Continuous Monitoring Center machine-learning risk triage
That crews should be dispatched mainly after a fault, outage, or visible field report.
Engineering design · California, United States
Executive brief
The operating-model shift, in one view.
The model is only the first step; value appears when an engineer can translate an anomaly into a precise, timely field dispatch.
AI value · Preventive catches and customer outage minutes avoided
Company-reported
PG&E reported 17 potential ignitions intercepted, 12 million unplanned customer-outage minutes avoided, 2,620 emergency response hours reduced, and about $6 million saved
Utility-reported avoided-event estimates; the independent articles repeat PG&E's figures rather than audit the counterfactual.
Before
Detect faults from alarms, outages, or reports after conditions escalate. → Locate and repair the failed or hazardous asset.
After
Scans sensor and smart-meter patterns for abnormal precursors. → Validate the signal, locate the asset, and perform preventive work before ignition or outage.
Human boundary
Models flag potential risk; engineers determine priority and crews verify and repair physical equipment.
Why it matters
That crews should be dispatched mainly after a fault, outage, or visible field report.
How the work changed
Before
How the work ran before the change.
Step 1 of 2
Grid operations
Detect faults from alarms, outages, or reports after conditions escalate.
ControlReactive response
Step 2 of 2
Field crew
Locate and repair the failed or hazardous asset.
ControlEmergency procedures
What changed
That crews should be dispatched mainly after a fault, outage, or visible field report.
Decision rightAI handles the default; humans own exceptions
After
How the same work runs now.
Step 1 of 2
Machine-learning monitoring stack
Scans sensor and smart-meter patterns for abnormal precursors.
ControlRisk alerts
Step 2 of 2
Engineer and field troubleshooter
Validate the signal, locate the asset, and perform preventive work before ignition or outage.
ControlHumans authorize and execute field intervention
Process model built from the published workflow evidence for Pacific Gas and Electric Company. Every step, actor, and control appears in full below.Every step, actor, and control
Exception path
Unconfirmed alerts can be monitored or closed; conventional protection, outage, and wildfire procedures remain in force.
Decision authority
Models flag potential risk; engineers determine priority and crews verify and repair physical equipment.
Before
#
Actor
Action
Control
01
Grid operations
Detect faults from alarms, outages, or reports after conditions escalate.
Reactive response
02
Field crew
Locate and repair the failed or hazardous asset.
Emergency procedures
After
#
Actor
Action
Control
01
Machine-learning monitoring stack
Scans sensor and smart-meter patterns for abnormal precursors.
Risk alerts
02
Engineer and field troubleshooter
Validate the signal, locate the asset, and perform preventive work before ignition or outage.
Humans authorize and execute field intervention
Work that left the path
Broad reactive search after failure
Some emergency response work that can be converted to planned preventive repair
Human role before
Engineers and crews responded after conventional alarms or customer impact.
Human role after
Engineers triage model alerts and dispatch targeted preventive inspection and repair.
AI role
Machine-learning anomaly detection across grid sensors and approximately 5.5 million smart meters.
Outcomes
Preventive catches and customer outage minutes avoided
Company-reported
Reactive response after equipment failure or ignition precursor escalated→PG&E reported 17 potential ignitions intercepted, 12 million unplanned customer-outage minutes avoided, 2,620 emergency response hours reduced, and about $6 million saved
Calendar year 2025 · PG&E electric grid monitored by tens of thousands of sensors and approximately 5.5 million meters
Utility-reported avoided-event estimates; the independent articles repeat PG&E's figures rather than audit the counterfactual.
What leaders can reuse
Anti-pattern
Counting every anomaly as a prevented wildfire or bypassing physical verification.
Questions
01How are avoided events estimated?
02What alert precision keeps field teams engaged?
Portability conditions
Dense asset telemetry
Location-aware diagnostics
Field capacity to investigate before failure
Reputation risk
medium
Evidence and authority
What the public record supports.
Current · updated
1 independent, 1 primary; publication outcomes are verified and reported.
Bundle 1.0.0 · reviewed 2026-09-06 · stable ID 6f860c6641d873cd
