Cases

ZenovaOS AI pilot cases

Implementation paths, before-and-after states, and measurable operating outcomes.

Priority accuracy and follow-up speed

Alarm governance pilot

Baseline alarm volume, validate P0/P1/P2 grouping, and document how attribution evidence changes daily operating review.

Report hours and review pass rate

Monthly report automation

The expert report team drafts conclusion-first monthly operating materials; humans only review and confirm.

Loss attribution coverage

PR / underperforming plant analysis

Locate PR swings and loss sources for underperforming plants, with a followable improvement list.

Publication path

Case studies start as measurable pilots.

Each case should record the pain, Monitor integration path, account scope, proof artifacts, unavailable-data explanations, and measurable result before becoming public-facing material.

Real problem

Start with where the previous workflow was stuck, such as alarm overload, slow reporting, or scattered evidence.

Traceable evidence

Keep key data, charts, field boundaries, and analysis logic so the business knows where conclusions came from.

Measurable outcome

Explain what changed after the pilot with agreed metrics, not just a polished answer sample.

Grounded expansion

End by showing which plants, roles, or operating workflows should expand in the next stage.

Acceptance criteria

A strong case explains before, after, and the basis for acceptance.

Acceptance is not about polished AI answers. It is about whether the business owner can confirm the result, use it in daily meetings, and rely on it for the next expansion stage.

Before the pilot

Record the original workflow problem, such as slow alarm judgment, report effort, or scattered device evidence.

After the pilot

Show how AI produces conclusions, evidence, charts, reports, and actions while reducing manual preparation.

Acceptance metrics

Measure with criteria agreed up front such as priority accuracy, report review pass rate, or evidence traceability.

Deliverables

Keep sanitized replays, chart samples, report samples, review conclusions, and next-stage expansion recommendations.

Sanitized replay evidence

Real operating questions, rewritten as public proof.

These examples come from approved conversation replays, with plant identities and device identifiers generalized for public use while preserving the operating logic.

Showing 6 of 16 replay items

Sanitized replayPR / underperforming plant analysisGeneration analysis

Multi-plant generation efficiency comparison

May reviewEquivalent hoursCross-plant ranking

Question: Compare May generation performance across two industrial PV plants.

Chart evidence

Equivalent utilization hours

9.7% stronger unit output

Large plant130.6 h

Efficient plant143.3 h

The smaller plant is the better efficiency story even though total MWh is lower.

Sanitized reply

After normalizing both plants to the same meter-based May period, Plant A wins on total generation because of larger installed capacity, while Plant B wins on unit output.

Evidence: The replay normalized both plants to the same meter-based period, compared capacity, generation, and equivalent utilization hours, then explained why the smaller plant showed stronger unit output.
Outcome: One plant produced 418.15 MWh because of scale, while the smaller plant reached 143.3 equivalent hours and outperformed on efficiency by 9.7%.
Website use: Turns a cross-plant question into an executive-ready efficiency story with traceable metric definitions.; /cases · /scenarios

Sanitized replayWeather impact and operating recoveryWeather attribution

Rainstorm impact analysis

12 plantsRain-period comparisonFault exclusion

Question: How much did rainstorms affect PV generation across the Shanghai region?

Chart evidence

Generation suppression

52.4% average rain-period drop

Average52.4 % drop

Largest74 % drop

Weather impact is large enough that a fault conclusion would be premature.

Sanitized reply

The replay first separates weather-driven suppression from equipment alarms by comparing generation before, during, and after the rain window across plants with both generation and weather coverage.

Evidence: The replay compared generation before, during, and after rainy days across 12 plants with weather and generation coverage, separating objective weather impact from equipment faults.
Outcome: Rain-period daily generation fell 52.4% on average; all reviewed plants showed significant weather-driven suppression, with the largest drop reaching 74.0%.
Website use: Shows ZenovaOS AI can explain low generation without over-blaming equipment when weather is the real driver.; /operations · /insights

Sanitized replayData governance and measurement consistencyData governance

Meter and inverter cross-check

Meter readingInverter readingDeviation threshold

Question: Is this week's generation data abnormal? Cross-check meter and inverter readings.

Chart evidence

Meter vs inverter deviation

10.72% lifetime deviation

Day0.4 %

Month1.3 %

Year1.8 %

Lifetime10.72 %

Only the lifetime window breaches the tolerance band, so the issue is likely historical or metering-related.

Sanitized reply

The answer compares meter and inverter generation across daily, monthly, yearly, and lifetime scopes, then applies the same deviation formula to every scope.

Evidence: The replay compared day, month, year, and lifetime meter-versus-inverter generation, applying a 2% deviation threshold and keeping the calculation formula visible.
Outcome: Short-cycle readings stayed within threshold, while lifetime data showed a -10.72% deviation and triggered a recommendation to inspect metering circuits or line losses.
Website use: A strong public proof point for traceable AI output and transparent data-quality governance.; /cases · /security

Sanitized replayO&M action verificationO&M verification

Cleaning effect assessment

CleaningBefore-afterIrradiation caveat

Question: How did generation change before and after module cleaning?

Chart evidence

Seven-day generation

63.4% seven-day uplift

Before11.32 MWh

After18.5 MWh

The uplift is visible, but the caveat keeps the claim credible.

Sanitized reply

The replay treats cleaning as a hypothesis to verify, not a guaranteed cause: it compares seven days before and after cleaning while also checking PR and irradiation.

Evidence: The replay compared seven days before and after cleaning, reported generation, PR, and irradiation together, and explicitly warned that higher irradiation affected the result.
Outcome: Post-cleaning seven-day generation increased by 7.18 MWh, or 63.4%, while the answer avoided claiming the full gain came from cleaning alone.
Website use: Useful for pilot acceptance because it proves ZenovaOS AI can verify actions without overstating causality.; /cases · /pricing

Sanitized replayAlarm storm governanceAlarm governance

Repeated alarm aggregation

Alarm groupingNoise reductionSame-device clustering

Question: There are many alarms. Which ones are repeated and can be grouped?

Chart evidence

Alarm batch shape

31 grouping candidates

Reviewed100 alarms

Active-analysis72 alarms

Groupable31 alarms

A small set of grouping rules can reduce a much larger review burden.

Sanitized reply

The answer reviews the latest alarm batch by status, type, device, and time window, then identifies where repeated same-device events are creating operating noise.

Evidence: The replay summarized 100 alarms, separated recovered and active records, identified active-analysis alarms as the dominant type, and found repeated events on the same devices.
Outcome: The answer recommended grouping short-lived same-device, same-type alarms to reduce operating noise and focus the team on root-cause follow-up.
Website use: Directly supports the alarm attribution pilot and the promise of turning alarm noise into action priority.; /operations · /cases

Sanitized replayInverter trend diagnosisDevice diagnosis

Power-curve anomaly judgment

5-minute curvePeak dropHypothesis ranking

Question: The afternoon power curve looks wrong. Is it an equipment problem?

Chart evidence

Afternoon power curve

34% afternoon average decline

The shape shows a decline, but the cause still needs one more evidence check.

Sanitized reply

The answer compares today's afternoon curve with the previous afternoon at five-minute resolution, then quantifies both peak and average power decline.

Evidence: The replay compared same-day 5-minute power data against the previous afternoon, calculated peak and mean changes, and called out sudden drops without forcing a fault conclusion.
Outcome: Afternoon peak and average output were both about one-third lower than the previous day, but the answer framed cloud cover, shading, or curtailment as hypotheses to verify.
Website use: Shows a field-engineer workflow where the agent helps judge uncertainty and choose the next evidence check.; /scenarios · /agent

Next step

Turn one real case into acceptable pilot evidence first.

If a public case does not exist yet, start with a pilot package and retain sanitized replays, chart samples, and report samples.

View pilot options Co-create a case