Hugging Face Datasets - Utilities Search (75+ datasets)

One search query, seventy-five collections, one delivery. Hugging Face Datasets - Utilities Search is the Hub's result set for the query 'utilities' as captured on 2026-08-21, and it is dominated by a single publisher: electricsheepafrica, whose Nigerian Energy & Utilities series contributes roughly 62 of the 75 hits, each a tabular collection shipping both a snappy-compressed Parquet file and a CSV twin with a README card stating row counts between 60k and 220k rows and an explicit column-schema table.

The series reads like a syllabus for emerging-market grid analytics. Twenty-one named themes hang under it: AI grid optimization, billing and payments, carbon footprint, commercial and industrial consumption, customer complaints, demand forecasting, dynamic pricing response, energy storage cycles, energy theft anomalies, equipment sensors, EV charging sessions, field technician logs, grid load, household smart meter, line losses, microgrid operations, outage fault logs, peak demand, plant emissions, power flow and power plant output.

The other thirteen results split two ways. gunnybd01 publishes four Utilities_News text corpora around small modular reactor coverage - showing 12.3k, 12.2k, 7.83k and 7.8k downloads at capture - and ttn1410/utilities_business_data adds a 47-row company-profile table with ticker, name, sector, industry, country, market_cap, currency, employees and description. Every row of every table arrives exactly as documented upstream; nothing is inferred from field names alone.

Three billing records exactly as they arrive, one per distribution company across tariff bands A through D:

customer_id       : CUS-0000000
disco             : Jos
billing_month     : 2024-12
tariff_band       : C
kwh               : 144.82
price_ngn_kwh     : 120.0
amount_billed_ngn : 17378.33
amount_paid_ngn   : 15947.79
paid_on_time      : false
arrears_ngn       : 1430.54

customer_id       : CUS-0000001
disco             : Kano
tariff_band       : A
kwh               : 134.48
price_ngn_kwh     : 206.0
amount_billed_ngn : 27702.87
paid_on_time      : true
arrears_ngn       : 2618.15

customer_id       : CUS-0000002
disco             : Enugu
billing_month     : 2024-06
tariff_band       : D
kwh               : 85.94
price_ngn_kwh     : 90.0
amount_billed_ngn : 7734.78
paid_on_time      : true
arrears_ngn       : 635.81

Note what three rows already demonstrate: the DisCo names are real distribution-company geographies (Jos, Kano, Enugu), tariff bands run A-D within a single month's sample, and payment behaviour arrives as a boolean with its naira consequence sitting beside it in arrears_ngn - so revenue-assurance math needs no derivation. Get a sample of this dataset cut to your themes and Datadory returns rows against your own question.

The billing-and-payments table, typed and defined - ten columns verified from an actual file pull during research, not read off a field-name list:

Beyond this table, the remaining sixty-plus collections carry their own documented schemas - sensor readings, outage events, charging sessions, technician logs - published as schema tables on each card rather than restated here. Those definitions travel with the sample on request, folded under additional fields on request so this page stays honest about what was verified versus what is documented-but-not-yet-pulled.

• Geographic: Nigeria for the main tabular block - distribution companies, substations and plants; the four news corpora and the company-profile table cover global utility firms instead.
• Temporal: observed billing months run 2024-06 through 2024-12; the collection itself was captured 2026-08-21 with repo commits as recent as 2026-08-12.
• Granularity: row-level records throughout - customer-month billing rows, individual news articles, one row per profiled utility company. Collection sizes range from under 1K rows to the 100K<n<1M band.

Read plainly: this is a snapshot of what the Hub surfaced for one query on one day, not a flowing operational feed. The billing months sit two years behind the capture date, which makes the tables vintage snapshots suited to schema work and benchmarking rather than current-state monitoring. Stacked dated deliveries are how any time series gets built here.

API, files, or your warehouse. Daily, weekly, or hourly.

ML model training. Sixty-plus themed collections give a benchmarking bench few industries can match: train a demand-forecasting model on one theme and validate generalization on grid load, peak demand and power flow. Row volumes of 60k-220k per theme keep training runs meaningful without a data-engineering project first.

Schema and product prototyping. The documented schemas cover nearly every artifact a utility-data product needs - customer-month billing rows, outage fault logs, EV charging sessions, field technician logs. Teams building utility analytics stand up realistic test environments in an afternoon instead of negotiating for production data.

Emerging-market revenue assurance. The billing table supports arrears analysis directly: amount_billed_ngn, amount_paid_ngn and arrears_ngn on every row mean collection-rate math is arithmetic, not estimation, across tariff bands A-D and multiple Discos.

Teaching and journalism. Four news corpora plus a company-profile table give students and writers real-shaped material without exposing actual customer records - useful precisely because the engineered origin is disclosable up front.

The personas who pull this most: data scientists and ML engineers building models, developers and data-product builders wiring utility schemas into products, market researchers mapping emerging-market utilities, and journalists, academics and students teaching with it. Industry context sits on the multi-utilities data hub; persona workflows at data scientists × multi-utilities, developers builders × multi-utilities and market researchers × multi-utilities.

Within the multi-utilities shelf this is the volume play, answering a different question than the measured sources beside it. USGS Water Services - Instantaneous, Daily and Statistics REST APIs serve keyless readings for roughly 1.5 million monitored US sites, and World Bank Energy & Mining - World Development Indicators (Topic 5) ships 41 indicators across about 217 economies - official statistics built to survive scrutiny. This collection instead offers breadth and row volume: 75 collections, roughly 5-10 GB uncompressed, twenty-one named themes. Prototype and train here; cite the others when a number has to hold up. The full ranking sits on best multi-utilities datasets.

Stated up front, because they shape the analysis:

• Engineered origin. The cards describe the collections as engineered by Electric Sheep Africa; whether rows are fully simulated or partially sourced from real DisCo operations is stated nowhere. Treat them as prototyping and benchmarking material, not regulatory or financial evidence.
• Vintage snapshots. Observed billing months end at 2024-12 while the capture date is 2026-08-21 - there is no contractual refresh behind these tables.
• Perishable perimeter. The 75-repo set reflects one query on one day; ranking drifts, so the exact composition will move between captures.
• Concentration risk. Sixty-two of seventy-five hits come from one publisher, so a change to that single series reshapes the whole result set.

None of these surface at delivery time as surprises: the caveats ship attached, and Datadory flags which analyses they bite before anyone commits to one.

Because the interesting question is rarely "all seventy-five" - it is your themes, your tariff bands, your Discos. Requested samples come back cut to that shape with every documented column intact, Parquet twins included where the pipeline wants them. Scheduled deliveries land keyed on customer_id, so consecutive passes diff cleanly into arrears timelines and collection-rate trends even though the upstream tables themselves are vintage snapshots. The full catalog holds the rest of the multi-utilities shelf alongside this one.