A renewables fleet fails quietly before it fails loudly. One month at BrightPlain Energy's fleet desk, 84 wind turbines, a 200 MW solar farm, a 100 MW / 200 MWh battery: a gearbox bearing signature on WTG-A17 that says six weeks to failure, caught while a planned $380K swap is still possible instead of a $980K run-to-failure with a crane remobilization; a thermal sweep that finds 61 defective solar strings bleeding 2.1 MW nobody was counting; a battery being cycled to death for pocket-change arbitrage; and the trick that ties it together, the forecast finds the low-wind week, so the crane works while the wind doesn't blow. Then a front arrives on day 24 and settles the argument between the two ways to run a fleet.
| Without | With GridCORTEX | Δ |
|---|
| Without | With GridCORTEX | Δ |
|---|
BrightPlain Energy is a fictional company that owns and operates renewable power plants, and the simulation covers one month at its fleet desk. The fleet: 84 wind turbines across two farms (Cedar Mesa, 48 turbines of 2.4 megawatts each, and Ridgeline, 36 of 3.0 megawatts), the 200 megawatt West Mesa solar farm with 96 groups of panel strings, and a 100 megawatt battery that can store 200 megawatt-hours. The month opens at 99.1% fleet availability (the share of the fleet able to produce) and a maintenance calendar written before anyone looked at the weather. Then the health scan reads every turbine's live sensor feed (bearing temperatures, vibration patterns, metal particles in the oil) against the fleet's own history of past failures, and turbine A17 stands out: a bearing inside its gearbox is running hot with a matching vibration pattern, 94% confidence, about six weeks from failure. Letting it run to failure would cost about $980K: a rush-ordered gearbox, bringing the giant crane back at emergency rates, and roughly six weeks of lost electricity sales. A planned swap costs about $380K. And the weather forecast has already found the moment to do it: days 18 to 22 are the lowest-wind week of the month, when a stopped turbine loses almost nothing.
Four decision points follow, each approved by the fleet manager; the AI ranks and schedules, humans decide. First, the calm-week swap: order the gearbox now at normal freight, book the crane for the calm window, fold a second turbine's (B29) early bearing fix into the same crane visit, and run A17 at 10% reduced power while it waits. Second, the solar heat-camera sweep: a drone photographs every panel string, the AI compares each frame against the sunlight actually hitting the panels, and finds 61 failed strings and 3 stuck sun-tracking rows silently bleeding 2.1 megawatts; 64 location-tagged repair orders go out, the stuck rows alone return 0.9 megawatts, and the full recovery is worth about $184K per month. Third, the battery schedule reshape: the battery is being fully drained twice a day to chase small buy-low-sell-high profits, wearing it out three times faster than the profits justify and putting it on track to miss its contractual year-12 storage-capacity test by 9%. The reshaped schedule keeps revenue within 2% while cutting wear 38%. Fourth, the storm-front commitment: fine-grained weather models show a strong front arriving on day 24, and the forecast is accurate enough (about 4.1% error) that the desk confidently promises tomorrow's delivery at locked prices instead of playing it safe and underselling.
The month then proves the calls. The crane goes up at A17 on day 18 in 9 mph winds, exactly the predicted lull; A17 is back online by day 21.5 with a clean vibration reading, at the planned $380K. The front arrives on day 24 and drives the fleet to 92% of its maximum rated output; the promised delivery schedule holds at 4.1% forecast error with zero shortfall penalties, while A17, six days past its old failure date, runs flat out through the best winds of the month. Ending state: 97.8% availability with the planned swap inside that number, 2 developing faults caught early, 61 strings recovered, battery wear down 38% for only 2% less revenue, and a 30-day net impact of plus $994K. Ignore the warnings and the same month ends at 91.2% availability and a net of minus $1.19M, with A17's gearbox tearing itself apart on day 26 in the best wind of the season.
Each warning sits in a different silo, so nobody adds them up. A17's hot bearing is buried in a monthly manufacturer's report nobody has opened. The solar farm reads "within normal range" while 2.1 megawatts of dead panels hide inside that range. The trading desk books the battery's small daily profits while the wear-and-tear bill appears on no one's screen. The front arrives on day 24 and the averages-based forecast is 11% short, so the farm delivers less than it promised: about $210K in shortfall penalties. On day 26, in the highest winds of the month, A17's gearbox fails hard, scattering debris through its internals: the emergency crane is quoted 3 weeks out, the rush-ordered gearbox costs 40% extra, about $980K plus roughly six weeks of lost sales. The month closes at 91.2% availability and falling, an insurance claim open, and a battery projected to fail its year-12 capacity test by 9%. Net: minus $1.19M.
The software listens, predicts, and recommends; people approve. Turbine health prediction (use case UC 10.2) hears the gearbox six weeks before it fails. Generation forecasting (UC 10.1) finds the calm week for the crane and holds 4.1% error through the storm front. Automated solar inspection (UC 10.3) turns the invisible 2.1 megawatt bleed into location-tagged repair orders with measured payback. The battery optimizer (UC 10.4) keeps revenue within 2% at 38% less wear. Every recommendation waits for a person: the fleet manager approves the swap, the repair orders, the battery schedule, and the storm-front commitment, and the maintenance and trading teams own every decision. The month nets plus $994K, a $2.18M swing, and finishes at 97.8% availability with a $380K planned repair instead of a $980K emergency.
| KPI | Without GridCORTEX | With GridCORTEX | Delta |
|---|---|---|---|
| WTG-A17 gearboxthe failing gearbox bearing on wind turbine A17, the fault the month turns on | failed day 26, 6-wk outage | planned swap in the calm week | $600K + the season |
| Cranethe giant crane every gearbox swap needs; emergency bookings cost more and take weeks | emergency remob, 3-wk wait | booked at standard rate, day 18 | the calm week was free |
| WTG-B29 main bearinga second turbine's early bearing wear, caught while the crane was already on site | still developing, unseen | fixed in the same mobilization | one crane, two saves |
| 61 solar strings + 3 trackersfailed panel strings and stuck sun-tracking rows at the solar farm, invisible in daily totals | bleeding inside "tolerance" | 2.1 MW recovered, measured | $184K/month back |
| Battery cyclinghow hard the battery is drained and refilled each day; deep daily cycling wears it out fastest | deep cycles for pennies | degradation −38%, revenue −2% | year-12 test passes |
| Fleet availabilitythe share of the fleet able to produce power | 91.2% and falling | 97.8% with the swap inside | the number the power-sale contract pays on |
| 30-day net impactthe month's bottom line across all four decisions | −$1.19M | +$994K | $2.18M swing |
| A17 event costwhat fixing turbine A17 actually cost | $980K + lost season | $380K planned | −$600K |
| Front-day schedulethe delivery promised for the day the storm front hit; miss it and penalties apply | 11% short, $210K imbalance | 4.1% error, clean settle | a forecast that holds |
| Solar productionoutput quietly lost inside "normal" day-to-day variation | 2.1 MW hiding in averages | recovered + receipted | $184K/month |
| BESS lifeBESS is the battery energy storage system; this tracks whether it will still hold its contractually required capacity at year 12 | year-12 capacity miss (−9%) | obligation met w/ margin | the capital plan holds |
| Insurance & claimsthe claims record insurers look at when setting next year's premium | gearbox claim file opens | clean month, clean record | premiums notice |
There is no direct field exposure in a forecast, and we will not pretend otherwise. The indirect mechanism is real: a better ramp forecast means fewer unplanned fast starts and fewer plant callouts at odd hours to cover a miss, and those callouts are where the driving and the rushed work happen.
Counted in units you already track:
Desk hours spent adjusting and second guessing the vendor forecast come back to the schedulers, and the monthly error review stops being a manual build.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Imbalance and deviation charges | your own imbalance or deviation charge per MWh x the MWh of forecast error removed, with the error sized from your own settlement history |
| Reserve and commitment | unit starts avoided x your start cost, plus reserve MW carried against forecast uncertainty x your reserve cost per MW |
| Desk labor | desk hours avoided x your loaded scheduler rate |
| Curtailment and spill | curtailed or spilled MWh avoided because the schedule matched the resource x your realized price per MWh |
| Forecast subscription | any portion of your current vendor forecast you retire x its annual cost |
You pay for the forecast service and the compute behind it, for integration into your scheduling workspace and historian, and for a parallel run period where the desk compares it against the incumbent forecast before trusting it. Budget the parallel run honestly, because no desk will switch without one.
Payback is normally computed from imbalance and deviation charges avoided, because that is a settlement number you already hold line by line. Start there and treat reserve and start cost savings as the second argument.
Up tower work is elevated work inside a confined nacelle, and emergency up tower repair happens whenever the weather allows rather than when the crew is fresh. Predicting the fault means the same climb happens on a planned day with the right parts staged, and it means fewer major component exchanges done under schedule pressure.
Counted in units you already track:
Emergency response and diagnostic climb hours come back to the site technicians, and the reliability engineer works a ranked queue instead of browsing a portal.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Planned versus forced repair | repairs moved from forced to planned x the difference between your own planned and forced cost for that component, crane mobilization included |
| Lost generation | forced outage hours avoided x turbine rated output x your capacity factor for that season x your realized price or PPA rate per MWh |
| Crane and logistics | crane mobilizations avoided or combined into a single campaign x your mobilization cost |
| Technician labor | emergency response and diagnostic hours avoided x your loaded technician rate, including the overtime differential you pay |
| Downstream component damage | exchanges avoided by catching a fault before it damages the next component in the drivetrain x your exchange cost, with the catch rate set by you |
You pay for the scoped engagement that builds and runs this and for the integration to get SCADA history out of the turbine vendor's system, which is more often a contractual problem than a technical one. You also pay for reliability engineer time to confirm the first flagged turbines, and some of those will be false alarms you climb for nothing.
Payback is dominated by the gap between planned and forced major component repair plus the generation lost while waiting on a crane. Both come from your own work order and production history, which makes this one of the more auditable cases in the set.
Solar inspection exposure is heat and direct current. Technicians walk rows for hours during peak insolation because that is when thermal signatures appear, and chasing a defect means opening combiner boxes and pulling string fuses on live DC. Both go away for finding defects. Qualified technicians still do the repair, under your normal lockout and arc flash rules.
Counted in units you already track:
Walking and searching hours come back to the site technicians, and the site supervisor stops building the repair list by hand and starts approving one.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Inspection labor | walk down and search hours avoided x your loaded technician rate |
| Contracted thermal survey | the annual survey scope you stop buying x your contracted survey price per MW or per site |
| Recovered production | measured MWh recovered after each repair x your PPA or capture price per MWh, measured against the pre repair baseline rather than predicted |
| Warranty recovery | modules and trackers identified as defective while still under warranty x your average claim value per unit, which only counts if you actually file |
| Vehicle and site travel | utility vehicle and pickup miles avoided x your fleet cost per mile |
You pay for the drone platform or the flight service, for a Part 107 certified remote pilot on staff or on contract, for charging and for enough site network to move thermal imagery off the aircraft, for the GridCORTEX layer that turns imagery into a ranked defect queue, for the integration into your asset or work management system, and for supervisor time to validate detections through the first few sweeps. If you want to fly a large site from one launch point rather than repositioning, budget for a beyond visual line of sight waiver and the operational work that comes with it.
Payback is usually driven by recovered production and by the contracted survey scope you stop buying, because both are auditable against a meter and an invoice. Treat labor as secondary, since most sites will redeploy the technician rather than remove the position.
The exposure here is thermal and it is gradual, so we will describe it plainly rather than dramatically. Cells cycled hard at high state of charge and high temperature age faster, and accelerated aging is what moves a container toward the conditions people worry about. Keeping cycling inside the design envelope is a slow safety benefit, plus fewer troubleshooting trips into an energized container after an out of envelope event.
Counted in units you already track:
Daily schedule building hours come back to the storage operations engineer, and the warranty and capacity reconciliation stops being a manual exercise.
The numbers we need from you to run that formula:
| Cost driver | How it is calculated, from a rate you supply |
|---|---|
| Deferred augmentation | years you push out a capacity augmentation x your installed cost per MWh of augmentation x your cost of capital, with the deferral length coming from your own degradation curve and not from us |
| Capacity obligation shortfall | shortfall events avoided x your penalty or replacement cost per MW of unmet obligation |
| Warranty position | cycle equivalents saved against your warranty allowance x what a cycle is worth under your specific warranty terms, which only you have |
| Engineering labor | schedule build and reconciliation hours avoided x your loaded engineering rate |
| Market revenue | the revenue difference between the optimized schedule and your current schedule, valued at your own settled prices, which can go either way on any given day |
You pay for the scoped engagement that builds and runs this, for integration into the plant control system and your market scheduling, and for the commissioning work to fit a degradation model to your specific cells and warranty terms. This is also the one case in this set with a closed loop pathway, so expect your operations and protection teams to want a longer advisory only period before they enable it, and budget that time.
Payback is usually carried by deferred augmentation, because a delayed capacity addition is a large, dated capital item your finance group already tracks. Market revenue differences are real but noisy day to day and should not be the headline.
What is this, exactly? It is AI software: intelligent agents and models built and delivered by SoftServe, running on NVIDIA accelerated computing. It is not a hardware appliance and it does not replace the systems you run today. It deploys in your own cloud or on your premises, connects read-only to your existing systems, and recommends; your people approve every action, starting in shadow mode until it earns trust.
A forecast feed for the generation desk and system operators: hour-ahead and day-ahead output forecasts for each wind and solar plant, built from kilometer-scale weather models fused with each plant's own performance history. The demo above uses synthetic data; everything below describes what the real deployment needs from your organization.
| Your system | Typical products | How we connect |
|---|---|---|
| SCADA historian | AVEVA PI System, AspenTech eDNA, GE Proficy | historian mirror (one-way feed) |
| Weather and environment | National Weather Service feeds, commercial forecast services | read-only API |
| Market and grid operator interfaces | PJM, MISO, ERCOT, CAISO portals | read-only API |
| Plant control (DCS) for generation | plant SCADA, plant historian | historian mirror (one-way feed) |
| Asset / work management (EAM/CMMS) | IBM Maximo, SAP PM, Hitachi Asset Suite | database replica refreshed nightly |
| Document and knowledge stores | OEM service bulletins, end-of-warranty inspection reports | document upload |
| Geographic Information System (GIS) | Esri ArcGIS, plant as-built drawings | scheduled file export (CSV or CIM XML) |
Runs in your own cloud account on GPU instances; an on-premises NVIDIA server is an option. Connections are read-only through your existing data zone, with no connection to plant controls and no dispatch actions; the forecast runs in shadow mode next to the incumbent until it earns trust.
The Approve button you just clicked in the demo above is the real workflow. This is what it looks like on the screen of the generation desk scheduler in the GridCORTEX console:
Releasing writes the forecast into the desk's scheduling workspace through its API as an advisory feed. The desk builds and submits its own day-ahead schedule; GridCORTEX submits nothing to the market.
Fully automatic from the connected weather models and plant SCADA history, including derates the telemetry shows; there is nothing to type.
Plant telemetry streams every few minutes and kilometer-scale weather refreshes hourly with model runs every 6 hours; every forecast card shows the as-of timestamp of its inputs.
Forecasts stream to the GridCORTEX console and the desk's scheduling tools; revisions beyond a set MW threshold send a Teams push. The console runs in a browser beside your existing screens on day one; embedding into your own systems is a roadmap step once the read-only phase has earned trust. Approve, Modify, and Decline are all captured in an audit trail your compliance team can pull, and GridCORTEX never blocks or overrides anything in the systems you run today.
The fair question from any renewables O&M director: "We have the OEM's monitoring contract, a SCADA historian, a drone vendor, and an EMS for the battery, what's new here?" Here's the honest answer.
When someone asks "what did it actually calculate?", this is the list. In the simulation these factors drive the storyline; in a pilot they are computed from your fleet's data.
Presenter's one-liner: "A gearbox bearing on A17 started whispering six weeks before it would have started screaming. The fleet model heard it, the forecast found the week the wind wasn't blowing, and the crane did a $380K planned swap instead of a $980K emergency. Same month: 61 solar strings recovered 2.1 MW nobody knew was missing, and the battery stopped burning years to earn pennies. Then the front came, and the only story was that there was no story."