AEIC Annual Load Research Conference September 12, 2006 - Reno, NV ERCOT Residential Profile ID Assignments – Dealing with Assignment Accuracy and Migration Presented By: Diana Ott Carl Raish 1 Overview • ERCOT Settlement highlights • Residential Annual Validation • Heating Fuel Type Residential Survey • Impact of Miss-Assignment of Residential Load Profile ID Assignment • New Residential Algorithm • Q&A 2 Settlement • ERCOT requires a fifteen (15) minute settlement interval • Vast majority of Customers do not have this level of granularity. • Profiles are created using adjusted static models • Models are dependent on season, day of week, time of day and weather • Backcasted Profiles are generated the day following a trade day and used for all settlements (initial, final and true-up) • Load Profiling: • Converts monthly NIDR reads to fifteen (15) minute intervals • Enables the accounting of energy usage in settlements • Allows the participation of these Customers in the retail market (reduces barrier to entry) 3 3 Load Profile Groups, 9 Segments Residential (2) Non-Metered (2) Business (5) Low-Winter Ratio (Non-electric Heat) High-Winter Ratio (Electric Heat) Low Load Factor Medium Load Factor High Load Factor Non-Demand IDR Default Lighting (Street Lights) Flat (Traffic Signals) 4 8 ERCOT Weather Zones Stars represent the location for the 20 ERCOT Weather Stations for each Weather Zone 5 Annual Validation of Profile Assignments • ERCOT in conjunction with Profiling Working Group establishes the rules for Profile ID assignment and publishes in the form of a Decision Tree on the ERCOT website • Annual Validation is a process established by the Market to annually review and update Profile ID assignments based on the rules defined in the Decision Tree • Historically, May 1 thru April 31 meter reads were used to determine the Annual Validation assignment. The process normally began in June and completed in January. 6 History of Residential Annual Validation • Oct. 2001 Initial Validation • • • 2002 Annual Validation • • • Large volume of migrations (1.5 million out of 4.9 million ESIIDs) 2004 Annual Validation • • • Not performed due to 2001 Initial Validation still in progress PWG sub team changed methodology from using billing month to usage month 2003 Annual Validation • • Profile IDs were assigned by TDSPs prior to Market Open Validation started in 2001 and was not completed until Sept. 2002 Large volumes of changes were identified (1.0 million out of 5.4 million ESIIDs) Annual Validation suspended to allow time to improve assignment process 2005 Annual Validation • • Some methodology changes were identified which still resulted in large volumes of migrations (0.5 million out of 5.1 million ESIIDs) Market delayed sending in transactions and ultimately decided to only send in a subset of changes identified 7 Residential Assignment Rules 2001 - 2004 Winter Ratio >=1.5 RESHIWR Winter Ratio < 1.5 RESLOWR Max ( ADUsedec , ADUse jan , ADUse feb ) WR * Average ( Fallbase , Springbase ) * * Round to two decimal places Where ADUsedec = Average Daily Use in the December Usage Month, ADUsejan = Average Daily Use in the January Usage Month, ADUsefeb = Average Daily Use in the February Usage Month, FallBase = minimum ADUse for the Usage Months of October and November SpringBase = minimum ADUse for the Usage Months of March and April. 8 Preliminary Residential Assignment Rules for Annual Validation 2005 • Do not replace a non-default assignment with a default assignment • Apply Dead-Bands • RESHIWR goes to RESLOWR if WR ≤ 1.0 • RESLOWR goes to RESHIWR if WR > 1.8 • Dead-Bands do not apply if currently a default assignment • kWh Minimums • WR numerator ≤ 20 then assign RESLOWR 9 Additional Profile Assignment Improvement Ideas • Use a statistical approach to correlate premise usage to profile usage. • Use a residential survey to obtain the necessary data to relate usage patterns to heating system type. • • More accurately account for weather variations Account for periods of low/no occupancy • Move calculation responsibility to ERCOT from TDSPs • Change time period for submission of assignment change transactions • • During the original October/November timeframe for submitting changes, the RESHIWR and RESLOWR profiles are significantly different RESHIWR and RESLOWR profiles are quite similar during the summer months 10 Residential Heating & Fuel Type Survey • Design: • • 41,000 bilingual survey forms mailed Stratified by Weather Zone and Profile Type • • • 2,563 RESHIWR per Weather Zone 2,562 RESLOWR per Weather Zone Response • Survey responses were identified to allow connecting the response to usage history • 4,669 responses as of 09/30/2005 • 11.4% response rate 11 Questions from Residential Survey pertinent to Electric Heat Analysis What classification best fits this address? (Check only one box) • Single-Family Dwelling • Multi-Family Dwelling (Duplex, Apartments, etc.) • Other (Please skip the remaining questions and disregard the survey.) What is the primary type of home heating used at this residence? (Check only one box) • Electricity • Natural gas or bottled gas (propane/butane) • Other or not sure • • • • Have you added central electric cooling or heating in the last 2 years? (Check all that apply) Yes, I added central air conditioning Yes, I added central electric heating No Not sure • • • • • What is the approximate age of your residence? (Check only one box) Less than 5 years 5 – 15 years 16 – 30 years More than 30 years Not sure 12 Residential Heating & Fuel Type Survey Number of Survey Responses by WZone & Profile Type 700 600 350 333 289 500 HIWR 252 231 HIWR 400 HIWR HIWR 176 HIWR 205 268 HIWR HIWR HIWR 300 200 355 262 100 LOWR LOWR 365 309 303 LOWR LOWR LOWR 325 LOWR 371 275 LOWR LOWR 0 Return rate by WZone COAST EAST FWEST NCENT NORTH SCENT SOUTH WEST 8.5% 12.6% 10.5% 9.9% 13.6% 11.3% 10.6% 14.1% 13 Residential Heating & Fuel Type Survey Overall - Percent of Home Heat Type (RESLOWR & RESHIWR) 100% 1.45% 0.35% 0.80% 0.98% 1.02% 0.33% 1.19% 1.18% 1.14% 52.2% 52.0% 46.7% 46.9% WEST ERCOT 90% 29.2% 80% 47.5% 70% 47.7% 57.3% 62.7% 62.5% 54.8% 60% 50% 40% 69.6% 30% 52.2% 20% 51.4% 41.6% 36.5% 36.0% 44.9% 10% 0% COAST Electricity EAST FWEST NCENT NORTH SCENT Natural/Bottled Gas SOUTH Other/Not Sure 14 Residential Heating & Fuel Type Survey RESLOWR Respondents - Percent of Home Heat Type 100% 1.5% 0.6% 1.0% 1.0% 1.1% 0.3% 1.5% 1.3% RESHIWR Respondents - Percent of Home Heat Type 1.2% 90% 90% 35% 80% 68% 74% 78% 68% 0.4% 1.0% 11% 13% 0.9% 18% 0.4% 0.4% 13% 11% 86% 89% SCENT SOUTH 0.9% 20% 0.9% 14% 29% 69% 80% 60% 50% 50% 40% 40% 81% 88% 87% 81% 79% 85% 71% 63% 30% 30% 20% 31% 10% 18% 0.0% 70% 69% 73% 1.1% 80% 70% 60% 100% 25% 21% 31% 26% 30% 30% 19% 10% 0% COAST EAST FWEST NCENT NORTH 20% SCENT SOUTH WEST ERCOT 0% COAST Overall 30% misclassified Electricity Natural/Bottled Gas EAST FWEST NCENT NORTH WEST ERCOT Overall 14% misclassified Other/Not Sure 15 Residential Heating & Fuel Type Survey ERCOT 2005 Residential Survey Results Added Central Electric Cooling or Heating (last 2 years) RESLOWR COAST EAST FWEST NCENT NORTH SCENT SOUTH WEST ERCOT Cooling 1.9% 4.8% 5.8% 3.0% 4.4% 1.8% 8.0% 5.1% 3.3% Heating 0.0% 2.0% 2.6% 1.0% 1.6% 0.6% 3.6% 2.2% 1.0% Both 0.0% 1.7% 1.9% 1.0% 1.6% 0.3% 3.3% 1.6% 0.9% No 96.2% 93.2% 91.6% 93.4% 93.2% 94.2% 88.4% 91.6% 93.9% Not Sure 1.1% 0.6% 0.6% 1.7% 0.3% 1.8% 1.5% 0.8% 1.3% No 93.8% 94.8% 90.5% 90.2% 94.9% 92.5% 91.0% 92.3% 91.6% Not Sure 1.7% 1.4% 2.2% 2.4% 0.6% 2.4% 1.1% 1.1% 2.0% Survey indicates low rate of heating system type change RESHIWR COAST EAST FWEST NCENT NORTH SCENT SOUTH WEST ERCOT Cooling 3.4% 2.4% 6.1% 6.3% 4.2% 4.4% 6.3% 5.4% 5.4% Heating 2.3% 1.7% 2.2% 2.9% 3.0% 2.8% 5.2% 2.9% 2.9% Both 1.7% 1.4% 1.7% 2.9% 2.4% 2.8% 4.5% 2.6% 2.6% 16 Residential Heating & Fuel Type Survey Primary Home Heat % for Homes Less than 5 years 100% 9.3% 90% 19.6% 23.8% 34.2% 38.9% 80% 44.2% 51.8% 70% 63.2% 60% 50% 83.8% 40% 80.4% 76.2% 65.8% 61.1% 30% 55.8% 48.2% 20% 36.8% 10% 6.9% 0% Coast East Fwest Other/Not Sure NCent North Electricity SCent South Natural/Bottled Gas West 17 Residential Heating & Fuel Type Survey • Performed visual inspection of usage patterns for each survey response • 4,630 responses indicated either a “Single-Family Dwelling” or “Multi-Family Dwelling” and a primary home heating type of either “Electricity” or “Natural gas or bottled gas (propane/butane) • 673 (14.5%) responses to the home heating type were deemed invalid by examination of their seasonal usage pattern • 3,957 (85%) responses were used to develop an improved Profile Type classification algorithm 18 Survey Response Validation - Electric Heat Example Note:Summer usage values omitted for clarity 19 What we found out from the Survey • Saturation of Electric Heat varied considerably across weather zones • Saturation of Electric Heat was inconsistent with breakdown between RESHIWR and RESLOWR • 30% of Survey responders reporting Electric Heat were assigned to RESLOWR • 14% of Survey responders reporting No Electric Heat were assigned to RESHIWR • There is very little year-to-year change in heating system fuel actually occurring • The percent of newer homes using electric heat varies considerably across weather zones (37% Coast – 84 South %) 20 Why Does Assignment Accuracy Matter? • Profile assignment errors create two types of load profile estimation errors • Assignment of billing kWh to the days within the billing period • (RESHIWR assigns more kWh than RESLOWR to cold days) • Assignment of daily kWh to the intervals within the day • (RESHIWR assigns more kWh to morning intervals) 21 Daily kWh as a Percent of Monthly kWh wz o n e = COA S T g r o u p = 2 0 0 4 / 0 5 Da i l y Pe r c e n t Reslowr 0. 08 mo n _ p c t _ l o Reshiwr mo n _ p c t _ h i 0. 07 Daily kWh Pct 0. 06 0. 05 0. 04 0. 03 0. 02 0. 01 0. 00 050104 060104 070104 080104 090104 100104 110104 usedat e Trade Day 120104 010105 020105 030105 040105 050105 22 Residential Profile Comparison - FWEST Reshiwr vs. Reslowr 2 0% -50% 1 -100% 0.5 -150% 0 -200% -0.5 -250% -1 -300% -1.5 -350% -2 -400% 12/10/05 12/11/05 12/4/05 12/5/05 12/6/05 12/7/05 Reslowr 12/8/05 Reshiwr 12/9/05 Pct_Diff % Diff kWh 1.5 23 Findings and Next Steps • ERCOT’s Profile ID Assignment process has resulted in unacceptably high migration rates • Dead - bands would reduce migration but could do more harm than good in terms of assignment accuracy • The impact of Profile ID miss-assignment is significant at the ESIID level • Undertake an effort to develop a new and improved assignment process with a goal of reducing migration and improving accuracy • More improvements are needed 24 Classification Algorithm Overview • Use Residential Survey response data in conjunction with responder usage data to build an algorithm to predict heating fuel • Use regression between actual meter readings for a premise and the RESHIWR and RESLOWR profile kWh for the same time periods • Use reads during shoulder and winter months for several (4.5) years • Omit reads during periods of very low use (no/low occupancy) • Omit outlier reads and require some reads to exceed a minimum kWh/day threshold in order to assign RESHIWR • Assign the better fitting profile to the ESIID 25 Classification Algorithm Development • For each ESI ID with a survey response usage values were selected from Lodestar for the January 2002 – September 2005 time period • Each usage value was converted into units of kWh/day and any read covering a period longer than 44 days was dropped • Each usage value was classified as a winter or shoulder reading • • Only shoulder and winter readings were used in the analysis • Winter/Shoulder: start > September 20 and stop < May 11 • Winter: start > November 15 and stop < March 15 • Shoulder: all others Usage values were screened for high and low outlier usage values 26 Classification Algorithm Development For each ESI ID compute a mean and standard deviation of the kWh/day values for the winter and shoulder readings and use these to “normalize” each usage value Z UsageValue Mean StDeviation Usage value dropped if: Z > 3 and kWh/day > 100 Z > 3.5 Outliers Z < -2 kWh/day < 5 Low Occupancy 27 Classification Algorithm Development Usage Screening Examples: Start Date Stop Date 3/4/2002 4/3/2002 4/3/2002 5/3/2002 10/30/2002 12/3/2002 3/4/2003 4/3/2003 4/3/2003 5/2/2003 10/1/2003 11/3/2003 11/3/2003 12/4/2003 3/4/2004 4/1/2004 4/1/2004 5/5/2004 9/30/2004 10/29/2004 10/29/2004 12/2/2004 3/4/2005 4/5/2005 4/5/2005 5/3/2005 1/4/2002 2/4/2002 2/4/2002 3/4/2002 12/3/2002 1/7/2003 1/7/2003 2/3/2003 2/3/2003 3/4/2003 12/4/2003 1/7/2004 1/7/2004 2/4/2004 2/4/2004 3/4/2004 12/2/2004 1/5/2005 1/5/2005 2/2/2005 2/2/2005 3/4/2005 mean standard deviation kWh 865 445 1,581 495 309 380 248 170 185 352 281 981 889 1,309 1,815 2,189 1,802 1,984 240 171 . 1,228 1,437 1,175 kWh/day 28.8 14.8 46.5 16.5 10.7 11.5 8.0 6.1 5.4 12.1 8.3 30.7 31.8 42.2 64.8 62.5 66.7 68.4 7.1 6.1 . 36.1 51.3 39.2 Z 0.02 -0.61 0.82 -0.53 -0.80 -0.76 -0.92 -1.00 -1.03 -0.73 -0.90 0.10 0.15 0.62 1.64 1.54 1.73 1.80 -0.96 -1.00 0.35 1.03 0.49 Dropped Dropped kWh kWh/day . . . . . . . . . . . . . . . . . . . . 132 . . . . . . . . . . . . . . . . . . . . . . . 4.6 . . . 28.3428 22.2221 Usage less than 5 kWh/day dropped 28 Classification Algorithm Development Usage Screening Examples: Start Date Stop Date 2/25/2002 3/25/2002 9/23/2002 10/22/2002 2/24/2003 3/26/2003 9/24/2003 10/22/2003 2/26/2004 3/25/2004 9/23/2004 10/22/2004 2/24/2005 3/24/2005 1/25/2002 11/20/2002 12/23/2002 1/24/2003 11/20/2003 12/26/2003 1/26/2004 11/22/2004 12/27/2004 1/26/2005 mean standard deviation 3/25/2002 4/25/2002 10/22/2002 11/20/2002 3/26/2003 4/25/2003 10/22/2003 . 11/20/2003 3/25/2004 4/23/2004 10/22/2004 11/22/2004 3/24/2005 4/26/2005 2/25/2002 12/23/2002 1/24/2003 2/24/2003 12/26/2003 1/26/2004 2/26/2004 12/27/2004 1/26/2005 2/24/2005 kWh kWh/day 1931 1570 1880 1784 2284 1710 68.964 50.645 64.828 61.517 76.133 57 . 1488 1561 1527 1791 1720 1424 1296 2492 2719 3349 2877 2526 2297 2676 2592 1980 1871 51.31 55.75 52.655 61.759 55.484 50.857 39.273 80.387 82.394 104.656 92.806 70.167 74.097 86.323 74.057 66 64.517 Z 0.22 -0.81 -0.02 -0.20 0.62 -0.46 -2.47 -0.78 -0.53 -0.70 -0.19 -0.54 -0.80 -1.46 0.86 0.97 2.23 1.56 0.28 0.51 1.19 0.50 0.05 -0.03 Dropped Dropped kWh kWh/day . . . . . . 595 . . . . . . . . . . . . . .. . . . . . . . . . 21.25 . . . . . . . . . . . . . . . . 65.1179 17.7623 29 Usage with Z < -2.00 dropped Classification Algorithm Development Usage Screening Results • 1,006 ESI IDs (21.7%) with one or more usage values screened • 2,414 usage values were screened out • 1,825 usage values screened out because < 5 kWh/day • If an ESI ID had fewer than 3 winter readings or fewer than 3 shoulder readings it was classified as “RESLOWD” (Residential Low Winter Ratio Default) and was not used for fine tuning the algorithm 30 Classification Algorithm Development Algorithm Basics • If an ESI ID has (and uses) electric heating, then the winter and shoulder usage values for that premise should be more similar to the RESHIWR profile kWh than to the RESLOWR profile kWh • The profile kWh for a day reflects the weather conditions associated with that day in the specific weather zone as well as the day type (day-of-week/holiday) and season of the year • To perform the comparison for an ESI ID, the profile kWh is summed across the intervals for the days in each of its meter reading periods (shoulder and winter months only) 31 Classification Algorithm Development Algorithm Basics • For each fall-winter-spring time period e.g., fall 2004 – spring 2005 the profile kWh is scaled to equal the sum of the ESI ID’s meter kWh for that time period • The correlation between the actual metered kWh and the scaled profile kWh for those readings is computed for each ESI ID • The R2 correlation is determined with a weighted linear regression analysis with no intercept term • Each reading is weighted as follows: Shoulder reading weight = 1 Winter reading weight = 2 RESHIWR kWh RESLOWR kWh Winter reading weight = 1 if RESHIWR kWh < RESLOWR kWh • The weighting process associates more importance with winter readings for which the RESHIWR kWh is greater than the RESLOWR kWh 32 New Algorithm Improvement Example ESIID Reshiwr Reslowr Note: New Algorithm improvement results from using multiple years of usage values 33 Fe b0 Ap 2 r- 0 Ju 2 nAu 0 2 g0 O 2 ct -0 De 2 c0 Fe 2 b0 Ap 3 r- 0 Ju 3 nAu 0 3 g0 O 3 ct -0 De 3 c0 Fe 3 b0 Ap 4 r- 0 Ju 4 n0 Au 4 g0 O 4 ct -0 De 4 c0 Fe 4 b0 Ap 5 r- 0 5 kWh Classification Algorithm Development Example ESIID Plotted 2,500 2,000 1,500 1,000 500 - Reading Mid-point ACTUAL SCALED RESHIWR SCALED RESLOWR 34 Classification Algorithm Development Algorithm - Classification Rules 1. If the highest winter reading kWh/day is less than 15 kWh/day then assign “RESLOWR” 2. If R2RESHIWR > 0.60 and R2RESHIWR > R2RESLOWR then assign “RESHIWR” 3. If the number of readings available > 9 and R2RESHIWR > 0.90 and (R2RESHIWR + 0.010) > R2RESLOWR and Winter Max kWh/day > 50 then assign “RESHIWR” 4. If the number of readings available > 9 and R2RESHIWR > 0.95 and (R2RESHIWR + 0.015) > R2RESLOWR and Winter Max kWh/day > 60 then assign “RESHIWR” 5. Otherwise assign “RESLOWR” 35 Classification Algorithm Development Algorithm – Rules Fine Tuning Algorithm fine tuning was an iterative process to tune each classification criterion on the previous slide individually Each classification criterion was adjusted to minimize misclassification error based on validated survey responses For each iteration, misclassified ESI IDs were examined graphically to assess the accuracy of the Profile Type assignment and to establish new criteria When the fine tuning was complete 184 (4.6%) validated survey responses regarding heating system type were different than the algorithm classification … most had usage patterns which were ambiguous 36 Survey and Algorithm Agree on Classification Reshiwr Reslowr Daily kWh ESIID 37 Survey and algorithm both indicated electric heat, “RESHIWR” Survey and Algorithm Disagree on Classification Reshiwr Reslowr Daily kWh ESIID 38 Survey said electric heat, algorithm said gas Classification Algorithm Results For the final version of the algorithm 3,773 (95.4%) validated survey responses regarding heating system type agreed with the algorithm classification Definitely not electric heat! 39 Applying Algorithm to Annual Validation 2005 62% of the 578,572 AV 2005 Profile Type changes agreed with the algorithm classification Changes to RESHIWR were significantly more accurate (78.4%) than changes to RESLOWR (43.5%) Accuracy of the changes by weather zone ranged from a low of 59.8% in the SOUTH zone to a high of 68.8% in the EAST zone The Residential population would have had somewhat more accurate Profile Type assignments as a result of conducting AV 2005 (81.4% vs. 78.7%) The market decided to allow only changes which were in agreement with the algorithm (358,000 changes were submitted) 40 Residential Changes for Annual Validation 2006 New algorithm adopted by market for AV2006, Calculation responsibility shifted to ERCOT Total Expected Changes 15.8% Break Down of Changes 18.1% HIWR to LOWR (156,798) 81.9% LOWR to HIWR (711,015) RESHIWR RESLOWR Current 1,789,799 3,685,490 5,475,289 Current % 32.7% 67.3% Expected Expected % 2,344,016 42.8% 3,131,273 57.2% 5,475,289 41 Estimates of Future Load Profile ID Migrations •Estimates below reflect migrations for 2006 if the new algorithm had been used exclusively for 2005 Annual Validation •The estimated migration rates are an indicator of what can be expected for year-to-year migration starting in 2007 •Estimates were developed from a sample of every 25th ESIID Year 1 to Year 2 Migration Estimates NonDefault Year to Year Migration Estimates 4.4% 3.6% 42 Conclusions The survey successfully provided data necessary to build a classification algorithm for electric heating and establish its accuracy. The classification algorithm at 96% accuracy was a significant improvement over the winter ratio method The improved accuracy will lead to assignment stability Profile assignments and shapes are in a feedback loop and improve each other The new algorithm uses load profile shapes to make profile assignments With updated load research analysis based on the new assignments, more accurate load profile shapes will be developed as a result of a more homogeneous population The more accurate load profile shapes should lead to better assignments ERCOT has completed load research analysis using the new profile assignments and is developing new profile models based on those latest estimates 43 44