Building a recommendation system for residential electricity consumption reduction

Abstract

Our project addressed a practical problem for the UK energy transition: utilities need to reduce electricity use during peak hours to reduce the grid’s stress. In households, existing “one-size-fits-all” messages and incentives work unevenly because each one has very different consumption habits. Using half-hourly smart meter readings, we developed a behavioural clustering pipeline that groupshouseholds into clear “behavioural profiles” based only on how they use electricity (for example, daytime vs evening use, weekday vs weekend patterns, and seasonality), without relying on surveys ordemographic data. We first designed and tested this approach on the UK Energy Demand Research Project dataset (2007–2010), and then applied it to EDF’s Beat the Peak+ randomised controlled trial (2023–2024) to evaluate how different behavioural groups changed peak-time consumption under tiered incentives. To quantify impact and identify who responds most, we combined standard causal methods with machine learning models for heterogeneous treatment effects and quantile-based analysis. The main contribution is clear evidence that behaviour-based segmentation predicts responsiveness much better than socioeconomic variables, and therefore can be used to target demand-side programmes more efficiently. In particular, the “Stay-at-Home” profile showed the most consistentreductions in weekday peak demand (around 7% on average), while low-consumption householdsshowed little to no change. A second key result is that average effects can mask importantheterogeneity: the largest reductions are concentrated in a smaller subset of high-respondinghouseholds, which our quantile analysis and heterogeneous treatment effect models helped identify. Overall, we produced a practical and scalable framework for using smart meter data to design targeted peak demand reduction interventions, improving programme cost-effectiveness for utilities and supporting national decarbonisation objectives.