GridGuardian

GridGuardian is an AI-native grid operation control room for the E.ON Energy / AI Hackathon.

It is a modular, explainable, safety-constrained operator decision-support system. It is not a chatbot and it is not production autonomous grid control. The product helps a human operator detect, simulate, validate, recommend, and explain grid-operation decisions.

Demo Case

Demo Video

Watch the GridGuardian demo GridGuardian is demonstrated through the Munich AI Factory Grid Stress Case.

This is a fictional but realistic grid-planning scenario inspired by growing AI/data-center and EV charging demand in urban electricity networks.

Fictional customer: IsarGrid Operations GmbH Location: Munich Tucherpark / Munich South grid corridor Scenario: AI Factory baseload + EV evening peak + low solar output Status: High-risk connection case under validation

Case Assumptions

• AI Factory initial baseload: 42 MW
• AI Factory expansion case: 65 MW
• Flexible AI training workload: 8–12 MW
• EV chargers: 10,000
• Charger power: 11 kW
• EV utilization: 65%
• EV charging window: 18:00–23:00
• EV added peak load: 71.5 MW
• Combined new grid stress: 113.5 MW
• Target grid frequency: 50 Hz

GridGuardian evaluates whether this new load can be safely connected under stressed evening-grid conditions.

Hackathon Story

GridGuardian supports both E.ON challenge directions.

Challenge A: Grid Operation Agents

GridGuardian helps operators monitor stressed grid states, coordinate regional agents, detect overloads, screen N-1 contingencies, recommend corrective actions, validate actions, and explain results before an operator acts.

Challenge A is demonstrated through:

• Command Center
• N-1 Guardian
• Night Shift
• Operator Copilot
• Agent Benchmark Lab
• Safety and corrective-action logic

Challenge B: Grid Foundation Models

GridGuardian adds a foundation-model-assisted workflow for fast what-if screening, rapid risk prediction, scenario ranking, and solver-style validation.

Challenge B is demonstrated through:

• GridFM Guardian
• EV Scenario Simulator
• AI Factory + EV what-if screening
• prediction vs validation comparison
• route-to-N-1 workflow

The core idea is:

Fast prediction where speed matters.
Physics-style validation where safety matters.
Human explanation where trust matters.
Operator approval where responsibility matters.

Product Modules

• GridGuardian Command Center
• GridGuardian N-1 Guardian
• GridGuardian Corrective Engine
• GridGuardian GridFM Guardian
• GridGuardian EV Scenario Simulator
• GridGuardian Night Shift
• GridGuardian Data Import Center
• GridGuardian Operator Copilot
• GridGuardian Agent Benchmark Lab
• GridGuardian Physics Engine
• GridGuardian Safety Layer
• GridGuardian Scenario Lab
• GridGuardian Foundation Engine

Architecture

Data Import Center
        ↓
EV Scenario Simulator
        ↓
GridFM Guardian
        ↓
N-1 Guardian
        ↓
Corrective Engine
        ↓
Night Shift
        ↓
Operator Copilot
        ↓
Human Operator Decision

GridGuardian connects both E.ON challenges in one workflow:

GridFM Guardian rapidly screens many possible scenarios.
N-1 Guardian identifies dangerous contingencies.
Corrective Engine proposes staged operator actions.
Physics Engine validates critical results.
Operator Copilot explains the result.
Human operator remains in control.

Key Demo Values

For the Munich AI Factory Grid Stress Case:

• AI Factory baseload: 42 MW
• EV charging peak: 71.5 MW
• Combined new stress: 113.5 MW
• Unmanaged loading: 1.58x
• N-1 contingency loading: 1.86x
• Stage 1 corrective action: 1.41x
• Stage 2 safe plan: 0.94x
• Final status: safe after staged mitigation

The staged safe plan combines:

• corridor switching
• smart EV charging
• AI workload deferral
• local storage support
• operator approval

Run Backend

cd gridguardian/backend
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

Backend URL:

http://localhost:8000

Health check:

http://localhost:8000/api/health

Run Frontend

cd gridguardian/frontend
npm install
npm run dev

Frontend URL:

http://localhost:3000

The frontend reads:

NEXT_PUBLIC_API_URL=http://localhost:8000

Demo Script

1. Open GridGuardian Command Center.

2. Show the active case: Munich AI Factory Grid Stress Case.

3. Explain the combined grid stress: 42 MW AI Factory baseload + 71.5 MW EV peak = 113.5 MW new load.

4. Open GridFM Guardian and show fast screening of the AI Factory + EV scenario.

5. Show predicted overloaded assets, confidence, inference time, and solver-style validation.

6. Open N-1 Guardian and show the line_17 contingency.

7. Show staged mitigation:

   • before action: 1.86x
   • stage 1: 1.41x
   • stage 2 safe plan: 0.94x

8. Open EV Scenario Simulator and show the 10,000 charger what-if case.

9. Open Night Shift and start the overnight autonomous supervision simulation.

10. Open Operator Copilot and ask: “Can IsarGrid approve the AI Factory connection?”

11. Open Data Import Center and show that CSV/JSON scenario data can be validated and registered.

12. Finish with Agent Benchmark Lab to show evaluation before trust.

Pages

• Command Center
• N-1 Guardian
• GridFM Guardian
• EV Scenario Simulator
• Night Shift
• Data Import Center
• Operator Copilot
• Agent Benchmark Lab
• Architecture

Documentation

• docs/ARCHITECTURE.md
• docs/PITCH.md
• docs/ROADMAP.md
• docs/CREDITS.md
• docs/LICENSES.md

Positioning

GridGuardian is a working hackathon prototype, not production grid-control software.

It demonstrates the full operator workflow:

import data
simulate new load
screen with a foundation-model-style layer
validate N-1 contingencies
recommend staged actions
simulate Night Shift supervision
explain through an Operator Copilot

A production deployment would require real grid topology, utility data, certified solvers, cybersecurity review, and integration with operational systems.