Empresa Eléctrica de La Habana · Telegram channel ·
Every outage notice, fault report and grid bulletin the Havana electric utility published to its Telegram channel — messages — parsed into the numbers the utility never charted itself.
Supply
How much generation Cuba's grid was short of demand, as reported daily by the Unión Eléctrica. This is the one quantitative series that runs the full span of the archive.
Peak shortfall reported each day, in megawatts. Filled dots are the maximum the utility reported for that day; the line is a 30-day rolling mean of those. Hollow dots are the Unión Eléctrica's forecast, kept only for the days with no reported actual and left out of the mean. Vertical marks are the days the national grid collapsed entirely.
The days, , on which the utility published both figures. The gap between the two lines is the shortfall that becomes a blackout. Lines break wherever more than a week has no bulletin: silence there means not published, not balanced.
Daily energy from Cuba's new photovoltaic parks, the one supply line that grew during the period: reported days, . Lines break across stretches of more than a week with no report; hover for the day's peak power.
Days on which the utility announced a desconexión total del Sistema Electroenergético Nacional — the whole country dark at once.
The Unión Eléctrica's daily Situación del SEN bulletin names the thermal units that are broken down or in maintenance. bulletins, to .
Mean number of units named in breakdown or in maintenance per bulletin-day. A dashed baseline marks a month with no bulletin at all.
The bulletin is reposted irregularly — — so a gap means not reposted, never nothing broken.
The fifteen units named most often, split into breakdown and maintenance. A plant named without a unit number is a single-unit plant.
Counts are days reported, not days actually down: a unit is only counted on a day the bulletin was reposted and named it. Last-reported dates are in the table.
Capital
From the utility began posting the capital's own shortfall several times a day. Window: onward — earlier months used a different reporting format and are not comparable.
Band shows the range across each day's reports; the line is the daily mean. The monthly mean went from to across .
From the capital notice also says how many circuits the cut is spread over (se afectan 220 MW distribuidos en 68 circuitos). A typical day affects circuits, with the busiest single report naming ; the median cut works out at per circuit ( days). Hover for the megawatts behind each day.
Mean megawatts cut by hour of day. Two peaks, not one: the deepest cuts land at () and again at the evening peak (). The easiest hour to have power is .
Mean megawatts cut, by weekday and hour. Darker is worse.
Havana is divided into six rotating blocks. Pairing each block's outage-start and restoration notices gives its outage length. All six sit within of one another on the median — the rotation is even. The tail is not: the worst single outage in every block ran past .
Circuits
From the utility has named the individual distribution circuits it cut and restored. Pairing those notices over days, to , measures how long each circuit was actually dark — the first thing in this archive that is a measurement rather than a proxy.
Total measured dark time divided by the number of circuits tracked there, so municipalities with more circuits are not penalised. Compare with the four-year mention counts under Network below — this one is measured, not inferred.
Distribution of every paired outage in the window; the overall median is . For scale, the block rotation in Capital has medians of .
Cumulative hours without power over the window. Hover for the median outage and the number of separate outages behind each total.
Circuit-hours of outage summed across every tracked circuit each day — one circuit out for ten hours counts the same as ten circuits out for one.
Paired outages and their median length, per month. Hover for total hours dark and circuits tracked. The first and last months are partial, so compare their medians, not their counts.
Circuit start notices per month, by the cause the notice itself gives (DAF is automatic frequency load shedding). Counts are start notices, not paired outages.
Until the standard notice named no cause, so most Sin causa indicada before then is deficit in the older template — the shift is a wording change, not a real change in the grid. This tags what the notice says, not a verified root cause.
Pairing start and restoration notices is inference, so it was checked against a source it does not use: the deficit updates state how long each circuit has been out so far, which gives an independent start time.
Your circuit
Everything the archive records about a single distribution circuit. Only the measured-outage window carries circuit detail, so this covers the same dates as the section above — the rest of the dashboard is city- or country-level and has no circuit dimension to filter by.
Every measured outage across the window. Each block is one outage, from the notice that cut it to the notice that restored it.
In order. The rule across the plot is the city-wide median outage, so you can see whether this circuit runs longer or shorter than typical.
The most recent notices naming this circuit, newest first — the raw text behind every number above.
Recovery
When Havana goes dark the utility posts a running count of households reconnected. Those bulletins are the only place the archive states how many people are affected rather than how many megawatts — and strung together they measure the city's recovery, outage by outage.
Every restoration episode, aligned to its first bulletin. The fastest and slowest are picked out; hover any line to identify it.
In chronological order. The spread is the finding, not a trend — anywhere from to across the episodes that got there, and episodes in all are too few to call a slowdown (rank correlation with time is only ). An n/r marks an episode that never reached half the city while the utility was still reporting.
Peak share reconnected while the utility was still reporting. Bulletins stop before full restoration, so treat each as a floor, not a final figure.
Network
Beyond the generation shortfall, the distribution network breaks. These counts are messages mentioning each fault type per month — a measure of reporting volume, which tracks but does not equal the true fault count.
dominates and keeps climbing — mentions in , in .
Mentions of each generation source. Distributed diesel sets and the rented Turkish powerships fade out of the record entirely; only the ageing thermal plants remain.
Messages about automatic frequency disconnection (Disparo Automático por Frecuencia) — the grid dropping load on its own to avoid collapse. A direct measure of how close to the edge the system runs.
Messages naming each of Havana's 15 municipalities alongside a fault. Reflects both network condition and municipality size.
Unlike the counts above, the utility sometimes posts the whole list of faults it has not yet repaired. such snapshots cover days between and ; the median day lists open items.
Monthly mean of each day's open items, from the fullest snapshot posted that day.
These are listed items, not households: one item can be a transformer serving a few homes or a feeder serving thousands. Paler bars are months the series starts or ends inside.
Mean open items per snapshot day, by municipality.
An item naming several municipalities counts toward each of them, so the bars do not sum to the daily totals.
Public
Telegram lets readers react to each post. Across reactions this becomes the longest continuous series in the archive — and needs no format normalisation to read.
👎 🤬 😢 😱 as a share of all positive-or-negative reactions, by month. The channel opened at negative and now sits near .
Total count per emoji across all messages.
Monthly volume rose from in to a peak of in as the crisis deepened.
se afectan NNN MW
(peak ). The
national series comes from the Unión Eléctrica's máxima afectación fue de NNNN MW
(peak ). Conflating them inflates Havana threefold.Bloque N.pdf schedule attachments were excluded from the export,
so published schedules could not be read.