Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
NEEA DEI Study Analysis Plan August 9, 2005 RLW Analytics, Inc. Roger L. Wright, Chairman, and Principal Consultant Outline Review our Clatskanie Substation Analysis Highlight issues in future analysis of CVR substation pilots Review HVR status Discuss plans for analysis of HVR studies Clatskanie Substation Analysis We used the Clatskanie data to test our analysis methodology We did not have information on the control status each day Our first attempt was to regress kWh on voltage, • Was not successful • Problem traced to simultaneity of relationship between voltage and kW Developed an algorithm to classify each day as a control or comparison day This gave more plausible results – but the data are still preliminary Initial Analysis Initial model: ln(kWh) = β0 + β1 ln(V ) where V = Voltage Equivalent to assuming a 1% drop in voltage yields a β1 drop in kWh The Observed Data Initial Results We were hoping for positive betas! Simultaneity of Voltage and kWh CVR effect: A drop in voltage is expected to yield a drop in kWh => + association. Load effect: An increase in kWh may cause the voltage to fall => - association A simple regression of kWh on voltage will reflect both effects and give an erroneous estimate of the CVR effect. Remedy Let C = voltage control status, 0 = off or 1 = on Or C = quantitative level of control variable Record the control status day by day and hour by hour Study the effect of control status on both kWh and Voltage Identifying the Control Status Control alternating off and on No clear control Energy Print of Control Status The energy print of voltage revealed periods of good control, periods of poor control, and periods of missing data Classification of Control Status When the circuit was in control the step function was set to 118; otherwise 122 Used to validate the classification visually Actual Voltage Control Indicator Verification of Control Status Effect of Control on Voltage (mnv) Figure 1: Change in Average Voltage Effect of Control on kWh Figure 2: Change in Average kWh β (Beta)= ΔkWh/ΔMNV For Phase A Feeder A - Divide the - 4% change in kWh by the - 3.2% change in MNV to obtain a Beta of 1.2 Across All Feeders and Phases - Divide the - .5% change in kWh by the - 3.1% change in MNV to obtain a Beta of .2 Estimated Beta by Feeder and Phase Erratic 2 Stable 1 0 E.C E.B E.A D.C D.B D.A C.C C.B C.A BB.C BB.B BB.A B.C B.B B.A A2.A A.C A.B A.A -1 Figure 3: Beta, the Change in kWh for a 1% Change in Voltage Impact By Season Summer Smaller Loads Negligible Cooling Loads Loads are mostly Lights and Plugs Winter Heating load increases the overall load Voltage control expected to have little or no effect on Electric Heating Voltage Control, therefore should have Modest Effect on Lights and Plugs Smaller percentage effect in winter than summer Figures 5 and 6 Summarize results for the Winter period Overall Beta was only 0.1 Figures 7 and 8 Summarize results for the Summer period Overall Beta was 0.3 Figure 5: Winter Change in Average Voltage Figure 6: Winter Change in Average kWh Figure 7: Summer Change in Average Voltage Figure 8: Summer Change in Average kWh Effect of Temperature Fit a regression model of the form kWh = β0 + β1 C + β2 T + ε kWh – Observed Energy Use of the feeder and phase in any hour of any Control period C – Indicator Variable that is equal to 1 if control was on in the hour, 0 otherwise T – Heating degrees If temperature < 650 then T = 650 – temperature T = 0 otherwise Interpretation of the Coefficients β0 = Least Squares Estimate of the expected kWh use in an hour with Control Off and with 0 Heating Degrees β1 = Least Squares Estimate of the change in kWh use in an hour with Control On vs. Control Off β2 = Least Squares Estimate of the change in kWh use in an hour per unit increase in heating degrees Figures 9 and 10 Separate Winter and Summer regressions for each combination of feeder, and phase kWh_off = Estimated value of β0 del_kWh = Estimated value of β1 pct_kWh = del_kWh/kWh_off Finally, used change in voltage from Figures 5 and 7 to calculate the Beta as pct_kWh / pct_MNV Winter results Figure 9: Winter Change in Average kWh Summer results Figure 10: Summer Change in Average kWh Figures 9 and 10 Support the hypothesis that voltage control has little or no effect on the heating component of the feeder load Indicate that the average value of Beta is about 0.3 in both the winter and summer, once the heating load has been excluded A 1% reduction in voltage appears to reduce the non-heating kWh load on the feeder by 0.3% on average across these feeders regardless of the season Effect By Hour Repeat this analysis for each hour of the day, from 1 to 24 For each combination of feeder, and phase, and each of the 24 hours, estimate a separate regression model of the form kWh = β0 + β1 C + β2 T + ε Combined Winter and Summer seasons into a single regression – as model captured effect of winter heating Hourly results for Feeder A Phase A Figure 11: Hourly Load Profile of Base Load with Voltage Control Off (0) and On (1), Feeder A Phase A Average hourly results Figure 13: Hourly Load Profile of Base Load with Voltage Control Off (0) and On (1) Average of all Feeders and Phases Figure 13 Provides graphs of average non-heating hourly load profile of all combinations of feeder and phase with and without voltage control Voltage regulation has on average a very small effect Effect is most consistent in the early morning hours when the load is smallest During peak load effect is negligible Lessons Learned from Clatskanie The importance of clean voltage and kwh data and accurate information about the status of experimental control Naive regression analysis can lead to biased findings Beta seems to vary by end use and season Careful regression analysis can ferret out effects (betas) by season or end use Unresolved Question How do the three phases of a feeder interact? Is it best to analyze each phase separately or can they be combined? HVR Studies - Objectives Estimate the customer-side portion of the CVR effect Help estimate how the CVR effect varies with end use Help adapt the findings to various utiities and service areas Targeted End Use Categories Effects shown are from prior BPA end use study We want to estimate the betas for these four end use categories Approach Install HVR devices in a stratified sample of homes to control the voltage (off or on) on a known schedule Do an onsite audit of each sample home Collect whole-house load data on hourly kWh and voltage Analyze the resulting data much like substation data, but rolling in the end use information to estimate the end-use effects Foundations for the HVR Analysis β = ∆ kWh / kWh Total House kWh = Sum of kWh by End Use, i.e. kWh = Σ kWhEU Similarly ∆ kWh = Σ ∆ kWhEU where ∆ kWhEU = βEU kWhEU So β = ∆ kWh / kWh = Σ βEU (kWhEU / kWh) Approach A) Analyze each home’s data to estimate 1. the overall β of the house 2. The end use energy share of the house kWhEU / kWh for each of the four end uses B) Regress the overall β on the four end use energy shares to estimate the four end-use betas Results will be developed by Market segments: 1. 2. 3. 4. Western region, all electric Western region with gas service Eastern region all electric Eastern region with gas service Measures of energy and demand: 1. Annual kWh 2. Seasonal kWh 3. Class peak kW The Keys to Success Reliable estimates of the whole-house betas for most of the sample homes. Accurate estimates of the end use energy shares. Substantial variation in the end use energy shares from home to home in the sample Whole-house Betas Our Clatskanie analysis indicates that we must have accurate information on HVR control status Each house can be on a different control schedule, but we must know the schedule End-use Energy Shares We will integrate the information from the onsite audits and whole premise load data Space heating, water heating, and AC have recognizable energy prints Must rely on the audits for – Resistance space heating vs. heat pumps – Incandescent vs fluorescent lamps Other plug loads will generally not be identifiable Probably will have to settle for annual or seasonal end use shares but not hourly Variation in End-use Shares from Home to Home Expect variation due to availability of natural gas, vintage of home, climate zone and service area Will need to combine all sample homes across utilities Can hope to borrow strength using seasonal analysis Concerns Limited time and money for the analysis Uncharted territory CVR effects are relatively small and hard to detect May depend on severity of weather during study period