Centre for Net Zero’s Faraday dataset (OpenSynth) is excellent and effectively<\/p>\n
unusable casually \u2014 it’s ~6GB of parquet with the load profiles stored as<\/p>\n
delimited strings. So I aggregated it and published the result.<\/p>\n
150 cohorts, each with a mean daily total, a 24-hour load shape, and deciles<\/p>\n
where the cell was thick enough:<\/p>\n
– baseline (no solar\/battery\/EV\/heat pump)<\/p>\n
– property type, EPC band, and the cross-tab<\/p>\n
– month<\/p>\n
– tariff type (standard \/ Economy 7 \/ smart \/ automated)<\/p>\n
– heat pump: with vs without, unmatched, matched on property+EPC, by month,<\/p>\n
and by tariff<\/p>\n
– EV: with vs without, and by tariff<\/p>\n
– LSOA k-means cluster<\/p>\n
Cross-check, which is the reason to trust any of it: baseline comes out at<\/p>\n
9.737 kWh\/day and a median of 8.31. SERL Statistics Report 1 \u2014 separate<\/p>\n
source, ~13,000 real metered GB homes \u2014 publishes a mean of 9.8 and a median<\/p>\n
of 8.2. Two moments of the distribution, two unrelated datasets, ~1% apart.<\/p>\n
Limitations, up front:<\/p>\n
– Synthetic. Faraday is a generative model trained on ~1bn smart meter<\/p>\n
readings from Octopus customers, who over-index on smart tariffs and LCT.<\/p>\n
– No household identifier, so the deciles are over household-DAYS, not<\/p>\n
households. That spread is wider than the between-home spread \u2014 treat it<\/p>\n
as an upper bound. It’s recorded in `distributionOver` in the JSON.<\/p>\n
– EPC band moves consumption by <0.1% within a property type, including for<\/p>\n
heat pump households where insulation should dominate. I read that as the<\/p>\n
model being under-conditioned on EPC rather than a finding about housing,<\/p>\n
and I’ve built nothing on those cuts. Published anyway \u2014 a null result is<\/p>\n
still a result.<\/p>\n
– `cluster_label` is a k-means grouping of LSOAs on socio-demographic<\/p>\n
features, NOT geography. Published as clusters, never as regions. Each<\/p>\n
cluster’s country mix (from LSOA code prefixes) is included so you can join<\/p>\n
your own geography.<\/p>\n
– Cohorts under 500 profiles are dropped rather than published thin.<\/p>\n
CSV and JSON, CDLA-Permissive-2.0 (same as the source), no registration.<\/p>\n
Derivation script is in the repo and reproduces the file exactly.<\/p>\n
https:\/\/www.energycosting.co.uk\/data\/uk-domestic-electricity<\/a><\/p>\n All credit to Centre for Net Zero for Faraday \u2014 I’ve only aggregated it.<\/p>\n<\/div>\n submitted by \/u\/aussiesteveau <\/a>
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
