LIVE_GRID_STATUS
LOAD
87.3%
VOLTAGE
1.020pu
FREQ
50.01Hz
RENEWABLE
34.2%
Grid Control Systems

RL-Powered
Topology Switching
& Redispatch

Real-time grid control with Grid2Op and LightSim2Grid. Maintain N-1 security under variable renewables and contingencies.

RTE France
L2RPN Validated
N-1 Security
Action Masking

Learn policies that keep the grid secure and economical in real time — topology switching, redispatch, storage dispatch under uncertainty.

Operational Challenges
01
Variable Renewables
02
N-1 Security
03
Real-Time Constraints
04
Topology Complexity
The Problem

Static rules and offline optimizers struggle with fast, sequential decisions. Power systems must maintain reliability while integrating renewables, uncertain demand, and contingencies under strict N-1 security constraints.

RL Solution

Policies learn bus-bar reconfigurations, redispatch optimization, and curtailment decisions with safety shields and action masking — validated on Grid2Op digital twins before deployment.

Grid2Op

RTE France • L2RPN • Production-Grade
Framework

Open-source platform created with RTE (France's TSO) for intelligent power-grid control. Sequential decision-making with Gym-like APIs.

LightSim2Grid

C++ power-flow solver accelerates AC/DC calculations. Enables fast safety checks and distributed RL training at grid operator speed.

L2RPN Competitions

Powers Learning-to-Run-a-Power-Network challenges. Operator-aligned tasks with long-horizon evaluation and N-1 security constraints.

REPOSITORY
github.com/Grid2op/grid2op

Core simulator with operational constraints

BACKEND
github.com/Grid2op/lightsim2grid

Fast C++ backend for large-scale RL

System

Telemetry → Operations

SCADA/EMS
Telemetry
Feature Store
States & forecasts
Grid2Op Twin
Digital simulation
RL Stack
PPO/SAC/DQN
Policy Serve
Shadow→Live
Monitor
Governance
Telemetry (SCADA/EMS)
↓
Feature Store (states, contingencies, forecasts)
↓
Grid2Op Digital Twin ⇄ LightSim2Grid
↓
RL Control Stack (safety layers, action masking)
↓
Policy Serving (shadow → canary → gated live)
↓
Monitoring & Governance

Observation → Action → Reward

OBSERVATION
→ Line loading % (thermal limits)
→ Bus voltages/angles
→ N-1 post-contingency estimates
→ Generator limits & ramp rates
→ Renewable forecasts
→ Topology status
→ Storage SoC
→ Locational marginal prices
ACTION
Topology Switching
Bus-bar reconfiguration
Redispatch
Within P/Q capability
Curtailment
Load/RES shedding
Storage/DR
Battery dispatch
REWARD
Long-horizon uptime
Penalty: Overloads
Penalty: Voltage violations
Penalty: Load shedding
Penalty: Redispatch cost
Penalty: Topology churn
RL Core

Algorithms

PPO / SAC
Continuous control for generation redispatch with ramp limits
DQN / QR-DQN
Discrete topology switching decisions
Hierarchical
Supervisor picks mode, specialized policy executes
Safety

Multi-Layer Shields

Action Masking
Forbid unsafe actions causing overload or islanding
Pre-Check Simulation
Fast solve via LightSim2Grid before commit
Ramp & Interlock
Hard constraints on generator rates and protection
Training Strategy
01 → Congestion
Fixed topology
02 → Contingency
N-1 scenarios
03 → Stochastic
Random outages
04 → Co-optimize
Storage + DR
Data Pipeline

Scenario Engineering & Fidelity

01
Historic Replays
Load/RES Time-Series
Actual historical demand and renewable generation patterns
Outages/Switchings
Real operational events and topology changes
Market Redispatch Prices
Historical locational marginal prices and redispatch costs
02
Synthetic Stress Testing
N-1/N-k
Contingency scenarios
Line Deratings
Thermal limit reductions
RES Ramps
Renewable variability
L2RPN Benchmarks
Standardized tasks
03
Fidelity Levers
Power-Flow Models
→ DC vs AC power-flow
→ Solver tolerances
→ Protection trip curves
Control Models
→ Tap-changer models (fixed/controllable)
→ Generator capability curves
→ Ramp rate constraints

RL2Grid and L2RPN academic benchmarks (built on RTE Grid2Op) illustrate standardized tasks and constraints

Evaluation & KPIs

Reliability
% time within thermal/voltage limits
Secure operation rate
Security
N-1 survivability
Post-contingency feasibility
Resilience
Overload clearance time
Response speed under faults
Cost
Redispatch/Curtailment cost
Economic efficiency
Stability
Topology churn / action rate
Operator friendliness
Market
Congestion rent reduction
Coordinated with market outcomes
Robustness
Performance under unseen outages
Generalization (L2RPN-style)
L2RPN Competition Standards

L2RPN competitions and public studies evaluate these dimensions using Grid2Op as the leading RL testbed for realistic grid control. Episodic scoring aligned with operator goals and long-horizon security evaluation.

Validation Layers
→ Offline backtesting on historical data
→ Digital twin stress testing
→ Shadow mode recommendations
→ Canary deployment with rollback triggers

Case Blueprints

CASE A

Congestion Management

A
Goal
Clear thermal overloads with minimal cost
Approach
Grid2Op + PPO (redispatch) + DQN (topology) + LightSim2Grid safety gatekeeper
CASE B

Contingency Response (N-1)

B
Goal
Survive single-element outages
Approach
Preventive topology + corrective redispatch + RL2Grid stress tests
CASE C

Renewable Coordination

C
Goal
Reduce curtailment via storage
Approach
Grid2Op with storage + charge/discharge + topology co-optimization
Evaluation Metrics
Reliability
Security
Resilience
Cost
Stability
Market
Robustness
L2RPN competitions evaluate these dimensions using Grid2Op as the leading RL testbed

MLOps & Deployment

Shadow
Recommendations only
Canary
10% feeders
Live
Full deployment
Training
→ Distributed rollouts
→ LightSim2Grid acceleration
→ Model registry
Observability
→ Overloads avoided
→ Action audits (SHAP-like)
→ Constraint alerts
Governance
→ Immutable logs
→ Config diffs
→ Operator override
Config Example
environment:
  platform: grid2op
  backend: lightsim2grid
  timestep_min: 5
  security: N-1
  actions: [topology_switch, redispatch, curtailment]

safety:
  action_mask: on
  ramp_limits: on
  thermal_margin: 95%

rl:
  topology_agent: dqn
  redispatch_agent: sac

deployment:
  mode: shadow
  canary_share: 0.1
  rollback_on: [limit_compliance < 99.5%]

Deploy RL-powered grid control with Grid2Op

Secure, economical real-time operations with topology switching, redispatch, and N-1 security compliance.

Limit Compliance99.5%
Security StandardN-1
Action Latency<5ms
ValidationL2RPN