An Approach for Assessing Clustering of Households by Electricity Usage

Abstract

How a household varies their regular usage of electricity is useful information for organisations to allow accurate targeting of behaviour modification initiatives with the aim of improving the overall efficiency of the electricity network. The variability of regular activities in a household is one possible indication of that household's willingness to accept incentives to change their behaviour.An approach is presented for identifying a way of representing the variability of a household's behaviour and developing an efficient way of clustering the households, using these measures of variability, into a few, usable groupings.To evaluate the effectiveness of the variability measures, a number of cluster validity indexes are explored with regard to how the indexes vary with the number of clusters, the number of attributes, and the quality of the attributes. The Cluster Dispersion Indicator (CDI) and the Davies-Boulden Indicator (DBI) are selected for future work developing various indicators of household behaviour variability.The approach is tested using data from 180 UK households monitored for over a year at a sampling interval of 5 minutes. Data is taken from the evening peak electricity usage period of 4pm to 8pm.

Investigating Underlying Drivers of Variability in Residential Energy Usage Patterns with Daily Load Shape Clustering of Smart Meter Data

“…We use the iterative truncation algorithm with violation rates (V ) of 10% and 30%, which results in sets of cluster centers numbering 608 and 99, respectively. We evaluate these two sets of representative clusters by using the Davies-Bouldin index (DBI), which is a metric of cluster separation ( [37], [20]). The DBI for the 608 and 99 cluster center sets are 2.23 and 2.22, respectively, indicating both sets of clusters have similar performance with respect to cluster separation.…”

“…Residential customers are characterized by highly volatile behavior, which challenges the application of clustering methods to individual load curves [10]. Using a large sample of residential daily load profiles (>100,000) and six performance metrics Jin et al [19], following [20], conducted a comparative study to evaluate eleven direct clustering methods under four families of algorithms: centroid based, hierarchical, density based, and model based methods. They found whole time series clustering of residential load profiles exhibits a trade-off between cluster compactness and distinctness and the number of clusters required to achieve adequate performance was 50 to 100, much larger than that of non-residential customers.…”

A review on clustering of residential electricity customers and its applications

“…Usually one or a few of clustering validity indexes (CVI) are used along with the clustering techniques. These indexes are used to evaluate the performance of clustering methods and hence, they can be utilized to determine the suitable number of clusters [29] [22], to compare the performance of different clustering techniques [2] and to evaluate the performance of clustering when some attributes (features) are added or removed [30]. Table II reports some of the most used CVIs along with the corresponding references.…”

“…Three different attributes are used in [30] as the input of KM method to cluster residential customers. The attributes (flexibility measures) defined in a way to represent the flexibility of each household, hence, allowing applying demand response programs to those customers with highest possibility of changes in their loads.…”

Clustering Methods for Electricity Consumers: An Empirical Study in Hvaler-Norway

“…These features could be statistics of the time-series (e.g. total usage, maximum usage, standard deviation at peak hour) [13], load shape indexes (e.g. load factor, night impact, lunch impact) [14], frequency domain indexes (e.g.…”