Autonomous AI agents that forecast, schedule, and submit revisions to the grid every 15 minutes
OpsAgent runs two coordinating AI agents: a Meteo Agent that pulls fresh weather data (cloud cover, irradiance, rain) for every plant, and a Solar Agent that reads the weather forecast alongside live SCADA actuals and decides whether to submit a corrected schedule or skip. The loop runs every 15 minutes, 24x7, with every decision logged and visible to operators.
Manual forecasting shifts miss the 03:00 AM storm and forget the day-ahead deadline.
Human forecasters work in shifts. They get tired, they take breaks, and they go home at night. A sudden weather front at 03:00 AM goes unnoticed until the morning operator logs in and discovers the plant has been generating outside the penalty band for hours.
Day-ahead schedule deadlines are missed because the person responsible is in a meeting. Intra-day revision windows pass because the control room is managing an unrelated substation trip.
Every missed revision is money lost to DSM penalties: STU Rajasthan charges 35% revenue-loss for deviations outside its 15%–25% tolerance band.
How OpsAgent operates in production
Meteo Agent pulls fresh weather
Every 15 minutes, the Meteo Agent fetches cloud cover, irradiance, and rain data for every plant from multiple NWP sources and satellite feeds.
Weather forecast published to shared inbox
The Meteo Agent publishes a structured weather forecast and briefing onto a shared MongoDB message bus, readable by the Solar Agent.
Solar Agent reads weather, schedule, and SCADA
The Solar Agent reads the peer weather message, the current submitted schedule, and live SCADA generation actuals from RealSync.
Deviation check against DSM no-penalty band
The Solar Agent decides: is the existing schedule still within the state-specific DSM no-penalty band? If yes, skip. If no, generate a corrected forecast and submit a revision.
Revision submitted or skipped with reasoning
Every decision (submit or skip) is logged with its full reasoning chain, visible to operators in the dashboard console. LLM prompts and responses are retained for audit.
Agents brief each other in plain English
Each agent ends its run by briefing the other in 2–3 plain-English sentences. Operators can read the conversation history to understand why a revision was or was not submitted.
Engineered capabilities without the fluff
Autonomous 15-minute loop
- Runs every 15 minutes, every day, including weekends and holidays.
- Does not miss the 03:00 AM storm or the day-ahead gate closure deadline.
- Conserves revision budget: STU allows 8 revisions/day, CTU allows 15/day.
- Coordinates two independent agents through a shared message inbox with no single point of failure.
DSM-aware decision engine
- Knows the per-state DSM no-penalty band (e.g. Rajasthan 15%–25%, Gujarat 7%–15%).
- Optimises for ₹ exposure, not just MAPE: a 12% deviation in Rajasthan is a skip, in Gujarat it triggers a revision.
- Conserves revision slots for when they matter: saves revisions for genuine weather-driven deviations.
- Built-in penalty model ported 1:1 from the same formula library the RealSync dashboard uses.
Audit trail and observability
- Every step is logged: weather pull, deviation calculation, decision, and submission result.
- All reasoning is visible in the operator console in real time.
- LLM prompts and responses are stored and retrievable for regulatory audit or dispute filing.
- Agent-to-agent briefings are written in plain English for non-technical operators to read.
Agent Infrastructure
Same REFlux forecasting engine, now running autonomously every 15 minutes without human intervention
Questions regarding OpsAgent
Deploy OpsAgent across your portfolio
Schedule a 30-minute operational demo with our domain engineering team. We will review your plant single-line diagram and current SLDC submission workflows.