Uncertainty management for classification and benchmarking of energy-use preference profiles

In the present state of information technologies, Big Data and Smart Metering allow very precise readings of human behavior, like e.g. energy consumption from households and firms. Focusing on detailed micro-data profiles representing the users’ preferences through time, preferences have to be represented for aggregation into an initial set of informative classes. For such an initial classification, a general frame is required which can be later exploited according to a specific purpose/application, while appropriately handling the uncertainty in human behavior classification. For the former, it is noticed that the aggregation of preferences profiles should maintain its characteristic shape, thus proposing a method for inferring class membership on the basis of proximity between any given profile, and specific (target) examples of the desired profiles. For the latter, a classification based on opposites and neutral categories is suggested. In consequence, identifying the profiles that should be targeted for specific applications, as in market design for demand response. The proposed methodology is applied to the energy sector, using a sample of 1243 firms in Denmark and their hourly energy use throughout 2013, classifying and benchmarking firms according to their green energy efficiency.