Data-Driven Energy Efficiency in Buildings Amir Kavousian, PhD Candidate Stanford University February 20, 2013 Photo: David Clanton
What is my research about? to demonstrate the power of data mining. [ ] results emerge from [ ] analysis of various databases, and asking the right questions. is a science with excellent tools for gaining answers but a serious shortage of interesting questions. 2
Data Driven Energy Efficiency GOOGLE POWERMETER STUDY 3
Google PowerMeter Home electricity Consumption monitoring 4
Energy Consumption Data 10 min interval smart meter data 1628 households From February 28, 2010 through October 23, 2010 5
Geographical Distribution of Households 6
114 Survey Questions Describe Household Characteristics External Factors Climate and Geography (6) Building Design and Construction Building Systems and Appliances Occupants Building (5) Home Improvements (12) Fuel Use (6) Appliances (14) Occupants (12) Energy Efficiency Habits (14) Payment items, method, estimate, feedback (6) How informed about appliance usage (5) Motivation Level (17) Effort to learn EE measures (7) Personal Info of the Survey Respondent (6) Thermostat Setpoint (4) 7
Data Driven Energy Efficiency EXPLORATORY ANALYSIS 8
Purpose: Quantify the impact of different factors on residential energy consumption 9
Analysis Method Forward stepwise regression model Ranks most important factors Estimates how much of the variation in energy consumption is due to each factor 10
Variability of Daily Maximum, Minimum and Average Consumption 11
Summer energy consumption is significantly affected by weather * Variance explained Average of cooling degree day values (38%) Nb of refrigerators (4%) Climate zone (3%) Type of Building (2%) Nb of freezers (2%) Nb of entertain t devices except TV s (2%) Thermostat setpoint habits (2%) Weather and Location House Appliances and Electronics Occupants * After normalizing for floor area 12
Summer daily maximum consumption is mostly affected by a number of high-consumption end uses Variance explained Average of cooling degree day values (31%) Elec. water heater (4%) Nb of elec. clothes dryers (4%) Nb of freezers (4%) Nb of refrigerators (4%) Nb of occupants (sqrt) (2%) Climate zone (2%) Weather and Location House Appliances and Electronics Occupants * After normalizing for floor area
Summer daily minimum consumption is mostly affected by less-volatile end uses and long-term habits Variance explained Average of cooling degree day values (26%) Nb of refrigerators (7%) Nb of entertain t devices except TV s(4%) Pet ownership (2%) Purchasing E Star Appliances (2%) Efficient thermostat settings (2%) Weather and Location House Appliances and Electronics Occupants * After normalizing for floor area
Summary of Findings: Effect of Location Location has a major role in determining energy consumption showed correlation with weather, building type, type of systems, and some behavioral factors 15
- Effect of Location: Why does it matter? - Community Effect 16
Summary of Findings: Effect of Building Characteristics Building type and floor area showed the largest impact when heating loads were dominant (winter) Building age did not show significant correlation with energy consumption Dual effect of equipment stock and air leakage or system efficiency 17
Summary of Findings: Effect of Appliances and Equipment Number of refrigerators is a significant factor in all models, mostly contributing to the base load. High consumption appliances such as water heaters, clothes dryers, air conditioners are highly correlated with peak load. 18
- Effect of Appliances and Equipment: Why does it matter? - Energy efficiency programs 19
Summary of Findings: Effect of Occupants Number of occupants has a non linear relationship with energy consumption. Pet ownership is a significant factor in all models. Long term habits (e.g., purchasing energy star equipment or consciously adjusting thermostat settings) have more correlation with consumption compared to short term habits. No significant income effect was observed. 20
Occupants and behavioral factors: practical insights 21
Data Driven Energy Efficiency BENCHMARKING 22
Purpose Identify most efficient buildings. Describe the characteristics of efficient buildings. 23
Method: Energy Efficient Frontier Analysis What are the services of a building? Provide space (floor area) Service Efficient Buildings Serve people (# of occupants) Serve business (revenue) Serve customers (# of customers) Hot shower and cold beer! e * e k Example: Floor Area x No. of Occupants Efficiency = e * / e k Energy
Energy Efficient Frontier for PowerMeter Data 25
Frontier analysis revealed trends that were not identified by regression analysis 26
Frontier analysis shows systematic differences in energy consumption of different groups of buildings (e.g., intervention analysis) 27
Benchmarking helps identify buildings with irregular consumption patterns. 28
Comparing frontier analysis with traditional kwh/sq.ft metric for energy efficiency Frontier Analysis Results Energy Intensity Results 29
Benchmarking: why does it matter? 30
Summary Different load features (baseload, peak load) are driven by different factors. Location, weather, occupancy level and large equipment/appliances are the usual suspects. Long term behavioral factors (purchasing preferences and thermostat settings) show high correlation with energy consumption. While regression model explains the overall trend in data, benchmarking helps identify specific features that improve efficiency. Frontier benchmarking methods help identify the impact of interventions, identify irregular consumption patterns identify efficient buildings across a wider range of building types and sizes. 31
Photo: Google/Connie Zhou