Forecasting and Verifying the Energy Savings for Web-Enabled Thermostats in Portable Classrooms: A Spreadsheet M&V Tool Developed for BPA William E. Koran, P.E. Quantum Energy Services and Technologies Mira Vowles, P.E. Bonneville Power Administration Contents Need for this tool IPMVP Option C Tool Introduction and Demo Forecasting Statistics and Uncertainty Potential Enhancements Comments, questions, additional ideas for enhancements Need for this Tool Measurement and Verification Definition M&V is the process of using measurement to reliably determine actual savings. Verification of the potential to generate savings should not be confused with M&V. Verification of the potential to generate savings does not adhere to IPMVP since no site energy measurement is required. The intent of this tool is to provide true M&V. Visualization of Savings Chart is similar to IPMVP Figure 1, Example Energy History 700 Actual Baseline Data Baseline Actual Post Data 600 Post Modeled Baseline 500 400 300 200 100 8/31/2010 7/1/2010 5/1/2010 3/1/2010 12/30/2009 10/31/2009 8/31/2009 7/1/2009 5/1/2009 3/1/2009 12/30/2008 10/30/2008 8/31/2008 7/1/2008 5/1/2008 3/1/2008 12/31/2007 10/31/2007 0 9/1/2007 Average kWh per day during billing period Electricity Use History and Adjusted Baseline IPMVP Savings Reporting Options Reporting Period Basis (“Avoided Energy Use”) • Baseline is Projected to Reporting Period Conditions • Avoided Energy Use = Projected Baseline Energy Use minus Actual Reporting Period Energy Use Fixed Conditions Basis (“Normalized Savings”) • Baseline and Post period energy use are Projected to a set of fixed conditions • Normalized Savings = Projected Baseline Energy Use minus Projected Post Energy Use IPMVP Option C Whole Facility Savings are determined by measuring energy use at the whole facility level. Most commonly, utility meter data is used for the energy use measurement. Routine adjustments are required, such as adjustments for weather conditions that differ between pre-and post. Routine adjustments are often made using regression analysis Approach Taken by this Tool This Tool Uses a Fixed Conditions Basis. The Energy Use is projected for a typical year, using TMY3 weather data. Routine adjustments are made using regression analysis Tool Introduction: Worksheets Instructions User Interaction • BillingData • WthrQuery • WthrData Outputs • ForecastSavings • VerifiedSavings Background Calculations • PastProjectsData • Calcs • RegressionBase • RegressionPost Tool Introduction: Calculation Approach Based on ASHRAE Guideline 14-2002 Measurement of Energy & Demand Savings, Annex D, Regression Techniques Independent Variable • Average Heating Degree-Hours per Day during billing period (base 65 ºF) Dependent Variable • Average kWh per Day during billing period Tool Introduction: Weather Data Web Query of Hourly Temperatures for Nearest Site Heating Degree-Hours are Calculated for Each Billing Period, divided by 24, and divided by the number of days in the billing period. Tool Demo Forecasting Savings For Proposed Projects Weather-dependent load is assumed to have the same relationship (slope) as past projects. Non-weather-dependent load is assumed to be proportional to number of scheduled hours. Uncertainty • uncertainty in the baseline regression • uncertainty in the post regression from past projects • uncertainty due to variation in the past projects. Statistics and Uncertainty International Performance Measurement and Verification Protocol, Volume 1, 2009. ASHRAE Guideline 14-2002, Measurement of Energy and Demand Savings, 2002, Annex B. CCC: Guidelines for Verifying Existing Building Commissioning Project Savings, Using Interval Data Energy Models: IPMVP Options B and C, 2008. National Institute of Standards and Technology. The NIST Engineering Statistics Handbook, http://www.itl.nist.gov/div898/handbook/index.htm Statistics and Uncertainty BPA Regression Reference Guide (in revision) Sections of Particular Relevance: • Requirements for Regression • Validating Models Statistical Tests for the Model Statistical Tests for the Model’s Coefficients Additional Tests Plus, Tables of Statistical Measures Statistics and Uncertainty T-statistic • The t-statistic is a measure of the statistical significance of a model’s coefficient. If it is greater than the comparison “critical” t-statistic, the coefficient is significant. • Critical t-statistics are a function of the required (input) confidence level and the number of data points. For 24 data points, and a 90% confidence level, the critical t-statistic is 1.72 Statistics and Uncertainty Confidence Intervals • Confidence intervals are a measure of the uncertainty of the regression line. • The uncertainty in the savings is dependent on the regression uncertainty. • The confidence intervals are a function of the t-statistic. Verified Savings Uncertainty Meter data measurement uncertainty is assumed to be zero. Uncertainty of baseline and post regressions are included. Uncertainty associated with the appropriateness TMY3 data is not included. Potential Enhancements Use a weighted regression. Adjust the regression for summer occupancy. Limit baseline to whole years. Input project start and end dates (use 2 dates). Use Heating Degree-Hours for Forecast Savings as well as Verified Savings. Use variable-base heating degree-hours. Adjust heating degree-hours for the occupancy schedule. Incorporate more completed projects in the forecasting. Protect cell formatting. Allow multiple weather sites in WthrData Add capability to benefit from interval meter data Comments and Questions Thank You Bill Koran Quantum Energy Services & Technologies 503-557-7828 wkoran@quest-world.com Mira Vowles Bonneville Power Administration 503-230-4796 mkvowles@bpa.gov Statistics and Uncertainty Normalized Demand, Watts per Square Foot 8 We are 80% confident that the true regression falls between these lines. 7 6 We are 95% confident that an individual point will fall between these lines. 5 4 3 Data Upper Confidence Line, 95% Confidence Level Lower Confidence Line, 95% Confidence Level 2 We are 95% confident that the true regression falls between these lines. 1 Upper Confidence Line, 80% Confidence Level Lower Confidence Line, 80% Confidence Level Upper Prediction Line, 95% Confidence Level Lower Prediction Line, 95% Confidence Level Linear (Data) 0 20 30 40 50 60 70 Ambient Temperature, ºF 80 90 100 Statistics and Uncertainty p-value • The p-value is the probability that a coefficient or independent variable is not significantly related to the dependent variable. • Rather than requiring an input confidence level as for the t-statistic, the p-value provides probability as an output.