GridOS by BECX

AI-Powered Grid Intelligence

Forecast demand at <5% MAPE. Optimize power procurement with MILP. Detect outages in 60 seconds. 7 autonomous AI agents.

Physics-informedOpen-source coreHuman-in-the-loop
GridOS Control Room
LIVE

Delhi SLDC · 73 substations · 137 lines

Frequency

50.02Hz

nominal

Load

2,840MW

Solar

312MW

24h load forecastMAPE 4.1%
7 agents active

The status quo

The Problem

Indian grid operators run on spreadsheets, react to outages after the fact, and cannot attribute where revenue leaks.

12-15% Forecast Error
Industry average MAPE causes over-procurement penalties and DSM charges.
₹75,000 Cr AT&C Losses
Indian DISCOMs lose revenue annually with no attribution between technical and commercial.
15-30 min Outage Detection
Operators learn about outages from consumer calls, not from the grid itself.

The GridOS platform

One Platform. Six Modules.

From physics-informed forecasting to autonomous agents — every module shares the same grid model and telemetry stream.

Forecast Engine
Spatio-temporal GNN on grid topology. 96-step, 24h forecast at 15-min resolution. <5% MAPE vs 8-12% industry average.
Dispatch Optimizer
HiGHS MILP 24h rolling MPC. Same solver as CAISO/PJM. Piecewise-linear heat rates, BESS cycling, multi-objective scoring.
Digital Twin
AC load flow (Newton-Raphson + BFS). WLS state estimation at 5-15% sensor coverage. N-1/N-2 contingency in <5 seconds.
GeoAI
Satellite rooftop solar detection (ResNet-UNet). 280 MWp mapped from 47,000 tiles. Feeds directly into load flow + forecast.
AI Agents
7 autonomous agents: forecasting, procurement, outage triage, maintenance, compliance, market intel, copilot. Human approval for critical actions.
GenAI Copilot
Natural language grid queries. Auto-generated DERC/CEA regulatory reports. Pluggable LLM (OpenAI/Claude/Ollama).

End-to-end

How It Works

Ingest every signal, analyze with physics-informed models, and act with human oversight.

01
Ingest
  • SCADA (15s)
  • AMI (15min)
  • Satellite
  • Weather (IMD/ECMWF)
  • IEX Market

Streamed via Kafka

02
Analyze
  • GNN forecast
  • AC load flow
  • State estimation
  • MILP optimization

Physics-informed pipeline

03
Act
  • Dispatch plan
  • DAM bid
  • BESS schedule
  • Outage response
  • Regulatory reports

Human approval for critical actions

Proof points

Numbers, not narratives

Real topology, real satellite coverage, real targets.

<5% MAPE

Forecast Accuracy

vs 8-12% industry average

73

Substations

Delhi 400/220kV real SLDC topology

280 MWp

Rooftop Solar

Mapped from 47,000 satellite tiles

3-8%

Cost Reduction

Projected procurement savings (pilot target)

7

AI Agents

Autonomous grid operations with human oversight

Engineering

Built on proven technology

Open-source, auditable, and battle-tested in production grid software.

GNN (Graph Neural Networks)HiGHS MILPPyPSAKafkaFastAPIYOLOv8LLM (OpenAI/Claude/Ollama)IEC 61850TimescaleDB

Ready to transform your grid operations?

Deploy GridOS in a 12-week pilot and validate forecast accuracy, procurement savings, and outage response on your network.