Quantifying occupant energy behavior using pattern analysis techniques

Occupant energy behavior is widely agreed upon to have a major influence over the amount of energy used in buildings. Few attempts have been made to quantify this energy behavior, even though vast amounts of end-use data containing useful information lay fallow. This paper describes analysis techniques developed to extract behavioral information from collected residential end-use data. Analysis of the averages, standard deviations and frequency distributions of hourly data can yield important behavioral information. Pattern analysis can be used to group similar daily energy patterns together for a particular end-use or set of end-uses. Resulting pattern groups can then be examined statistically using multinomial logit modeling to find their likelihood of occurrence for a given set of daily conditions. These techniques were tested successfully using end-use data for families living in four heavily instrumented residences. Energy behaviors were analyzed for individual families during each heating season of the study. These behaviors (indoor temperature, ventilation load, water heating, large appliance energy, and miscellaneous outlet energy) capture how occupants directly control the residence. The pattern analysis and multinomial logit model were able to match the occupant behavior correctly 40 to 70% of the time. The steadier behaviors of indoor temperature and ventilation were matched most successfully. Simple changes to capture more detail during pattern analysis can increase accuracy for the more variable behavior patterns. The methods developed here show promise for extracting meaningful and useful information about occupant energy behavior from the stores of existing end-use data