Clustering electricity consumption time series into two customer groups.

Hi all,

I have hourly electricity consumption data of 250 customers over a two-year time period. This customer base is made up of two groups: 1) customers who use electric heating (high electricity consumption) and 2) those who use district heating (significantly lower electricity consumption). However, I do not know which or how many customers belong to which group.

I need to cluster the customers into two groups and get the customerIDs belonging to those respective groups. I know how to do the clustering of the data points at each individual point in time, but the clusters would be different at each point in time, so this does not work. How would I go about clustering the entire time series data in MATLAB?

[Adrian Roos]

The raw consumption data is stored in a .csv file with the first row containing the customerIDs and the first column containing the timestamps. I would also appreciate a tip about what would be the smartest way to import it into MATLAB; as what type (table, array, time series, etc).

Answer 1

[the cyclist]

I'm actually not sure if there are MATLAB functions that support these analyses, but I think the models you want are known as "Latent Class Growth Analysis" (LCGA) and "Growth Mixture Model" (GMM). Don't confuse the latter with "gaussian mixture models". You might want to search on those, plus MATLAB as a keyword.

If you are also handy with R, the lcmm package has functionality for these models.