PROPOSED PROJECT
Solar Performance Monitor
Give solar operators a daily view of arrays producing less than expected. Compare inverter readings with weather, site history, and maintenance logs, then rank unusual drops with supporting charts and possible explanations for an engineer to investigate.
The sun stayed up. One reading dropped.
Illustrative hourly records
10:00–16:00 / same site
Find the interval.
Then inspect the evidence.
A fictional six-hour window has 1,200 kWh expected output and 906 kWh recorded output, a 294 kWh difference. Inverter B records zero in the two midday intervals while A continues producing. The maintenance log has an inspection scheduled later, but no confirmed outage explaining those readings. An engineer must distinguish a real stop from a reporting problem.
WHAT IS USUALLY DONE
An operator checks daily totals across dashboards and maintenance notes. A cloudy day can hide a equipment-specific drop, while a missing reading may be mistaken for zero output. Engineers spend time aligning intervals and deciding which difference deserves attention.
WHAT SOVRUN CAN DO FOR YOU
Sovrun would build a daily review that aligns inverter readings, weather, site history, and maintenance records. It compares actual output with a documented expected range, separates missing readings from measured zero, and ranks unusual differences with supporting charts. Engineers see possible explanations and the evidence still needed before deciding on an inspection or maintenance action.
Two zero readings.
Several possible explanations.
Inverter A continues producing while B reports zero at noon and 13:00. A missing record would be labeled missing, not zero. The engineer still needs the device status and logger evidence to determine what actually happened.
Readings received
Readings received
Recorded zero / investigate
Recorded zero / investigate
Readings received
Readings received
Inspection scheduled at 13:30. No confirmed outage entry explains the earlier zero. Check the equipment status, logger, and actual work record before calling this a hardware fault.
A focused investigation.
Evidence an engineer can use.
The first build joins one defined group of inverters with its weather and maintenance records. Alerts guide a review; operating decisions remain with the site team.
10:00–16:00 / 294 kWh difference
Zero is a record.
- Readings
- 0 kWh / 0 kWh
- Data state
- Received, not missing
- Cause
- Unconfirmed
- Next check
- Device + logger status
Check before diagnosing.
- Evidence
- Intervals + ML-014
- Owner
- Site engineer
- Action
- Inspection to approve
- Recovered energy
- Not established
How we’d make it happen.
Align the records
Confirm time zones, intervals, units, and equipment identifiers. Separate zero readings, missing data, and known outages before comparing output.
Find the unusual differences
Compare observed output with the agreed baseline and inspect the equipment-level pattern. Attach weather context and maintenance notes without assuming a cause.
Review and learn from the finding
An engineer approves the next investigation. Preserve the original alert and compare later evidence with the initial explanation.
Where AI helps
Time-series models estimate expected output using the agreed weather and history inputs. Language models summarize linked maintenance records and draft investigation notes. Code checks timestamps, units, and totals. A flagged difference is not a fault diagnosis; expected output needs validation on the actual site.
What people approve
Engineers approve the baseline, review data quality, confirm known outages and site constraints, and decide which checks are appropriate. Maintenance actions remain with authorized staff. The tool does not remotely operate equipment or promise recovery of all modeled output differences.
What we’d need to start
- Timestamped inverter output, meter definitions, and explicit missing-data markers
- Relevant weather records, site configuration, historical output, and maintenance logs
- Engineer-approved comparison method, alert criteria, and review workflow
How we’d test it
Backtest on held-out site periods with known weather changes, outages, sensor gaps, and normal operation. Check false alerts, missed events, interval alignment, and expected-range accuracy. Evaluate follow-up outcomes separately; modeled output differences are not recovered energy.
Prepared for Energy & sustainability. We would agree on the work after reviewing your brief, materials, and requirements.
This is a proposed project, not completed client work. The visuals and numbers show example outputs, not measured results.
