Electric — Address-only extract

Lakeside Electric — billing export (synthetic)

Sample data · no sign-up

Aggregation Weather normalization Benchmarking (EUI) Load & demand Seasonal & outdoor use Distribution & inequality Map & clustering Class × time Differential privacy Data quality

Disclosure control Suppression k = 15 1 customer ≤ 15% complementary on

An alternative to suppression: rather than hide small groups, differential privacy adds calibrated random noise to each count so no individual's presence can be inferred, while totals stay useful. Lower ε = stronger privacy = more noise.

Privacy budget (ε)

1.0

smaller = more private

Noise scale

1.0

Laplace b = 1 ÷ ε

Mean abs. error

±0.7

across 3 groups

True vs. differentially private counts by customer class

customers per group

Laplace mechanism, ε-differential privacy. Dwork, McSherry, Nissim & Smith (2006); Dwork & Roth (2014); US Census Bureau 2020 DAS.