Power Grid Monitoring: California (CAISO)

Findings report · scored 2020-01-08 to 2026-04-01 · hourly · all times UTC

Summary

Over 6.2 years of hourly CAISO data the monitor fired 19 alerts. Of 3 target events: 2 predicted (Aug 2020 blackouts, 1.6 h lead and Sep 2022 record demand, 136 h lead); 1 detected (Sep 2020 heat wave, <1 h lag); 0 missed. 15 alerts fell outside every target window: 7 inside explanatory windows declared in the run config and 8 unexplained.

At a glance

Scored period6.2 years53,973 hours

Target events3

Predicted2 / 3

Detected1 / 3

Missed0

Alerts fired193 first, 1 repeat

Explained7hindsight label

Unexplained81.3/yr; 2.4/yr with explained

How it is scored

An event is PREDICTED if the first alert falls within 168 hours before onset, DETECTED if it falls within 48 hours after onset, and MISSED otherwise. The same windows apply to every event. The reasoning is in “Too early is the same as wrong”.
Onset is the time the grid operator declared the highest emergency level the event reached, converted to UTC.
Consecutive alerting hours count as one alert. A later alert inside a target window is a repeat, with no credit and no penalty.
An alert outside every target window is “explained” if it starts inside an explanatory window declared in the run config, and “unexplained” otherwise.

Target events

Event	Outcome	Lead / lag	First alert (UTC)	Onset (UTC)	Onset basis and source
August 2020 Rolling Blackouts Stage 3 Emergency, rotating outages on Aug 14 and 15	PREDICTED	1.6 h lead	2020-08-15 00:00	2020-08-15 01:38 2020-08-14 18:38 local	CAISO declared a Stage 3 Emergency. CAISO, CPUC, CEC, Preliminary Root Cause Analysis: Mid-August 2020 Heat Storm, section 3.3.2
September 2020 Labor Day Heat Wave Stage 2 Emergency, no outages ordered	DETECTED	<1 h lag	2020-09-07 01:00	2020-09-07 00:55 2020-09-06 17:55 local	CAISO declared a statewide Stage 2 Emergency. CAISO, system conditions bulletin (September 2020)
September 2022 Record Demand EEA 3 declared, no load shed ordered by CAISO	PREDICTED	136 h lead	2022-09-01 08:00	2022-09-07 00:17 2022-09-06 17:17 local	CAISO declared an EEA 3. CAISO, Grid Emergencies History Report (EEA3, 09/06/2022 17:17 to 20:00)

August 2020 Rolling Blackouts PREDICTED

Scoring window: 2020-08-08 01:38 to 2020-08-17 01:38 UTC (168 h before onset to 48 h after).

First alert at 2020-08-15 00:00 UTC, 1.6 h lead. Models flagging in that hour: demand, imbalance residual, and net generation.

CAISO had already declared a Stage 2 Emergency at 3:25 pm local time (22:25 UTC), so this alert came after the first emergency declaration and 1.6 hours before the Stage 3 declaration. That is too short a lead to act on.

September 2020 Labor Day Heat Wave DETECTED

Scoring window: 2020-08-31 00:55 to 2020-09-09 00:55 UTC (168 h before onset to 48 h after).

First alert at 2020-09-07 01:00 UTC, <1 h lag. Models flagging in that hour: demand, imbalance, and imbalance residual.

The alert is on the hourly reading stamped 01:00 UTC, the first one after the 00:55 UTC declaration.

September 2022 Record Demand PREDICTED

Scoring window: 2022-08-31 00:17 to 2022-09-09 00:17 UTC (168 h before onset to 48 h after).

First alert at 2022-09-01 08:00 UTC, 136 h lead. Models flagging in that hour: demand and forecast error.

Repeat alerts inside the window: 2022-09-02 04:00 UTC.

Every alert

All 19 alert episodes in the scored period, in order. Nothing is left out.

#	Start (UTC)	End (UTC)	Hours	Class	Matched to	Models flagging at start
1	2020-02-04 03:00	2020-02-04 03:00	1	Unexplained		demand, imbalance
2	2020-04-12 03:00	2020-04-12 03:00	1	Explained	Explanatory window: COVID-19 demand shift, Apr to May 2020	demand, imbalance residual
3	2020-05-27 00:00	2020-05-27 00:00	1	Explained	Explanatory window: COVID-19 demand shift, Apr to May 2020	demand, imbalance, net generation
4	2020-08-15 00:00	2020-08-15 00:00	1	First alert	August 2020 Rolling Blackouts	demand, imbalance residual, net generation
5	2020-09-07 01:00	2020-09-07 03:00	3	First alert	September 2020 Labor Day Heat Wave	demand, imbalance, imbalance residual
6	2020-10-08 20:00	2020-10-08 22:00	3	Explained	Explanatory window: Wildfire power shutoffs, Oct 2020	demand, imbalance, imbalance residual, net generation
7	2022-09-01 08:00	2022-09-01 10:00	3	First alert	September 2022 Record Demand	demand, forecast error
8	2022-09-02 04:00	2022-09-02 04:00	1	Repeat	September 2022 Record Demand	demand, imbalance, net generation
9	2022-11-29 09:00	2022-11-29 10:00	2	Unexplained		demand, imbalance, net generation
10	2023-04-29 18:00	2023-04-29 18:00	1	Unexplained		demand, imbalance
11	2023-05-17 08:00	2023-05-17 09:00	2	Unexplained		demand, imbalance
12	2023-11-10 08:00	2023-11-10 14:00	7	Explained	Explanatory window: Atmospheric river, Nov 2023	imbalance, imbalance residual, net generation
13	2024-05-19 14:00	2024-05-19 14:00	1	Unexplained		imbalance, net generation
14	2024-12-08 19:00	2024-12-08 19:00	1	Unexplained		imbalance, net generation
15	2025-02-23 19:00	2025-02-23 19:00	1	Unexplained		imbalance, net generation
16	2025-10-27 22:00	2025-10-28 00:00	3	Unexplained		demand, forecast error
17	2026-03-17 04:00	2026-03-17 05:00	2	Explained	Explanatory window: Atmospheric river, Mar 2026	demand, forecast error, imbalance residual
18	2026-03-18 04:00	2026-03-18 05:00	2	Explained	Explanatory window: Atmospheric river, Mar 2026	demand, forecast error, imbalance residual, net generation
19	2026-03-19 04:00	2026-03-19 04:00	1	Explained	Explanatory window: Atmospheric river, Mar 2026	demand, imbalance residual, net generation

Configuration of this run

Signals, one HTM model each: demand, net generation, forecast error, imbalance, imbalance residual.
A model flags an hour when its anomaly likelihood is at or above 0.98. A model's warmth is the share of flagged hours over its last 6 hours.
Signals are grouped (table below). The system alerts when 2 of 2 groups are hot for at least 2 consecutive hours.
Warm-up: the first 168 hours are not scored. Learning stays on for the whole run; no labels are used.

Group	Signals	Hot when
structure	demand, net generation	2 of its 2 signals flag in the same hour, or the group's mean warmth is at least 0.45
residual	forecast error, imbalance, imbalance residual	1 of its 3 signal flags in the same hour, or the group's mean warmth is at least 0.5

Limits

Three target events is a case study. It shows how the detector behaves on these events; it does not establish a detection rate.
The replay treats each hourly value as available at its timestamp. EIA publishes demand about an hour after the operating hour and net generation the next day, so a live deployment at these lead times would need the operator's own telemetry.
"Explained" is assigned in hindsight from explanatory windows declared in the run config. It never counts as a catch, and the alert rate is given both with and without those alerts.
The alert-rule settings listed above were set per region. Only the HTM model parameters, the encoder shape, the one-week warm-up and the consensus method are shared between regions.
The imbalance residual signal subtracts an hour-of-day median that was computed over the whole data file, not only over past data. A causal version of that signal needs a rerun.
628 hours are absent from the run's timebase because at least one input signal was missing in the source data.