Downloads & Quick Reference

Important

This repository is self-contained. After cloning, run ./setup.sh to set up the environment automatically.

Setup Script

Run after cloning to set up the environment:

git clone https://github.com/NAICNO/wp7-UC2-pem-electrolyzer-digital-twin.git
cd wp7-UC2-pem-electrolyzer-digital-twin
./setup.sh

Available Models

Name	Class	Parameters	Description
`teacher`	`HybridPhysicsMLP`	9,354	8 physics params + MLP residual correction
`student`	`PhysicsHybrid12Param`	12	6 physics + 6 hybrid correction params
`pure_mlp`	`PureMLP`	8,961	No-physics MLP baseline
`big_mlp`	`BigMLP`	43,393	Large no-physics MLP baseline
`transformer`	`SteadyStateTransformer`	529,793	Self-attention baseline

Command Line Options

python scripts/pem_electrolyzer/main.py [OPTIONS]

Option	Default	Description
`--mode`	`full`	Execution mode: `full`, `quick-test`, `teacher-only`, `ablation`
`--epochs`	`100`	Number of training epochs
`--batch-size`	`4096`	Training batch size
`--lr`	`0.01`	Learning rate
`--device`	`auto`	Device selection: `cuda`, `cpu`, or `auto`
`--seed`	`42`	Random seed for reproducibility
`--data-dir`	`dataset/`	Path to dataset directory
`--results-dir`	`results/`	Path to output directory

Available Datasets

Dataset	Description
`dataset/test4_subset.csv`	Training data – long-term stability test
`dataset/test2_subset.csv`	OOD evaluation – current sweep
`dataset/test3_subset.csv`	OOD evaluation – pressure swap

Example Commands

Quick Test (< 1 minute)

python scripts/pem_electrolyzer/main.py --mode quick-test

Full Training (GPU)

python scripts/pem_electrolyzer/main.py --mode full --device cuda --epochs 100

Teacher Only

python scripts/pem_electrolyzer/main.py --mode teacher-only --epochs 50

Background Training (tmux)

tmux new -s training 'python scripts/pem_electrolyzer/main.py \
    --mode full --epochs 100 2>&1 | tee training.log'

# Monitor: tail -f training.log
# Attach: tmux attach -t training

For AI Coding Assistants

If you’re using an AI coding assistant (Claude Code, GitHub Copilot, Cursor, etc.), the repository includes machine-readable instruction files:

AGENT.md – Markdown format (human and agent readable)
AGENT.yaml – YAML format (structured data for programmatic parsing)

These files contain step-by-step instructions that agents can follow to:

Set up the environment on the VM
Run the Jupyter notebook
Execute command-line experiments
Verify results

Quick prompt for your AI assistant:

Read AGENT.md and help me run the PEM electrolyzer PINN
demonstrator on my NAIC VM.
VM IP: <your_vm_ip>
SSH Key: <path_to_your_key.pem>

The agent will execute the setup and run experiments based on the structured instructions.