Sep 30, 2026 · @Sidharth
Forecasting and scheduling software earns its fee only when it changes a decision: the schedule your QCA submits, an intraday revision, a DAM or RTM bid, a battery dispatch or a maintenance window. A low average forecast error says little about whether those decisions got better.
This guide sets out the criteria that separate a forecasting tool from a platform that improves commercial outcomes, with a checklist you can put to any vendor, Arkco included. For how forecasts, physical limits and market economics fit together in operation, see our companion article, From Forecast to Dispatch.
Key takeaways
Judge software by the decisions it improves at each horizon (day-ahead, intraday revision, RTM), not by one average accuracy score.
Look for models trained on each asset's own data, and schedules chosen for commercial outcome across DSM, trading revenue and PPA compliance.
Plant availability, including outages, derating and planned maintenance, must feed the forecast before it becomes a schedule.
In India, the platform should fit your existing QCA and trader, or connect directly to IEX and the load despatch centres if you trade on your own licence.
Agree the baseline and measures before a pilot, and prefer vendors who are paid on measured results.
What decisions should the software support?
For an Indian renewable portfolio, the software has to support at least five decisions, each on its own clock. A vendor that is strong on one horizon can be weak on another, so ask for evidence on each.
Decision | When it is made | Who acts on it |
|---|---|---|
Day-ahead schedule for each 15-minute block | The day before delivery | QCA or scheduling desk, submitted to the SLDC or RLDC |
Intraday schedule revisions | Through the day, taking effect a few blocks after submission under IEGC | QCA or scheduling desk |
DAM and RTM bids on IEX | DAM the day before; RTM every half hour for delivery shortly after | Generator's trader or own trading desk |
Battery charge and discharge | Block by block, against DSM cover, market prices and PPA peak obligations | Site control room or EMS |
Maintenance windows | Days to weeks ahead, revisited as plans change | O&M team with the scheduling desk |
The useful output for each is a proposed action with a quantity, a time and a reason, not just a megawatt forecast.
Evaluation checklist
Use these ten questions to compare vendors on the same terms. The right-hand column is what a strong answer looks like.
Criterion | What to ask | What good looks like |
|---|---|---|
Asset-level models | Is the model trained on this plant's own SCADA, availability and settlement history? | A model tuned per asset and retrained as the plant changes, not one generic model across sites |
Commercial objective | What does the schedule optimise for? | Expected commercial outcome across DSM, trading and PPA compliance, with the trade-off against pure accuracy shown |
Horizon coverage | How does performance differ day-ahead, intraday and in RTM windows? | Results reported per horizon, tied to the revision or gate-closure deadline |
Uncertainty | Does it give a range or only a point value? | A range per block that actually shapes the schedule, not a chart shown alongside it |
Plant availability | How do outages, derating and planned work enter the forecast? | Live availability data adjusts capacity before the schedule is built |
Storage and markets | Can it plan BESS dispatch, DAM/RTM bids and PPA peak delivery together? | One timeline and one decision, not separate tools competing for the same MWh |
Integration | Does it work with our QCA and trader, or connect to IEX and the LDCs directly? | Fits your existing set-up, with either route available |
Control and audit | Which actions run automatically, and who approves the rest? | Set per customer at onboarding, with every action and override logged |
Proof | Are results replayed through actual settlement, at pool and SPV level? | Pre-agreed baseline, eligible blocks defined in advance, results statistically tested |
Commercial model | How is the vendor paid? | A platform fee plus a share of measured gains, so the vendor carries part of the risk |
Why is one accuracy number not enough?
A single MAE or MAPE figure averages all blocks equally, while money is lost unequally. Three things it hides are worth asking about directly.
Direction and timing of errors. A forecast that repeatedly over-predicts around evening ramps can post a respectable average and still be the most expensive forecast in the shortlist. Ask for error broken down by horizon, time of day, weather regime and ramp events, with over- and under-forecasting shown separately.
The measurement basis. Deviation for wind and solar is measured according to the regulation that applies to your asset, and that framework is moving towards measuring deviation against the submitted schedule. Ask how the vendor calculates DSM exposure today and how quickly its optimisation can follow a rule change.
The commercial unit. Results can differ between a pooling station and an individual SPV, because pooling nets deviations across plants. A vendor should report at the level where you actually settle, and should not present a pool-level number as an SPV result, or the reverse.
How should plant availability and maintenance be handled?
A weather forecast becomes a schedule only after it is adjusted for what the plant can actually deliver. Turbines offline, inverters derated and feeders under work all reduce deliverable output, even when the weather forecast is right.
Ask three things. How does live availability reach the model, and what happens when that feed fails? How does planned maintenance adjust capacity, and does the schedule update if the work moves or overruns? And can the platform recommend maintenance windows in low-value, low-risk blocks, so planned downtime costs as little generation and revenue as possible?
Arkco's platform uses availability data in every forecast and advises on maintenance timing against expected output and prices. Safety and equipment condition stay with the O&M team; the platform shows the commercial cost of each window.
How should it fit into Indian scheduling and trading?
The software should fit how you already schedule and trade, not force a new chain of custody. In India that usually means one of two routes.
Through your QCA and trader. Many generators schedule through a Qualified Coordinating Agency and trade through a licensed trader. The platform should hand them schedules, revisions and bid recommendations in the formats and on the timelines they already use.
Directly with IEX and the load despatch centres. Generators that trade on their own licence need a platform that can integrate with the exchange and the SLDC or RLDC for scheduling and bidding.
Arkco supports both. On either route, ask the vendor to show exactly where a recommendation becomes an action: which steps run automatically, which need sign-off, and what happens when a data feed fails or a curtailment instruction arrives. A good platform lets you set those boundaries at onboarding and logs every action and override for settlement reconciliation.
How do you run a fair pilot?
A pilot is only persuasive if the rules are fixed before the results arrive. Agree these in writing at the start:
Scope: which assets, which period, and whether results are judged at pool or SPV level.
Baseline: the incumbent forecast and schedules, the actual DSM settlement, and realised market prices over the same blocks.
Eligible blocks: how curtailment, outages and missing data are handled, so nothing is quietly excluded later.
Measures: forecast error by horizon, DSM charges, and trading or PPA-compliance gains where those streams exist.
Evidence: forecasts time-stamped when issued, schedules replayed through the actual settlement rules, and a test of statistical significance.
Include difficult periods such as ramp seasons, monsoon transitions and planned outages. A pilot run only in calm weather proves little.
Arkco's pilot result. In a six-month pilot on a 126 MW wind asset in India, run in parallel against real DSM settlement across 3,360 data points, Arkco's engine improved day-ahead MAE by about 27% and 90-minute-ahead MAE by about 30% over the incumbent forecast. Replayed through actual settlement, DSM charges fell by 44.9% on a pool-adjusted basis and 47.9% at SPV level (p < 0.001). The engine behind it, e2m, has been developed and run by Metro Power in Australia for about 16 years, and the India pilot was backed by the India–Australia RISE Accelerator.
Frequently asked questions
What is the best renewable forecasting software in India?
The best choice is the one that improves your settled results, not the one with the lowest headline error. Compare vendors on asset-level models, commercial optimisation across DSM, trading and PPA compliance, integration with your QCA or trader, and results replayed through your own settlement data.
What is the difference between forecasting software and scheduling software?
Forecasting software predicts generation. Scheduling software turns that prediction into the schedule, revisions and bids you actually submit. Commercial value comes from the second step, so a buyer should evaluate both together.
Do we need to replace our QCA to use a new platform?
No. A platform like Arkco can work through your existing QCA and trader. Generators that trade on their own licence can instead have the platform integrate directly with IEX and the load despatch centres.
How long should a pilot run?
Long enough to include difficult periods such as seasonal transitions, ramp events and planned outages. Arkco's India pilot ran for six months on a 126 MW wind asset.
Can forecasting software help plan maintenance?
Yes. A platform that knows expected output and prices by block can recommend maintenance windows that cost the least generation and revenue. The O&M team still decides on safety and equipment grounds.
How are forecasting and scheduling platforms priced?
Models range from flat licence fees to outcome-based pricing. Arkco charges a platform fee plus a share of measured gains against an agreed baseline, so the customer pays mainly for results that show up in settlement.