%0 Report
%A Ling Jin
%A C Anna Spurlock
%A Sam Borgeson
%A Daniel Fredman
%A Liesel Hans
%A Siddharth Patel
%A Annika Todd-Blick
%D 2016
%G eng
%T Load Shape Clustering Using Residential Smart Meter Data: a Technical Memorandum
%8 09/2016
%X <div>The rise of Advanced Metering Infrastructure has enabled large volumes of electricity&nbsp;consumption data to be captured at an hourly frequency or even higher. A thread of research has demonstrated methods for coupling this fast growing data stream with data mining techniques&nbsp;such as cluster analysis for categorization of electricity load patterns. In past research on residential customers, such categorization has usually been conducted on aggregated load data,&nbsp;partly due to large variability exhibited within and across customers. However, in the context of demand response and efficiency programs, load patterns of individual customers and their daily and inter-daily variability directly relate to each customer’s ability to respond to program incentives.</div><div>&nbsp;</div><div>This document is a technical memorandum of application of an innovative clustering technique&nbsp;to individual customers’ daily load data resolved at the hourly level across a large sample of residential customers over a full year period. An additional innovation of our work is that we&nbsp;focus our analysis on the timing of discretionary1 electricity usage in particular, as opposed to total electricity use. We document the innovations and hyperparameter selection in the clustering process specific to our residential smart meter dataset and derive a diverse set of archetypical discretionary loadshapes.</div><div>&nbsp;</div><div>While typically utilities and system operators focus on the aggregate residential load shape,&nbsp;application of this improved clustering method will shed light on the considerable heterogeneity and variability across days and customers. In the future, more behavioral features associated with household consumption schedules and variability can be extracted based on our results and can be used in future studies to segment customers for better program targeting and designing tailored recruitment strategies.</div>