Condensed representation and individual prediction of consumer demand

Consumer Demand Response (DR) is an important research and industry problem, which seeks to solve three problems. First, grouping consumers into useful categories, second, predicting a given consumer's energy consumption, and third, estimating for that consumer if and when a specific device will be used. Unfortunately, measured consumer energy consumption patterns (24-hour load curves) show great variability even for an individual consumer, making it difficult to classify consumers into stable representative groups and to predict individual energy consumption. Traditional clustering methods have resulted in many hundreds of clusters, with a given consumer often associated with several clusters. In this paper, we present a new method that better classifies and predicts fine grain consumer energy consumption behavior. The method is based on Dynamic Time Warping. DTW seeks an optimal alignment between energy consumption patterns reflecting the effect of hidden patterns of regular consumer behavior. Using actual consumer 24-hour load curves from Opower Corporation, this method results in a 50% reduction in the number of representative groups and an improvement of 20% in prediction accuracy. We extend the approach to predict which electrical devices will be used and at what times for a given day, based on partial day data.