Toward cost-effective residential energy reduction and community impacts: A data-based machine learning approach

Many U.S. utilities incentivize residential energy reduction through rebates, often in response to state mandates for energy reduction or from a desire to reduce demand to mitigate the need to grow generating assets. The assumption built into incentive programs is that the least efficient residences will be more likely take advantage of the rebates. This, however, is not always the case. The main goal of this study was to determine the potential for prioritized incentivization, i.e., prioritizing incentives that deliver the greatest energy savings per investment through an entire community. It uses a data mining approach that leverages known building and energy characteristics for predicting energy consumption of houses that collectively can be considered representative of all residences within an entire community. From this model, it estimates natural gas consumption and savings, and corresponding implementation costs associated with the adoption of the most impactful energy reduction measures. The resulting savings and cost estimates allow us to develop a sequential energy reduction strategy whereby the most economic measures within the whole utility district are addressed. The results show that an energy reduction of 36% can be achieved at a levelized cost of less than $14 per mmBTU ($14,780 per MJ), demonstrating the strong potential of this approach. A corresponding Economic Input–Output Analysis captures the cascading community economic impacts of this strategy. The results show that for the roughly 45,000 single-family residences in the studied region, an initial energy efficiency investment of $26M could result in a total cascading multiplier economic impact of $41M and additional economic impacts of $2.2M for the lifetime of the considered energy efficiency measures.