Solar PV performance monitoring that finds, explains and prices every lost megawatt-hour
Find the megawatt-hours the desert dust is taking.
See one case, screen by screen ↓Threshold alarms catch an inverter that stops. They miss the slow losses: dust building up 0.3–0.5 points a day, an inverter derating in the afternoon heat, tracker rows left flat after a storm stow, strings drifting apart. At Dunemark, the 960 MWdc sample plant, they come to 284 MWh on a normal day — $6,816 at a $24/MWh energy value — and most of that energy trips no alarm.
Cleaning runs on a fixed calendar, crew blocks every 30 days and robots every 3 nights, so after a dust storm the calendar is weeks late and in a calm month it cleans clean panels. The performance analyst rebuilds expected energy and the loss table by hand from SCADA exports, and by the time the monthly report shows the gap, the O&M contractor's response clock and the availability guarantee month have gone by.
Estimated split of one analyst's week on one plant, in hours, by hand and with the solution.
Six specialist agents find, explain and price every lost megawatt-hour; the solar asset manager decides.
Dunemark Solar Park is Lumina Utilities' 960 MWdc, 800 MWac plant: 12 blocks, 48 inverter stations, 1,641,600 modules on single-axis trackers. On Friday 9 October a dust storm crossed its western half. It is Tuesday 13 October, 07:10. Here is what happens next, screen by screen, in the working solution.
Nadia owns yield and revenue for Dunemark. Her morning opens on one waterfall: 6,120 MWh expected from the 7.62 kWh/m² the plant actually received, 5,720 MWh at the revenue meter. The plant was 400 MWh short of expected (6.5 %), and 284 MWh of it is recoverable — $6,816 at $24/MWh. Soiling is the largest step down: 174 MWh, 2.8 %, $4,176, with blocks 4–7 at 5.3–5.6 % and the robot blocks at 1.4–1.7 %.
Built between 05:40 and 05:46 by six agents from 1.94 M SCADA readings — pyranometer P-07 left out, 3.1 % low against its neighbours.
Behind the 67 MWh inverter loss ($1,608) are three causes. INV-09-3 tripped on a DC ground fault from 10:40 to 15:10 (48 MWh), with insulation still at 0.38 MΩ. INV-02-1 is derating 12 % above 41 °C on a failed cooling fan (14 MWh). Two combiner inputs, CB-05-11 and CB-07-04, have open string fuses (5 MWh). Each one links to its fault case.
“Alarms caught the INV-09-3 trip. They did not catch INV-02-1: a derating inverter raises no alarm — it shows only against its sister inverters in the afternoon heat.”
The 55 MWh curtailment step is hatched: by design or excused. The analyst matched it to grid instruction GC-1012-04 from Lumina Grid Control, which capped the plant at 720 MW from 12:10 to 13:40. Under the setting “Deemed energy, claimable”, the 55 MWh is paid as deemed energy, and Nadia can add it to the deemed-energy claim from the side panel.
Unexplained: 12 MWh (0.2 %) — under the 0.5 % limit, so it stays unexplained rather than being moved into a named loss.
Blocks 4–7 are crew-cleaned, 547,200 modules, last cleaned 28 September and next due on the fixed date of 28 October. The dust is taking 100 MWh a day that a clean would return — $2,400 a day. A wet clean costs $13,954: crew labour $11,280, 219 m³ of desalinated water $679, vehicles and supervision $1,995. It pays back in 5.8 days and nets $20,846 by 28 October, when soiling would otherwise have reached 8.4 %.
“The calculator re-runs every morning after the soiling estimate. A plan is proposed only when it pays back inside 10 days.”
The same calculator says when a clean does not pay. On blocks 8–12, an extra robot pass every sixth night would lower average soiling by 0.31 points: $190 a day returned for $205 a day of extra passes. The robots already ran two extra passes after the storm on 10 and 11 October. Block 3's higher soiling is robot RB-03-14 stuck on its rail — a fault case, not a schedule change.
“Keep every 3 nights — every 2 nights would cost more than it returns.”
The plan books Sandline crews A–D for the nights of 13–14 and 14–15 October, reserves 219 m³ of water and moves the fixed date from 28 October to 13 November. The confirmation shows the cost, the daily return, the payback and the net by 28 October. Nadia's limit is $25,000; above it, Hessa Karim, head of generation, co-signs.
“It is under your $25,000 limit, so no co-signature is needed. It is written to the record with the calculator's inputs.”
Only after Nadia sets the plan does the planner book: work orders WO-88412 for blocks 4 and 5 tonight and WO-88413 for blocks 6 and 7 tomorrow night, 219 m³ from Lumina Water on tankers T-4471 and T-4472, and the fixed date moved to 13 November. At 07:26 Sandline Services confirms: Ravi Menon, crews A and B on site at 19:30.
The planner proposes only — it never books a crew or a robot run until the asset manager sets the plan.
Inverter, combiner, tracker and robot faults become cases for Sandline Services, the O&M contractor: 6 open, 111 MWh a day at stake ($2,664), 2 ready to send, 1.1 days average to fix in October. FC-2619 — 46 tracker rows in block 6 flat at 0° after the storm stow — has a crew on site. FC-2618, the INV-02-1 fan, and FC-2621, robot RB-03-14, wait for a person to send them.
“Energy at stake = MWh a day the fault costs until it is fixed, at $24/MWh.”
The case shows INV-09-3's power against its sister inverters on 12 October, with the 48 MWh trip shaded. The evidence: alarm ALM 0412, DC insulation fault at 10:40; insulation resistance 0.38 MΩ after the 15:10 reset against a 1.00 MΩ trip limit; and a similar past case, FC-2533 on INV-09-1 in August 2026, a cracked DC cable — 86 % match. Ravi acknowledged it at 11:02.
“The case closes only when SCADA shows the energy back for a full day. The contractor's word is not enough, and no agent closes a case.”
The O&M contract checker classed all 12 outages from 1 to 12 October: 9 contractor-caused under clause 7.2, 3 excused under clause 7.3 — Lumina Grid's feeder and switch work and 7 grid instructions. Energy-based availability is 98.70 % against the 99.0 % guarantee: 954 MWh lost to contractor-caused outages, 220 MWh below the guarantee, $5,280 in liquidated damages accrued. The notice goes to Hessa for approval.
The contract checker drafts only — the head of generation approves every notice.
The same morning after a dust storm, seen by the four people who own the plant's energy — what their week looked like, and what it looks like now.
Six specialist agents find, explain and price every lost megawatt-hour; the solar asset manager decides.
Works out what each block should have produced from the sun it actually received — measured plane-of-array irradiance, module temperature, commanded tracker angles and the plant model with its module, inverter and cable losses and degradation.
Splits the gap between expected and metered energy into soiling, inverter faults, clipping, tracker faults, availability, curtailment and unexplained, using alarms, power curves against sister inverters, tracker angles, grid instructions and the soiling estimate, and publishes the waterfall.
Estimates soiling per block from the daily soiling ratio of the six soiling stations and, for blocks without a station, from string current recovery after the last clean.
For each group of blocks, works out the energy a clean would return each day, the cost of a crew clean or of extra robot passes, and the payback; proposes a plan when it pays back within the limit, and says when not to clean.
Turns inverter, combiner, tracker and robot faults into cases with the asset, the fault, the evidence, similar past cases and the energy at stake a day, ready for the O&M contractor, and checks recovery from SCADA each morning.
Classes every outage as contractor-caused (clause 7.2) or excused (clause 7.3) under the O&M agreement, computes energy-based availability month to date and the liquidated damages, and drafts the monthly notice.
Sets cleaning plans; head of generation co-signs above $25,000. The agents propose; a named person decides.
Ask about yesterday's losses, a block, a fault case, cleaning or the O&M contract — or say what to change. Every answer points to its source.
Why is soiling so high in blocks 4 to 7?
The 9 October dust storm, on the four blocks that wait for crews. PM10 peaked at 1,850 µg/m³ from the west-north-west, and the soiling stations in blocks 4, 5 and 7 have read 5.3–5.6 % since. The robot blocks ran two extra passes and are back at 1.5–1.7 %; blocks 4–7 are not due until 28 October. A clean now pays back in 6 days.
What did INV-09-3 cost us yesterday?
48 MWh — $1,152. It tripped on a DC insulation fault at 10:40 and ran again only after the 15:10 remote reset. Insulation is still 0.38 MΩ against a 1.0 MΩ limit, so it can trip again. The same alarm on INV-09-1 in August was a cracked DC cable. Sandline acknowledged at 11:02.
Should the robots run every night this week?
No — keep every 3 nights. On blocks 8–12, every 2 nights would return about $190 a day for $205 a day of extra passes. Block 3's higher soiling is robot RB-03-14 stuck on its rail, not the schedule — that is case FC-2621.
Are we owed damages under the O&M contract?
Yes: $5,280 so far this month. Energy-based availability is 98.70 % for 1–12 October against the 99.0 % guarantee — 954 MWh lost to contractor-caused outages, 220 MWh below the guarantee. Curtailment and Lumina Grid's feeder work are excused and not counted. The notice needs Hessa Karim's approval.
13 screens from the working solution, on its sample data. Pick one to see it large.
The energy value, cleaning costs, thresholds, approvers and O&M terms are settings, chosen by the utility for each plant.
| Setting | Default | Choose from |
|---|---|---|
| Energy value for losses | $24.00/MWh | $18.00 · $24.00 · $31.00 per MWh — the avoided generation cost used in every dollar figure |
| Curtailment | Deemed energy, claimable | Deemed energy, claimable · Lost, not claimable |
| Unexplained loss that asks the analyst | 0.5 % | 0.3 % · 0.5 % · 1.0 % — the waterfall still publishes; the gap is sent to a person |
| Inverter derating to open a case | 5 % | 3 % · 5 % · 8 %, against its sister inverters over 2 hours |
| Propose a plan when it pays back within | 10 days | 7 days · 10 days · 14 days — the calculator stays quiet above this |
| Keep a fixed date as a backstop | On | On · Off — a clean at least every 45 days, whatever the return |
| Cleaning plans set by the solar asset manager | Up to $25,000 | Up to $25,000 · up to $15,000 — above the limit, the head of generation co-signs |
| Availability guarantee | 99.0 % | 98.5 % · 99.0 % · 99.5 %, energy-based, each calendar month; inverter response 12 h · 24 h · 48 h |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample plant data · “target” = the design goal, measured in the live solution · “estimated” = our arithmetic: 2,080 GWh a year × 1.5 % × $24/MWh ≈ $750,000; 3 fewer crew cleans × $13,954 ≈ $42,000 a year on blocks 4–7; the analyst's hours are an estimate · “typical” = published figures: IEA PVPS Task 13, Soiling Losses (2023) — soiling costs 3–5 % of PV energy worldwide, about €3–5 billion a year; Raptor Maps, 2025 Global Solar Report — 5.77 % underperformance from equipment issues in 2024; QEERI desert soiling tests — about 0.5 % loss a day, cost-optimal clean near 10 % loss. Lumina Utilities, Dunemark Solar Park, Sandline Services and the people named on this page are fictional — characters and sample data in the working solution.
Software that compares what a solar plant should have produced from the sun it received with what it delivered, and explains the difference. Solar PV Performance Monitoring adds six agents that attribute every lost megawatt-hour to a named cause, estimate soiling block by block, price each clean against the energy it returns and turn faults into cases for the O&M contractor. The solar asset manager sets every plan and sends every case.
Each morning the expected energy modeller works out yesterday's expected energy from measured irradiance, module temperature, tracker angles and the plant model. The loss attribution analyst splits the gap to the revenue meter into soiling, inverter faults, clipping, tracker faults, availability, curtailment and unexplained. Recoverable losses are drawn solid, excused and by-design losses hatched, and each step opens the blocks, inverters and grid instructions behind it.
The cleaning return planner weighs the energy the dust is taking each day against the cost of a crew clean or extra robot passes, and proposes a plan only when it pays back within the limit you set — 10 days in the sample plant. It also says when not to clean, and a fixed backstop date can stay in place.
The soiling estimator reads the daily soiling ratio from each soiling station, comparing clean and soiled reference modules. For blocks without a station it estimates from string current recovery after the last clean, and it says so when a block has no station and no clean in 45 days.
Yes. The O&M contract checker classes every outage as contractor-caused or excused under the agreement, computes energy-based availability month to date and the liquidated damages, and drafts the monthly notice. In the sample plant that is 98.70 % against a 99.0 % guarantee and $5,280 accrued; the head of generation approves the notice.
Yes. The agents model, attribute, estimate, propose and draft — they never book a crew without a set plan, send a case, close a case or send a claim. The asset manager sets plans up to $25,000, the head of generation co-signs above that and approves every notice, and SCADA data is read-only to the agents.
Yes. The energy value for losses, curtailment treatment, the unexplained-loss and inverter-derating thresholds, the morning run time, cleaning and water costs, the payback limit, the backstop date, the approval limits and the O&M guarantee and response times are all settings, with every change recorded.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your plant model, SCADA tag list, cleaning rates and O&M agreement — and it goes live once every quality gate has passed. We will walk you through it on your own plant's data first.
We'll run Solar PV Performance Monitoring on your own plant's SCADA, soiling and cleaning data.