How ikramkhaecu/madrl_blackstart works
This repository is a small PI-MADRL black-start restoration simulation: a training script runs episodes in a restoration environment using DQN or PPO agents, then saves training history and an episode replay for inspection. The environment models switching, generation and load restoration, battery state, observations, and reward. The README describes a much broader v2 framework, but its listed modules are absent from this tree; the graph therefore reflects the available implementation and treats visualization wiring as unverified.
Main entry points: train.py Training runner · visualizer.html Replay visualizer · blackstart_env.py Black-start environment · dqn_agent.py DQN agent
Components 7
Training workflow
- Training runner
scripts/train.py
Restoration simulation
- Black-start environment
src/envs/blackstart_env.py
RL agents and results
- DQN agent
src/agents/dqn_agent.py - PPO agent
src/agents/ppo_agent.py - Training and replay files
scripts/train.py - Replay visualizer
static/visualizer.html
Other components
- Experiment operator
Connections 6
- Experiment operator → Training runner: runs training
scripts/train.py - Training runner → Black-start environment: steps episodes
scripts/train.py - Training runner → DQN agent: trains with
scripts/train.py - Training runner → PPO agent: trains with
scripts/train.py - Training runner → Training and replay files: writes results
scripts/train.py - Replay visualizer → Training and replay files: may inspect