What happened
In 2021, Portland General Electric and Pacific Gas & Electric hosted pilots of Pano AI's Rapid Detect platform through Incubatenergy Labs, an EPRI program with utility sponsors. EPRI was the technical advisor. Cameras on mountaintop towers turn 360 degrees every minute, and AI-based methods detect smoke. The Pano Intelligence Center applies machine learning across all stations to confirm sightings before alerts go out. PGE had two stations and PG&E had access to 13. EPRI's report of March 2022 says that the final November model detected 60% of test vegetation fires within 4 minutes of visible smoke and 85% within 10 minutes. The false positive rate fell from about 1.6% in late July to 0.37% in early October as the models improved. Across more than 50 alerts to utilities, the alerting accuracy rate was above 90%. In the Round Fire in northern California, Pano AI alerted before the fire was reported to first responders. The report gives no count or location for the test fires. It gives no comparison with other ways of detecting fires.
Sources
-
Evaluation report, by an evaluator that the makers or users commissioned.
EPRI (Electric Power Research Institute). Incubatenergy Labs 2021 Pilot Project Report: 360-Degree, AI-Based Imaging for Wildfire Situational and Locational Awareness. Product 3002023034, March 2022. EPRI was the technical advisor to pilots hosted by Portland General Electric and Pacific Gas & Electric (evaluator's account).
Pano AI's platform was able to detect 60% of the test fires within 4 minutes from the time smoke was first visible and 85% within 10 minutes
Learning by AI-based models reduced the false positive rate during the pilot project, helping utility analysts focus attention on actual incidents posing potential wildfire risks.
Details
AI system
- Reported configuration
-
Final November production model
- Workflow
-
Cameras on existing mountaintop towers turned 360 degrees every minute and sent ultra-high-definition images. AI-based methods detected smoke, and the Pano Intelligence Center applied machine learning across all stations to confirm sightings before alerts. Analysts used triangulation and maps of utility infrastructure to decide which incidents to escalate.
Follow-up
- Conditions for replication
-
Stations need towers with power, connectivity and permits, about 10 miles apart for triangulation. Watch out to about 25 miles needs clear days.
Entry details
- Sector
-
Wildfire detection / electric utilities.
- Purpose
-
Pano AI built Rapid Detect to find wildfire smoke early with rotating mountaintop cameras and AI and to alert utilities and fire agencies.
30 more questions not yet checked
model, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). Pano AI spots most test fires within minutes in a utility pilot (AIBD-1135). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1135.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)