PNNL

Abstract

Occupant behavior can have significant and direct impact on building energy usage. With increasing efficiency improvement on building enclosures, lighting, HVAC systems, and major appliances, it is necessary to understand occupant behaviors and influence them effectively in order to achieve low-energy use targets. For example, incentives are a practical approach to discover and change behaviors. In this paper, we propose a data-driven framework to accurately quantify the impact on occupant behaviors with respect to certain incentives. The algorithm for quantifying impact uses machine learning techniques to learn the changes in the energy consumption of residential buildings. The proposed algorithm is developed using a kernel ridge regression model with a non-linear kernel and an unsupervised a clustering step. The results are given as a distribution of various behavior changes, which can help evaluate the effectiveness of the incentives. The performance of the algorithm is tested using simulated energy consumption in residential houses with an entropy-based clustering evaluation metric, which is consistently better compared to other bench-marking algorithms.