1. Settlement Accuracy Analysis Prepared by ERCOT Load Profiling

2. Introduction and Methodology

3. 3 Settlement Accuracy Analysis Advanced metering can be used to enable 15-minute interval settlement for all ESIIDs. Alternatively, advanced metering can be used to:  Enable 15-minute interval settlement for some ESIIDs, and  Improve the accuracy of load profiling and settlement for all other ESIIDs The intent of this presentation is to address the use of advanced meters to improve load profiling by:  Increasing the frequency of meter readings  Creating the feasibility for larger more frequently updated load research samples  Facilitating the introduction of new load profiles  Enabling dynamic (true/lagged) profiling

4. 4 Meter Reading Frequency Currently monthly meter reads are used in settlement and are profiled using adjusted static models  Load profile models are structured in 3 stages Stage 1: Monthly kWh - Daily kWh Stage 2: Daily kWh - Hourly kWh Stage 3: Hourly kWh – 15-minute interval kWh Increasing the meter reading frequency can eliminate one or more stages from the profile model and the estimation error associated with the stages Universal TOU meter reading (independent of pricing) could leverage the “chunking” functionality already existing in ERCOT systems to further reduce profiling error  TOU meter readings would individualize the profile shape to capture systematic difference in usage patterns across customers

5. 5 ERCOT utilized the interval data from the Load Research Sample customers to investigate the profiling error impact associated with the levels of meter reading frequency. ERCOT’s round 1 load research sample was approximately one-half the size anticipated to be needed The models developed from the load research data are not as accurate as would be estimated from larger samples Meter Reading Frequency

6. 6 ERCOT utilized the interval data from the Load Research Sample customers to investigate the impact of several levels of meter reading frequency on profiling error. Analysis window: July 2005 - June 2006 The meter reading frequency levels analyzed are listed below: Time Period Approx. # of Reads per month Monthly (calendar month) 1 TOU Monthly 8 Daily 28 TOU Daily 120 Hourly 720 Meter Reading Frequency

7. 7 Interval data for each of the sample customers was summed over the appropriate time periods to determine non-IDR meter readings for the various reading frequency levels The non-IDR meter readings were then profiled following the currently established process The actual interval values and the profiled versions of those intervals were then extrapolated to the profile class level using standard load research statistical methodology The difference between the actual and profiled class level estimates is a measure of the amount of profiling error (at the class level) associated with the various meter reading frequency levels Meter Reading Frequency

8. 8 For this analysis, profile specific TOU schedules were created to isolate periods with significant variation in load shape and high volume of consumption

9. 9 Example of Actual vs Profiled – Monthly Reads BUSMEDLF – COAST - July 14, 2005

10. 10 Example of Actual vs Profiled – Monthly TOU Reads BUSMEDLF – COAST - July 14, 2005

11. 11 Example of Actual vs Profiled – Daily Reads BUSMEDLF – COAST - July 14, 2005

12. 12 Example of Actual vs Profiled – Daily TOU Reads BUSMEDLF – COAST - July 14, 2005

13. 13 Example of Actual vs Profiled – Hourly Reads BUSMEDLF – COAST - July 14, 2005

14. Summary Reports - kWh

15. 15 Annual Average Interval kWh Difference Profile – Actual Profiling spreads kWh from meter readings across intervals but does not change the total kWh use. Consequently, the average difference between actual and profiled kWh is zero; differences shown above are attributable to rounding.

16. 16 Annual Average Interval Percent Difference Profile – Actual Increased meter reading frequency results in improved annual average percent differences.  Across all profile types Busnodem improves the most – 1.38 % improvement from monthly to hourly reads  Residential profile types improve by 1.12% from monthly to hourly reads All percent differences are positive – overestimation of low load intervals underestimation of high load intervals

17. 17 Meter Reading Frequency All percent differences are positive – overestimation of low load intervals underestimation of high load intervals  An inherent limitation of adjusted static models because the estimation of model coefficients is driven by the preponderance of medium load intervals  The load profiling process allocates kWh from meter readings across all intervals in the period; overestimated intervals must be offset by underestimated intervals The following slides illustrate the over/under estimation for selected profile models

18. 18 Average Day Across the Study Period RESHIWR - COAST underestimating overestimating

19. 19 Average Day Across the Study Period RESHIWR - NCENT underestimating overestimating

20. 20 Average Day Across the Study Period BUSMEDLF - COAST underestimating overestimating

21. 21 Average Day Across the Study Period BUSMEDLF - NCENT underestimating overestimating

22. 22 Distribution of Interval Percent Differences BUSMEDLF Percent Difference (Profile – Actual) Increasing meter reading frequency results in tighter distribution of differences

23. 23 Percent Difference (Profile – Actual) Distribution of Interval Percent Differences RESHIWR Increasing meter reading frequency results in tighter distribution of differences

24. Summary Reports - Dollars

25. 25 Annual Average Absolute Interval Percent Difference Profile – Actual Increased meter reading frequency results in improved annual average absolute interval percent differences.  Buslolf, Busnodem, Reshiwr, and Reslowr all improve by about 5 % going from monthly to hourly reads  Bushilf and Busmedlf improve by 1.6% and 2.8% respectively

26. 26 Annual Total Dollar Difference Profile – Actual Annual Dollars = ∑ i kWh i * MCPE i Annual total dollar differences are relatively small even with monthly reads The differences range from $0.39 for Busnodem up to $21 for Bushilf per month Residential profiles account for ~ 60% of annual dollars and have a difference of about $1.70 per month Increased meter reading frequency results in lower annual total dollar differences. In general dollar differences are negative Overestimation of low load intervals and underestimation of high load intervals Positive correlation between load and MCPE

27. 27 Average Day Across the Study Period RESHIWR - COAST

28. 28 Average Day Across the Study Period RESHIWR - NCENT

29. 29 Average Day Across the Study Period BUSMEDLF - COAST

30. 30 Average Day Across the Study Period BUSMEDLF - NCENT

31. Selected Weekly Profiles BUSMEDLF_NCENT RESHIWR_NCENT

32. 32 Summer Week – August 15-21, 2005 BUSMEDLF - NCENT

33. 33 Summer Weekday – Thursday, August 18, 2005 BUSMEDLF - NCENT

34. 34 Summer Weekend – August 20-21, 2005 BUSMEDLF - NCENT

35. 35 Fall Week – October 17-23, 2005 BUSMEDLF - NCENT

36. 36 Fall Weekday – Friday, October 21, 2005 BUSMEDLF - NCENT

37. 37 Fall Weekend – October 22-23, 2005 BUSMEDLF - NCENT

38. 38 Winter Week – February 13-19, 2006 BUSMEDLF - NCENT

39. 39 Winter Weekday – Friday, February 17, 2006 BUSMEDLF - NCENT

40. 40 Winter Weekend – February 18-19, 2006 BUSMEDLF - NCENT

41. 41 Spring Week – March 27 - April 2, 2006 BUSMEDLF - NCENT

42. 42 Spring Weekday – Tuesday, March 28, 2006 BUSMEDLF - NCENT

43. 43 Spring Weekend – April 1-2, 2006 BUSMEDLF - NCENT

44. 44 Summer Week – August 15-21, 2005 RESHIWR - NCENT

45. 45 Summer Weekday – Monday, August 15, 2005 RESHIWR - NCENT

46. 46 Summer Weekend – August 20-21, 2005 RESHIWR - NCENT

47. 47 Fall Week – October 17-23, 2005 RESHIWR - NCENT

48. 48 Fall Weekdays – Thursday-Friday, October 20-21, 2005 RESHIWR - NCENT

49. 49 Fall Weekend – October 22-23, 2005 RESHIWR - NCENT

50. 50 Winter Week – February 13-19, 2006 RESHIWR - NCENT

51. 51 Winter Weekdays – Thursday-Friday, February 16-17, 2006 RESHIWR - NCENT

52. 52 Winter Weekend – February 18-19, 2006 RESHIWR - NCENT

53. 53 Spring Week – March 27 - April 2, 2006 RESHIWR - NCENT

54. 54 Spring Weekday – Tuesday, March 28, 2006 RESHIWR - NCENT

55. 55 Spring Weekend – April 1-2, 2006 RESHIWR - NCENT

56. Extreme Events - kWh

57. 57 Hurricane Rita – September 21-23, 2005 BUSHILF - COAST

58. 58 Hurricane Rita – September 21-23, 2005 BUSMEDLF - COAST

59. 59 Hurricane Rita – September 21-23, 2005 BUSLOLF - COAST

60. 60 Hurricane Rita – September 21-23, 2005 BUSNODEM - COAST

61. 61 Hurricane Rita – September 21-23, 2005 RESHIWR - COAST

62. 62 Hurricane Rita – September 21-23, 2005 RESLOWR - COAST

63. 63 Extreme Temperature Event – April 17, 2006 BUSHILF - COAST

64. 64 Extreme Temperature Event – April 17, 2006 BUSHILF - NCENT

65. 65 Extreme Temperature Event – April 17, 2006 BUSMEDLF - COAST

66. 66 Extreme Temperature Event – April 17, 2006 BUSMEDLF - NCENT

67. 67 Extreme Temperature Event – April 17, 2006 RESHIWR - COAST

68. 68 Extreme Temperature Event – April 17, 2006 RESHIWR - NCENT

69. 69 Summary and Conclusions For large aggregations of customers: Profiling provides a good estimate of load even with monthly meter reads Annual total dollar differences are relatively small with monthly reads Hourly and daily meter reads improves profile estimates, especially during unusual or extreme weather events. Settlement with 15-minute interval load data provides marginal improvement over what could be achieved using profiling with hourly or daily meter reads.

70. 70 Summary and Conclusions Opportunities for Profiling Improvement using advanced metering capabilities: Significantly larger and more frequently updated load research samples are viable considerations Introduction of new load profiles become faster and less costly Dynamic (true/lagged) profiling would be facilitated  To replace adjusted static models for standard profiles  Demand Response Programs  New profiles