USC Center for Energy Informatics cei.usc.edu Towards Modeling and Prediction of Energy Consumption for a Campus Micro-Grid Saima Aman
USC Center for Energy Informatics cei.usc.edu Energy Consumption USC has 250 buildings (2009), up by 10% from from 2001 Annual consumption of electricity in 2009 was up by 37% from 2001 Majority of electricity is consumed in buildings Modeling useful for planning and implementing university energy policies 9-Oct-152
USC Center for Energy Informatics cei.usc.edu Overview Goal: Models to predict daily energy use Trained using energy data for two years (2008 & 2009) Campus-Scale Energy Use Model Covers 170 buildings on UPC and HSC campuses Building-Scale Energy Use Model Covers 23 buildings on the UPC campus
USC Center for Energy Informatics cei.usc.edu Related Work Modeling methods: Regression models, artificial neural networks, time series models Data used in Models Static consumption data Live data streams Synthetic data using building simulation programs (e.g., Energy Plus) Estimation based on utility bills.
USC Center for Energy Informatics cei.usc.edu Related Work : Data Attributes Weather Data Temperature measurements (max, avg) Heating degree day, cooling degree day Building Data Orientation of buildings; windows Wall insulation thickness, heat transfer coefficient; window to wall ratio, etc. Occupancy data Estimate presence/absence and number of people Based on sensors in rooms, building entrances Based on heuristics, such as open/close office door
USC Center for Energy Informatics cei.usc.edu Unique Features of our Work Single unified building energy use model Applicable to diverse buildings; other works focus on homogeneous buildings Information driven approach Indirect indicators of energy use plus domain attributes; data is typically available publicly Design and Operation phase Use attributes that can be applied during design phase as well
USC Center for Energy Informatics cei.usc.edu FMS Energy Data 15-min interval energy data available for 3 years (from Jul 09, 2007 to Nov 21, 2010) Covers 170 buildings on the UPC and HSC campus Data: One CSV file for each day 24*4 = 96 records for each building per day Issues: Missing values and timestamps 9-Oct-157
USC Center for Energy Informatics cei.usc.edu Model Attributes ENRG* - Energy Use TMP* - Max Temperature Value AVTMP - Average Temperature Value GAREA - Gross Area NAREA - Net Area CYR - Year of construction (1919 – 2006) BTYP - Type of building (Academic, Residential, Other) WKDY* - Day of the week (M,T,…Su) HLDY* - Holiday (None, Academic Holiday, Campus Holiday) SEM* - Semester (Spring, Summer, Fall) (Sources: FMS, Academic calendar, Weather Underground) ( Attributes marked * are used in campus-scale model)
USC Center for Energy Informatics cei.usc.edu Daily consumption for 2009 & Oct-159 SpringSummer Fall
USC Center for Energy Informatics cei.usc.edu Academic Buildings (2009) 9-Oct-1510 EEB – Hughes (5F+B, sq. ft., 1990 ) RTH – Tutor Hall (6F+B, sq. ft., 2003)
USC Center for Energy Informatics cei.usc.edu Residential Buildings (2009) WTO – Webb Twr (14F+B, sq.ft., 1972) PRB – Parkside (4F+B, sq. ft., 2006) 9-Oct-1511
USC Center for Energy Informatics cei.usc.edu Campus-scale Model Training Data 731 records (for each day of the year 2008 & 2009) Test Data 325 records (for the year 2010, up to Nov 21) Tool: Statistics toolbox of MATLAB
USC Center for Energy Informatics cei.usc.edu Decision Tree 1. if TMP =74.5 then node 3 else if WKDY /in/ {6/7} then node 4 else if WKDY/in/ {1/2/3/4/5} then node 5 else if WKDY /in/ {6/7} then node 6 else if WKDY/in/ {1/2/3/4/5} then node 7 else if HLDY /in/ {1/2} then node 8 else if HLDY=0 then node 9 else if HLDY/in/ {1/2} then node 10 else if HLDY=0 then node 11 else
USC Center for Energy Informatics cei.usc.edu 9-Oct-1514
USC Center for Energy Informatics cei.usc.edu Campus-scale evaluation Model used to make prediction for year 2010 Evaluated using observed values CV-RMSE value = 7.45%. The predicted values are able to capture the weekly patterns of rise and fall of energy load.
USC Center for Energy Informatics cei.usc.edu Campus-scale Energy Prediction 9-Oct-1516 (For the year 2010)
USC Center for Energy Informatics cei.usc.edu Building-scale Model Training Data records Test Data Separate test dataset for each building Each has 325 records
USC Center for Energy Informatics cei.usc.edu Decision Tree 1. if NAREA< then node 2 elseif NAREA>= then node 3 else if CYR = then node 5 else if CYR =1931 then node 7 else if CYR = then node 9 else if WKDY in {6 7} then node 10 elseif WKDY in { } then node 11 else if WKDY in {6 7} then node 12 elseif WKDY in { } then node 13 else if GAREA< then node 14 elseif GAREA>= then node 15 else if GAREA< then node 16 elseif GAREA>= then node 17 else if GAREA< then node 18 elseif GAREA>= then node 19 else if WKDY=7 then node 20 elseif WKDY=6 then node 21 else
USC Center for Energy Informatics cei.usc.edu Building-scale evaluation
USC Center for Energy Informatics cei.usc.edu Building-scale Prediction ASC (CV-RMSE = 12.03%)
USC Center for Energy Informatics cei.usc.edu Building-scale Prediction EEB (CV-RMSE = 9.19%)
USC Center for Energy Informatics cei.usc.edu Future Work Include fine-grained information in our model 15-min granularity energy use data Detailed occupancy data (classroom assignments, course enrolment, room use)
USC Center for Energy Informatics cei.usc.edu Thanks
USC Center for Energy Informatics cei.usc.edu Academic Buildings (2009) SAL – Salvatori CS (3F, sq. ft., 1976) Comparison 9-Oct-1524
USC Center for Energy Informatics cei.usc.edu Residential Buildings (2009) PTD – Pardee Tower (8F, sq. ft., 1982) Comparison 9-Oct-1525