The use of Multi-Attribute Trade-Off Analysis in Strategic Planning For an Electric Distribution utility: An Analysis of Abu Dhabi Distribution Company Richard D. Tabors, Ph.D. Rick Hornby Tabors Caramanis & Associates rtabors@tca-us.com rhornby@tca-us.com Abstract Multi-Attribute Trade-Off Analysis (MATA) provides decision-makers with an analytical tool to identify Pareto Superior options for solving a problem with conflicting objectives or attributes. This technique is ideally suited to electric distribution systems, where decision-makers must choose investments that will ensure reliable service at reasonable cost. This paper describes the application of MATA to an electric distribution system facing dramatic growth, the Abu Dhabi Distribution Company (ADDC) in the United Arab Emirates. ADDC has a range of distribution system design options from which to choose in order to meet this growth. The distribution system design options have different levels of service quality (i.e., reliability) and service cost. Management can use MATA to calculate, summarize and compare the service quality and service cost attributes of the various design options. The Pareto frontier diagrams present management with clear, simple pictures of the trade-offs between service cost and service quality. 1. Introduction The objective of this paper is to describe a methodology for evaluation of investments in electric distribution systems that allow the analyst and decision-maker to evaluate the trade-off between conflicting objectives or attributes. Multi-Attribute Trade-Off Analysis is based on the concepts of Pareto optimality. It provides for the visual presentation of the trade-off values. The paper is based upon extensive work by the authors providing guidance to management at ADDC on techniques for identifying and evaluating potential investments to improve the quality of electric distribution service, and reduce its cost. It provides MATA results for alternative potential investments using System Average Interruption Duration Index (SAIDI) and System Average Interruption Frequency Index (SAIFI) as service quality attributes, and net present value of potential investments as the service cost attribute. The Emirate of Abu Dhabi is the largest of the seven Emirates within the United Arab Emirates. The Emir of Abu Dhabi is also the President of the UAE. The principle city of the Emirate is the city of Abu Dhabi located on the Persian Gulf. A secondary city, Al Ain is located inland and at higher elevation. Al Ain is an active oasis while Abu Dhabi has no natural source of fresh water. The Abu Dhabi Water and Electricity Authority (ADWEA) was formed in 1999 from a previously existing governmental department. The mandate of ADWEA is to transform the governmental entity into a set of commercially operational entities in generation, transmission and distribution. To date the generation companies have been privatized with the government holding marginal majority interests. TransCo and the two distribution companies Abu Dhabi Distribution Company (ADDC) and Al Ain Distribution Company (AADC) remain at this time under the umbrella of ADWEA. Generation is based upon indigenous natural gas employing the Flash Process for cogeneration of water and electricity. TransCo receives the energy at the generating station and delivers it to the low side of the step-down transformers. The distribution companies such as ADDC (the focus of this paper) are responsible for delivery of energy as well as for all customer contact. As ADWEA has moved toward increased business based strategies for the operation of ADDC (the focus of this paper) they and the management of ADDC have become increasingly sensitive to the cost of service to their customers and with this concern the relationship between customer measures of reliability and both operating and capital costs. 2. Multi-Attribute Trade-Off Analysis Multi-attribute tradeoff analysis (MATA) gives management the ability to evaluate and compare system design alternatives that achieve the same goal for a given planning situation, i.e., provision of distribution service to a specific level of future demand, but with different levels of service cost and service quality. This approach is based on the premise that most, if not all, major system 1
planning problems do not have a single optimum solution but rather have several alternative solutions that can achieve the same overall outcome but with different levels of attributes, such as service cost and service quality. The MATA approach is best suited to a problem with three basic characteristics - numerous alternatives, conflicting attributes and uncertainty. 2.1 Numerous alternatives. The problem of choosing a fundamental system design for the electric distribution network at ADDC or AADC has this characteristic. First, there are several different system design alternatives that could be used. Second, there are several different timing strategies for installing each of those system design alternatives. Thus, management has a number of alternatives from which to choose. 2.2 Conflicting attributes. Again, the problem of choosing a fundamental system design for the electric distribution network at ADDC or AADC has this characteristic. In choosing a system design from the range of potential alternatives management will be concerned with a number of attributes. These include: Service cost. Alternative system designs will have different levels of capital and operating costs as measured in terms of net present value, fils per kwh or capital cost per MW. Service quality. Alternative system designs will have different levels of service quality as measured by indices such as SAIDI, SAIFI and un-served energy 1. Ease of operation and maintenance. Alternative system designs will vary in terms of ease of operation and maintenance. For example, an automated design can automatically isolate, and identify the location of, the failed equipment. This design, by eliminating the need for field technicians to manually locate and isolate failed equipment, should be easier for staff to operate. 2.3 Uncertainty. Management faced with choosing a fundamental system design for the electric distribution network do so in the face of considerable uncertainty in variables over which they have little or no control. For example, in the three case studies there is uncertainty associated with the rate of load growth over the 10 year planning horizon, the specific location of that growth within the case study area, 1 Un-served energy, or MWh, is the annual quantity of customer load not served due to outages. the future prices of system components and future interest rates. The decision-support model can help management deal with this uncertainty by testing the sensitivity of the relative attractiveness of each design alternative to changes in the input assumptions. The model can be used to answer what if type questions by re-running it using different input assumptions. 2.4 Pareto optimality example The analytic structure of Pareto optimality allows the analyst to develop a frontier or curve of feasible alternative solutions to a given planning problem. Each alternative on the frontier represents a combination of attributes the can not, at the same time, be improved upon. In other words, for any alternative on the frontier, one cannot improve the value of one of its attribute without causing a reduction in one or more of its other attributes. As an example, Figure 1 plots the service cost and service quality attributes of several system design alternatives for ADDC, each of which could be used as a configuration for the electricity distribution system in the high load density sectors of Abu Dhabi Island. The Pareto tradeoff frontier is the thick, multi-segment line that approximates a curve. In choosing one of the system design alternatives as a solution to this planning problem there is a potential tradeoff between service quality, as measured in SAIDI, and service cost, as measured in the net present value (NPV) of each project. This is not surprising since a design with higher security of supply generally comes at a higher cost than a design with a lower security of supply. The system design alternatives that lie on the curve are considered optimal or Pareto superior. The remaining system design alternatives are said to be dominated because there is an alternative on the Pareto frontier that is either less costly or more reliable or both. These key points are illustrated in Figure 1, as follows: The plot for the phased 11 kv mesh dbl bus system design (beige diamond) demonstrates that it is not a Pareto Optimal alternative because it is possible to move vertically down to the frontier to alternatives with a lower service cost - the phased 11 kv Mesh (green diamond) for example and the same service quality. Alternatively it is possible to move horizontally left to the frontier to other alternatives that have the same service cost but higher service quality. Thus all of the alternatives on the frontier are better solutions than the phased 11 kv mesh dbl bus. The plots of alternative system designs located along the frontier, the phased 11 kv Mesh (green diamond) for example, demonstrate that it is not possible to find a competing alternative with an improved service quality attribute (i.e., lower SAIDI). The 2
complete 11 kv mesh (green triangle) for example, has an improved SAIDI but with increased service cost. This example is very simple but highly graphic. The advantage of the technique is that numerous feasible alternatives can be displayed and described in sets of twodimensional graphs. Management can then easily see the alternatives that are near, or on, the frontier and quickly discard the alternatives that are clearly dominated on all dimensions. 3. Overview of System Design Alternatives Before addressing each alternative system design in detail it is useful to establish a clear understanding of the fundamental difference between them aside from the capacity of the system components. That fundamental difference is the configuration of primary feeders, i.e., ring or open-loop versus primary mesh. In a ring or open-loop design each primary feeder is interconnected to another primary feeder at the boundary between the areas they serve. The interconnection is achieved using switchgear, such as switching substations, with a normal open point. This approach, illustrated in Figure 2, reflects the current approach used by ADDC and AADC. In a primary mesh design at least two primary feeders from each primary transformer are interconnected with the feeders from either an adjacent transformer, or the same transformer, without a normal open between the feeders. This approach, illustrated in Figure 3, is used by Scottish Power in Liverpool, England; Singapore Electric in Singapore; PECO in Philadelphia, Pennsylvania; Progress Energy in Orlando, Florida; and by other utilities in the United States. 4. Calculation of System Parameters The decision-support model is an Excel workbook. In the workbook there is a worksheet for each system design alternative which the model uses to estimate the key physical, service cost and service quality attributes of that alternative in each year of the planning horizon. The worksheet calculates and tracks the major physical changes in the network in each year under that particular system design alternative, i.e., the number, type and capacity of each major new system component installed in that year. The model uses that physical data to calculate the resulting service cost and service quality in each year. In order to make those calculations, the model requires input data and assumptions regarding the key physical components and engineering design guidelines for each system design alternative; capital costs for each key physical component, as well as an estimate of the annual cost of its operation and maintenance (O&M); technical losses associated with each key physical component; security of supply performance data by key physical component. The input data and assumptions used to develop and model each specific system design alternative for the case studies are based upon ADWEA and ADDC engineering design guidelines. 4.1 Key physical components and design guidelines. Each system design alternative is characterized according to the following key physical components and design guidelines: Primary substations number, capacity and peak loading of transformers; number of busses; Feeders number per primary transformer, cable size, configuration, peak loading, average length; Switching substations number per primary transformer and the number of express feeders per each switching substation; Distribution Transformer Utilization. The key physical components and design guidelines of each system design alternative considered are presented in the case study description below. 4.2 Unit cost assumptions. The capital and annual O&M cost assumptions for each system component of each system design alternative were developed from working experience and data developed at ADWEA and ADDC. The capital cost assumptions are based upon information from ADDC, as well as from international consulting reports prepared for ADDC. The annual O&M cost assumptions are based upon estimates of annual O&M hours for each component multiplied by an estimated loaded hourly rate for ADDC technicians. Our analysis has assumed that these unit costs will increase at an annual inflation rate of 2.5%. 4.3 Technical loss assumptions. The assumptions regarding technical losses associated with each system component of each system design alternative are based on detailed analysis and reporting by ADDC. The assumptions for the primary substations, feeders and distribution transformers are based upon calculations of technical losses under peak loading 3
conditions using standard loss data for the electrical components. 4.4 Performance assumptions. The model estimates outage indexes for each system design based on two major sets of assumptions performance per existing system component, and improvement in performance due to new system components and/or new system design. The performance per existing system component estimates assume that annual interruptions, annual duration of interruptions and annual un-served energy by cause of outage will be directly proportional to the following physical characteristics: Cable km of cable; Equipment number of feeders; Planned outages number of distribution substations due to the outages that are scheduled for maintenance on the substations; Distribution Transformers number of distribution transformers; Low voltage outages km of LV cable. We used those physical characteristics to develop estimates of unit annual interruptions, annual duration of interruptions and annual un-served energy by cause from security of supply data from ADDC over the period January 2001 through June 2003. That data provides a detailed break-down by region, voltage and cause of outage. The assumptions regarding improvement in performance due to new system components and/or new system design are based on a review of the causes of interruptions and research done on the impact of equipment conditions on the security of supply. For example, the mesh option will eliminate the majority of outages due to cable faults due to the redundant loop supply so the mesh option assumes the outages due to cable will improve (be reduced) by 90% if the feeders are installed as a mesh configuration. 5. Key Attributes of Each System Design Alternative As noted above, the worksheet for each system design alternative tracks the number, type and capacity of each major new system component installed in each year. The model then uses that physical data to estimate the following attributes of the system design alternative in each year: Technical Losses. The worksheet calculates the aggregate annual technical losses associated with the system design alternative each year. This calculation is made by applying the technical loss factors by system component, e.g., primary transformers, feeders, distribution transformers, to the number of those system components in each year and summing up for all system components. Service Costs. The worksheet calculates the aggregate capital costs and annual operation and maintenance (O&M) costs of the system design alternative for each year. The capital costs consist of a one-time set of transition investments to upgrade the existing network consistent with the engineering design guidelines, followed by annual investments to implement the new system design. For example, the transition costs could include the costs of installing indoor substations and circuit breaker switchgear to replace package units and distribution substations with TRM and QRM switchgear. The investments to implement the new system design include costs of new system components such as new primary substations, cable, and distribution transformers. The calculations of aggregate capital and annual O&M costs are made by applying the relevant unit capital and O&M costs to the number of new system components in each year, and then summing up for all system components. The worksheet also calculates the net present value of these aggregate annual capital and O&M costs over the planning horizon using a discount rate of 6%. Service Quality. The worksheet for each system design alternative calculates three measures of security of supply performance for each year - SAIDI, SAIFI and un-served energy. The calculation is made by applying the relevant factor for each security of supply measure to the number of existing and new system components forecast to be in place in year ten (2015), and summing up for all system components. For example, annual SAIDI per Km of existing 11 kv cable times km of existing 11 kv cable in place in 2015. Similarly, implementation of a 20 kv system design alternative to meet new load in Abu Dhabi Island north would require all new system components, so outages due to age and system component failure on that new 20 kv system should be minimal. 6. Case Study: High Density Region of Abu Dhabi Island 6.1 Analytic engine The MATA is driven by the decision-support model, which is governed by the distribution system design 4
criteria developed for, and accepted by, ADDC. The analytic engine uses the unit cost information reflective of the most recent (2003/2004) hardware purchases and installation costs for Abu Dhabi. 6.2 Load and geographic description The case study covers the highest load density portions of Abu Dhabi Island. The distribution delivery sectors represent a combination of high density high rise commercial structures as well as high density residential areas. Within the Emirate this segment of the island is the most reliability sensitive as well as representing the area that has grown the most densely over the past decade. The analysis begins with a forecast of peak demand by year for the 10 year planning horizon (to 2015). This forecast is used to determine the system expansion investments, e.g., new primary substations that will be required over the planning period. The forecast of peak demand in MW by sector has to be converted into a forecast of peak demand in kva by primary substation in order to estimate the number of primary substations and associated distribution equipment that is required to serve the load. 6.3 System design alternatives In the case study we evaluated six system design alternatives under three different implementation strategies for a total of 18 alternative scenarios. The three implementation strategies relate to the speed and extent of the transition to a new system design instant, phased and ten-year complete. The instant strategy assumes that the entire existing network is re-configured to the alternative system design in 2006, and that all load growth is met using the new design; The phased strategy assumes that the existing network continues to serve existing load, and that only new load from 2006 onwards is met using the new design; the ten-year complete strategy assumes that the existing network is gradually re-configured to the new system design over the ten year planning horizon, at a rate of 10% per year, and that new load from 2006 onwards is met using the new design. The six system design alternatives evaluated are: 11 kv Ring design (11 kv ring HD sw/s). 11 kv open-loop non-express and express feeders interconnected at switching substations, indoor substations, and 4x40 MVA primary substations. This system design alternative essentially represents the design of the current network after adoption of the engineering design guidelines. 11 kv Ring design (11 kv ring HD sw/s 3x55 MVA). 11 kv open-loop non-express and express feeders interconnected at switching substations, indoor substations and 3x55 MVA primary substations. This system design alternative represents the design of the current network except with a 3x55 MVA primary substation configuration instead of the 4x40 MVA configuration. 11 kv Mesh Design (11 kv mesh). 11 kv twofeeder primary mesh, indoor substations, 40 MVA primary transformers. 11 kv Mesh Design (11 kv mesh dbl bus). 11 kv two-feeder primary mesh, double buses, indoor substations, 40 MVA primary transformers. 20 kv Ring Design (20 kv ring HD sw/s). 20 kv open-loop non-express and express feeders interconnected to adjacent transformers via switching substations, indoor substations, 60 MVA primary transformers, concrete encased ducts. 20 kv Ring Design double bus (20 kv ring dbl bus) 20 kv open-loop feeders interconnected to the same or adjacent transformer, double busses, indoor substations, 60 MVA primary transformers, concrete encased ducts. Each of the alternative system designs utilizes the accepted engineering design guidelines and operation of the Distribution Management System (DMS) scheduled to be in effect by 2006 in Abu Dhabi. 6.4 Case study results We used the MATA analytic engine to determine the service cost and service quality of each system design alternative, and then to compare the alternatives according to those attributes. An analysis of the cost composition of the phased system design alternative, Figure 4 below indicates that the capital and O&M costs associated with primary transformers represent approximately 50% of the NPV. The annual level of capital investments on all other distribution system components, as well as the annual O&M costs, underlying the NPV of each alternative are consistent with the current levels approved in the budget for ADDC. We show below example Multi-attribute tradeoff charts used to present and compare the relative merits of each system design alternative. The tradeoff between service 5
quality in 2015, measured in SAIDI, and service cost, measured in NPV, are presented in Figure 5. This comparison indicates that the 18 potential system design scenarios can be quickly reduced to approximately seven that are Pareto superior, as well as five that are close. The seven are: the phased 11 kv ring HD sw/s the phased 11 kv ring HD sw/s the phased 11 kv mesh the complete 11 kv mesh the instant 11 kv mesh the complete 11 kv ring HD sw/s the instant 20 kv ring sw/s 7. Sensitivity Analyses One of the major difficulties in making a decision such as this is the uncertainty associated with the input assumptions. We used the model to test the sensitivity of the attributes of each system design to changes in the input assumptions. In particular we were seeking to determine whether the multiple Pareto superior system design alternatives would change position relative to each other, or the other design alternatives, in response to a change in input assumptions. We tested our base case results against variation in both load growth and discount rate, the two input assumptions where uncertainties were felt to be most critical. The results of these sensitivity analyses runs indicate that the relative positions of the three system design alternatives identified as Pareto superior under the base case input assumptions are not sensitive to changes in those assumptions, as shown in Figures 6a and 6b. 8. Conclusion Multi-Attribute Trade-Off Analysis represents a highly visual technique that can be used to present investment alternatives with conflicting attributes to decision-makers. The trade-offs between those attributes implicit in the decision process become highly visible allowing the decision maker (as opposed to the analyst) to understand quickly and easily what is given up to improve one or another of the attributes. In addition, and probably equally critically, the concept of Pareto optimality provides means of eliminating the truly inferior alternatives as has been shown in the case study of ADDC presented in this paper. 9. References [1] White, D.C. et al. Strategic Planning for Electric Energy in the 1980s for New York City and Westchester County. MIT Energy Laboratory, Technical Report, Cambridge MA, 1981 [2] Tabors, R.D., and D.P. Flagg. Natural Gas Fired Combined Cycle Generators, Dominat Solutions in Capacity Planning. IEEE Transactions on Power Systems PWRS 1986, 1 No 2 pp. 122-127 [3] Tabors, R.D. and B.L. Monroe, III Planning for Future Uncertainties in Electric Power Generation: An Analysis of Transitional Strategies for Reduction of Carbon and Sulfur Emissions, IEEE Transactions on Power Systems, 1991. [4] Tabors, R.D. et al A Computer Design Assistant for Induction Motors, Using Monte-Carlo Design Synthesis to Augment a Design Database, Conference Record of the 1991 IEEE IAS Annual Meeting, 1991 (With J. A. Moses, J. L. Kirtley, J. H. Lang and F. Cuadra). [5]Tabors, R.D A Simulator of the Manufacturing of Induction Motors, Conference Record of the 1991 IEEE IAS Annual Meeting,, 1991 (With C. L. Tucci, J. H. Lang, and J. L. Kirtley). [6]Tabors, R.D. and R. Hornby Decision Support Models for Evaluating Productivity Improvement Options and Capital Investments Task IV Final Report to ADWEA, April 2004. 6
Service Quality (SAIDI in 2015) vs Service Cost (Present Value AED m) Trade-off Results High Density - Abu Dhabi Island - Base Case 04 13 04 3,500 Phased 11 kv Mesh Phased 11 kv Mesh dbl bus 3,000 Phased 20 kv ring sw/s Phased 20 kv ring dbl bus 2,500 Instant 11 kv Ring HD sw/s present value (AED m) 2,000 1,500 Higher service cost Instant 11 kv Ring HD sw/s 55 MVA PRY Instant 11 kv Mesh Instant 11 kv Mesh dbl bus Instant 20 kv ring sw/s Instant 20 kv ring dbl bus 1,000 500 - Higher service quality 0 5 10 15 20 25 30 35 SAIDI in 2015 complete 11 kv Mesh complete 11 kv Mesh dbl bus complete 20 kv ring sw/s complete 20 kv ring dbl bus Figure 1: Example of Pareto Frontier SUBSTATION 1 SUB BREAKER SUBSTATION 2 FAULT OCCUR Figure 2: Open-Loop Feeder Configuration (Ring) OPEN POINT 7
SUBSTATION 1 S SUBSTATION 2 FAULT OCCUR NO OPEN POINT Figure 3: Primary Mesh Feeder Configuration Composition of system design costs by major category - Phased strategy High Density - Abu Dhabi Island - Base Case 04 13 04 100% 90% 80% 70% 60% 50% 40% all other O&M costs all other capital costs PRY Sub O&M costs PRY Sub capital costs 30% 20% 10% 0% Phased 11 kv Ring HD sw/s Phased 11 kv Ring HD sw/s 55 MVA PRY Phased 11 kv Mesh Phased 11 kv Mesh dbl bus Phased 20 kv ring sw/s Phased 20 kv ring dbl bus Figure 4: Comparative Costs of Alternatives Evaluated 8
Service Quality (SAIDI in 2015) vs Service Cost (Present Value AED m) Trade-off Results High Density - Abu Dhabi Island - Base Case 04 13 04 3,500 Phased 11 kv Mesh Phased 11 kv Mesh dbl bus 3,000 Phased 20 kv ring sw/s Phased 20 kv ring dbl bus 2,500 Instant 11 kv Ring HD sw/s present value (AED m) 2,000 1,500 Higher service cost Instant 11 kv Ring HD sw/s 55 MVA PRY Instant 11 kv Mesh Instant 11 kv Mesh dbl bus Instant 20 kv ring sw/s Instant 20 kv ring dbl bus 1,000 500 - Higher service quality 0 5 10 15 20 25 30 35 SAIDI in 2015 complete 11 kv Mesh complete 11 kv Mesh dbl bus complete 20 kv ring sw/s complete 20 kv ring dbl bus Figure 5: Base Case Trade-off Curve Service Quality (SAIDI) vs Service Cost (million AED) Trade-off Results High Density - Abu Dhabi Island - Higher Growth 04 13 04 4,500 Phased 11 kv Mesh 4,000 Phased 11 kv Mesh dbl bus Phased 20 kv ring sw/s 3,500 Phased 20 kv ring dbl bus AED m 3,000 2,500 2,000 Higher service cost Instant 11 kv Ring HD sw/s Instant 11 kv Ring HD sw/s 55 MVA PRY Instant 11 kv Mesh Instant 11 kv Mesh dbl bus Instant 20 kv ring sw/s 1,500 1,000 500 Higher service quality Instant 20 kv ring dbl bus complete 11 kv Mesh complete 11 kv Mesh dbl bus - 0 5 10 15 20 25 30 35 SAIDI Figure 6a: Sensitivity Results for High Growth Case complete 20 kv ring sw/s complete 20 kv ring dbl bus 9
Service Quality (SAIDI) vs Service Cost (million AED) Trade-off Results High Density - Abu Dhabi Island - Lower Growth 04 13 04 3,000 Phased 11 kv Mesh Phased 11 kv Mesh dbl bus 2,500 Phased 20 kv ring sw/s Phased 20 kv ring dbl bus AED m 2,000 1,500 Higher service cost Instant 11 kv Ring HD sw/s Instant 11 kv Ring HD sw/s 55 MVA PRY Instant 11 kv Mesh Instant 11 kv Mesh dbl bus Instant 20 kv ring sw/s 1,000 Instant 20 kv ring dbl bus 500 Higher service quality complete 11 kv Mesh complete 11 kv Mesh dbl bus - 0 5 10 15 20 25 30 35 40 SAIDI Figure 6b: Sensitivity Results for Low Growth Case complete 20 kv ring sw/s complete 20 kv ring dbl bus 10