Theme 10 · AI for renewable generation forecasting

Karnataka renewable forecast console.

Day-ahead and intra-day forecasts with calibrated uncertainty, driver attribution, and flexible aggregation. Built for masked or synthetic data, with no hosted LLM dependency.

Asset type

Plant or cluster

Issue

Cluster aggregate

Next 24 hours

Pavagada Solar Park forecast

Median 90% interval Capacity

No forecast selected Choose an asset to populate the chart.

Expected energy 0 MWh 90% PI pending

Peak output 0 MW Pending

Capacity factor 0% 24 hour average

Interval width 0% Mean 90% width

Explainability

Operational drivers

13:00 IST

Driver narrative pending.

Scenario update

Weather what-if

Cloud cover delta +0% Wind speed delta +0.0 m/s GHI delta +0 W/m2 Temperature delta +0 C

Adjusted median 0 MW No movement

Aggregation

Karnataka cluster view

0 MW

Evaluation

Baselines beaten in masked replay

0% coverage

The sandbox replay adds deterministic forecast-issue perturbations so repeated synthetic data cannot make persistence artificially perfect. Replace the replay target with real SCADA to score production accuracy.

Architecture

Layer beside existing systems

1Weather + masked SCADANWP, satellite, plant meter history

2Global learnerAsset, geography, season and physics features

3Quantile calibration50% and 90% intervals by horizon

4Operator outputsForecast, drivers, reserve-aware ranges