionnaires directed to collecting data for scenario W can respond to questions concerning willingness to pay (to avoid also collect data for a small range of related conceptual events an interruption) and willingness to accept certain levels of (Y) or conceptual contexts (X) related to the actual event and CAIFI and CAIDI. In effect the CIC is then measured by context, provided the alternatives do not confuse the proxy. respondents. 4 B. Data collection for scenario X Eight scenarios were suggested in the survey but only one of Scenario X collects costs of actual events in a conceptual these is given on each form. context. If the conceptual is very different from the actual B. Forms for indirect method – type F context, customers may have difficulty assessing the cost The indirect method used eight different forms. Each impact of an interruption. For contexts with more serious suggested a particular hypothetical scenario in terms of interruptions (more frequent or longer) it will be useful to ask context and event. No questions are asked about actual about the contingency actions a customer might adopt, such as interruption events. A typical scenario would be: “Imagine paying a premium for higher reliability or purchasing a you had a power failure this week: Wednesday, 5 to 9 pm standby supply. without warning”. Questions, similar to those in the previous In the South African context of relatively high interruption section relating to the customer’s willingness to pay to avoid rates (such as in most rural areas and recently in urban areas an interruption, were also added to these forms. also), combined with the component of Free Basic Electricity that is not applied uniformly, it might not be useful to ask C. Survey results – Direct costing questions about the willingness to accept lower reliability for A total of 237 completed response forms were processed. It a lower electricity tariff. was apparent from the A-type questionnaires that the C. Data collection for scenario Y estimated cost of the most recent interruption had a high variation. This phenomenon is also reported by Ghajar et al Scenario Y collects costs of conceptual events in actual . context from customers. Although the research was conducted as a pilot survey the The conceptual events can be: sample size is large enough to disregard the effects of • The same as actual events experienced by other excessive sampling variation. Cost of interruptions are customers (as discussed in scenario W) in a similar or incurred in different ways. different context • It depends on appliance use: In some cases the loss • Events identified as happening at the worst time of of supply is easily deferred e.g. water heating. In some day/month and with durations of specified periods to cases it is not deferrable e.g. fixed meal times or cover the range of interest, e.g. ½, 1, 2, 4, 8, 16 hours. computer applications. • Events happening at other times of day/month, with the • It depends on damage related to the interruptions same durations. e.g. damage to hard-drives, security systems and food The “existence” of the actual context experienced needs to be spoilage. conveyed to the customer and that it from within this context Within these, the severity will also be affected by prior that the questions are to be answered. If the actual experience warning of an interruption. Short interruptions of events is too different from the actual experience (for characteristically without warning can have a relatively high example, the customer may not have experienced very long CIC, as shown in Table 2. In fact, a small number of very interruptions) the responses to some conceptual events might high estimates were received, suggesting that the customers not be reliable. It might be possible to check the applicability were not ‘homogeneous’ although belonging to the same of the conceptual range to actual experience by comparison ‘residential’ sector. Including these very high values in the with the questions describing CAIDI and CAIFI. measurement of average CIC significantly distort the results. D. Data collection for scenario Z It was therefore decided to separate estimates larger than R300 Scenario Z collects costs of conceptual events in conceptual ($45), which is approximately three times the largest standard contexts from customers. The events may be the same as or deviation, and to regard them as outliers. This accounted for similar to those for scenario Y but the context must be clearly approximately 10% of the respondents. As mentioned, defined. The range of characteristic interruption frequency willingness to pay (WTP), was measured as a response to a and duration in South Africa is indicated in tables from the hypothetical scenario. Tables 2 shows the results of CIC and NRS specification NRS048-2. WTP, excluding the outliers. TABLE 2 RESIDENTIAL SECTOR - WITHOUT OUTLIERS . V. RESULTS FROM RESIDENTIAL PILOT SURVEY Duration Direct-Cost [R] WTP [R] [h] mean stdev mean stdev Pilot studies were conducted in South Africa to measure 0.25 56.5 95.22 10.5 23.62 CIC in the residential and commercial sectors. Two sets of questionnaires were prepared for the application of both direct 1.25 34.88 68.47 13.33 24.31 and indirect measurement of CIC for the residential sector. 4.00 31.85 59.18 13.65 25.68 11.00 46.5 86.03 17.03 23.65 A. Forms for direct method – type A The survey form asked specific questions relating to the It was evident from the results that the outliers include most recent power failure that occurred during the past year. estimates that are far above the norm, suggesting that these The form also included a section in which customers were customers incurred significant consequent costs. From the asked to respond to hypothetical scenarios (frequency plus survey the highest estimates were for: duration). This section was included to measure the • Freezer (spoilage) – R1500 customer’s willingness to pay to avoid interruptions (WTP). • Computer – R1200 5 • Alarm/control – R1000 Beta Distribution of Willingness to Pay How the outliers should be represented and used in decision-making is uncertain. The figures reflect high 0.06 dispersion as well as skewness when represented as a probability distribution. In an earlier paper it was suggested 0.05 that a beta probability function could be used in such cases Probility Density 0.04 . Figure 2 shows the resulting p.d.f. for the directly measured CIC. Assuming that the goodness of fit is 0.03 acceptable the value of the CIC at the 90-percentile 0.02 confidence level is R149.19 while the mean is R37.54. Figures 3 and 4 show the resultant p.d.f. graphs for the 0.01 measurements of duration (CAIDI) and WTP, as reported by the surveyed customers. The maximum value of R70 was 0 0 10 20 30 40 50 60 70 arbitrarily set as the highest choice for the question on WTP. WTP - Rands Figure 5 shows the distribution of the all the customers’ perceived frequency of interruptions per year (CAIFI). Prob. density Fig. 4. Beta probability function of WTP for residential customers Beta Distribution of Direct Costs Beta Distribution of CAIFI 0.025 0.025 0.02 0.02 Probility Density Probility Density 0.015 0.015 0.01 0.01 0.005 0.005 0 0 0 50 100 150 200 250 300 0 50 100 150 200 250 300 CIC Rands Interruptions / year Prob. density Prob. density Fig. 2. Beta probability function of direct CIC for residential customers, excluding outliers Fig. 5. Beta probability function of interruptions/yr for residential customers Beta Distribution of CAIDI 0.45 VI. RESULTS FROM COMMERCIAL SURVEY 0.4 A fairly small but carefully executed survey of the CIC for 0.35 commercial customers was undertaken in the Durban area . The objectives were to determine methods and to uncover Probility Density 0.3 future pitfalls when estimating CIC within the commercial 0.25 sector. The Standard Industrial Classification (SIC) described 0.2 in Statistics South Africa (1993) was used as a guide to 0.15 identify suitable sub-sectors. These were defined as: 0.1 • Retail 0.05 • Hotel and restaurant 0 • Monetary intermediation (financial) 0 2 4 6 8 10 12 14 16 Duration [hrs] A. Methodology Prob. density From the literature it appears that there are three main CIC assessment methodologies: analytical methods, blackout case Fig. 3. Beta probability function of interruption duration for residential studies and customer survey methods. Generally, customer customers survey is preferred. But within this method there are three approaches: the contingent evaluation using the WTP concept, the indirect cost method and the direct cost approach when CIC is tangible and quantifiable . The last method was used in this pilot study. However, a major problem was encountered in accessing data in some cases because on-site 6 managers were reluctant to divulge information about their The results were again characterized by large variances. business for confidential reasons. In some cases, the Mean and standard deviations of the CIC values recorded are corporate nature of commercial enterprises isolated staff from shown in Table 3. operating principles and practices. TABLE 3 SUB-SECTOR WORST CASE CIC – MEAN & STANDARD DEVIATION CIC [R] Retail Financial Hotel& Rest B. Results for the retail sub-sector Duration Mean Stdev Mean Stdev Mean Stdev Worst-case (in terms of time-of-occurrence) data processed 2 sec 182 547 18936 29127 455 1508 20 min 6398 9983 37252 61170 2054 2855 from the responses submitted by the retail sub-sector yielded 1 hr 16303 26251 96765 179714 4958 5173 the graphs in figure 6. Here the ‘worst-case’ total cost is 4 hr 47042 73580 211703 403812 29542 31546 shown against a breakdown of some of the components 8 hr 67111 97358 249506 438927 39624 31229 (wages, sales and perishables). This is repeated in figures 7 and 8 for the other sub-sectors. VII. DISCUSSION OF RESULTS The results of the surveys in both the residential and commercial sectors were characterized by much greater variance than had been expected from the literature. The large spread of the results demonstrates a need to understand and use the values of ‘outliers’ correctly. A probabilistic approach that uses the distribution of the results instead of a single value may be the only way in which to cope with the large variance of CIC. The shapes of the Beta pdf of the direct costs and WTP results for the residential customers are similar, but there is a clear difference in the magnitude. The WTP is significantly lower than the direct cost of an interruption. The difference Fig. 6. Retail ‘worst-case’ CIC function and its components for the retail sub- might represent customers inherently being willing to share sector the risk of interruptions, but whatever the reason it is clear that the results of direct and indirect methods are not the same and need to be applied accordingly. The time of occurrence significantly affects the CIC. The difference according to time of day in the commercial sector can be extreme, with interruptions outside normal operating hours incurring virtually no cost, and there is significant variation according to the day of the month. CIC in the residential sector did not vary to the same extent with time of occurrence, but is clearly evident. The time of occurrence of outages in systems with chronic interruptions tends to correspond with the system demand because load shedding and network overloading occur when the demand is high. This means that time-averaged values of CIC cannot be used to determine the costs of system Fig. 7. Retail ‘worst-case’ CIC function and its components for the hotel and interruptions, as suggested in many approaches to optimizing restaurant sub-sector system reliability. VIII. CONCLUSIONS The recent stresses in the South African electricity system have provided an opportunity to identify the differences in the results of CIC surveys using direct costs and WTP in the residential sector. Similarly, the conditions enabled reliable direct cost results to be collected in three sub-sectors of commercial customers. It is recognized that the results measured under conditions of chronic interruptions might differ from the costs incurred by the same or similar customers experiencing only sporadic interruptions, so the results reported here may not be fully comparable with other published figures. The concept that Fig. 8. Retail ‘worst-case’ CIC function and its components for the financial CIC might differ for chronic and sporadic interruptions services sub-sector requires further research and measurement. 7 We suggest that where SAIFI figures are high enough for respondents in a survey to recall actual interruptions, the direct approach could be used successfully in the residential sector. Event chasing is recommended where possible. There were large differences between the mean values of CIC and the ‘outliers’ in all sectors measured, for both the direct cost and WTP approaches. This indicates that the practical variation of CIC must be taken into account in reliability studies for utility planning and in regulatory standards and compliance monitoring. Representing the results in the form of a Beta p.d.f. emphasizes the dispersion as well as the skewness. IX. ACKNOWLEDGMENTS The study in the commercial sector was carried out by Geoff Jordan under the direction of the authors, and his careful work is appreciated. Part of the research in the residential sector was sponsored financially by Eskom, and was carried out with collaboration from Eskom and the National Energy Regulator of South Africa. X. REFERENCES  Leite da Silva AM, da Fonseca Manso LA, de Oliveira Mello, Billinton R, “Pseudo chronological simulation for composite reliability analysis with time-varying loads”, IEEE-Trans. Power Syst., vol 15, pp73 – 80, 2000.  Billinton R (convenor), “Methods to consider customer interruption costs in power system analysis”, Cigré Task Force 38.06.01, Paris, 2001.  Wacker G, Wojcznski E, Billinton R, “Interruption cost methodology and results – Canadian residential survey”, IEEE T-PAS, vol. PAS-102, 1983.  Billinton R, Wangdee W, “Estimating customer outage costs due to a specific failure event”, IEE Proc.-Gener. Transm. Distrib. Vol 150, pp668 – 672, 2003.  Ghajar R, Billinton R, Chan E, “Distributed nature of residential customer outage costs”, IEEE-Trans. Power Syst., vol 11, pp 1236 – 1243, 1996..  Cross N, Herman R & Gaunt CT, “Investigating the usefulness of the Beta pdf to describe Parameters in Reliability Analyses”, PMAPS06, Stockholm, July 2006.  Jordaan G, P “Segment specific customer interruption costs in the commercial sector”, Project thesis, University of Cape Town, 2006.  Tiedemann KH, “Estimating the value of reliability for business customers”, PMAPS04, Iowa, September 2004. Dr. Ron Herman obtained a BSc (Eng.) degree from the University of Cape Town. After working with ESKOM he joined the staff of the University of Stellenbosch, completing MSc and PhD. degrees in Electrical Engineering. He developed the Herman-Beta method that is currently used for probabilistic low voltage distribution design in South Africa. He is presently a part- time Research Officer at the University of Cape Town, South Africa. Trevor Gaunt is a Professor in the Department of Electrical Engineering, University of Cape Town. He is married and has four children.