RefineryOpt

🛢️ Research-Grade Petroleum Refinery Optimization Platform

From Excel to Exascale: A complete toolkit for refinery blending, planning, and optimization.

Quick Start • Features • Installation • Documentation

---

🎯 Overview

RefineryOpt is an open-source Python framework for petroleum refinery optimization combining:

• Scientifically Accurate Blending — Non-linear correlations per ASTM/API standards
• Process Unit Modeling — CDU, FCC, Reformer, Hydrotreater yield models
• Environmental Compliance — EPA Tier 3, IMO 2020, RFS2 constraints
• Carbon Footprint Tracking — Scope 1, 2, 3 emissions per GHG Protocol
• Differentiable Simulation — GPU-accelerated optimization via JAX

---

🚀 Quick Start

# Clone and install
git clone https://github.com/thookham/RefineryOpt.git
cd RefineryOpt
pip install -e .

# Run simple blending optimization
python refinery_opt.py examples/gasoline.json --solver scipy

Output:

Status: optimal
Cost: $831,332.14

---

✨ Features

Core Engine (Phases 1-3)

Module	Description
Blending	RVP, viscosity, flash point, octane correlations
Process Units	CDU, FCC, Reformer, Hydrotreater models
Environmental	EPA Tier 3, IMO 2020, RFS2 constraints
Economics	Crack spreads, gross margin optimization

Advanced (Phases 4-5)

Feature	Description
Octane Modeling	Interaction matrix, Ethyl method, blending values
Carbon Tracking	Scope 1/2/3 emissions per GHG Protocol
JAX Integration	Differentiable blending with GPU acceleration

Future (Phases 6-7)

Feature	Description
Streamlit Dashboard	Web-based GUI for non-programmers
Chance-Constrained Programming	Probabilistic constraints
ML Surrogates	PyTorch neural network approximations
Decomposition	Benders, Lagrangean Relaxation

---

📦 Installation

Basic

pip install -e .

With GPU (Intel)

pip install jax==0.5.0 jaxlib==0.5.0 intel-extension-for-openxla==0.7.0

Dependencies

Package	Purpose
pyomo	Optimization modeling
pydantic	Data validation
numpy, scipy, pandas	Numerical computing
jax	Differentiable programming (optional)

---

📁 Project Structure

RefineryOpt/
├── refinery_opt.py           # CLI entry point
├── src/
│   ├── blending/             # RVP, viscosity, octane
│   ├── constraints/          # EPA, IMO, carbon
│   ├── crude/                # Assay handling
│   ├── units/                # CDU, FCC, Reformer
│   ├── economics/            # Margins, crack spreads
│   ├── uncertainty/          # Monte Carlo, CVaR, CCP
│   ├── models/               # Pyomo LP/NLP/MILP
│   ├── solvers/              # CBC, GLPK, Gurobi, SciPy
│   └── ml/                   # JAX, PyTorch surrogates
├── dashboard/                # Streamlit GUI
├── examples/                 # Sample configs
├── benchmarks/               # Li, Haverly, Pinto
└── tests/                    # Unit tests

---

🔬 Technical Details

Blending Correlations

Property	Model
RVP	(Σ xᵢ·RVPᵢ^1.25)^0.8 (Chevron Index)
Viscosity	Refutas Index
Flash Point	Inverse blending
Octane	Ethyl/Interaction matrix

Carbon Accounting

Scope	Source
1	Direct combustion (process units)
2	Purchased electricity
3	Upstream crude extraction

Solver Priority

1. CBC (open-source MILP)
2. GLPK
3. Gurobi (commercial)
4. IPOPT (NLP)
5. SciPy (fallback)

---

🧪 Testing

python -m pytest tests/ -v

---

📖 References

• Gary, Handwerk, Kaiser — Petroleum Refining, 5th Ed.
• Riazi — Petroleum Fraction Characterization
• Li et al. (2016) — Nonlinear blending optimization
• GHG Protocol — Scope 3 Standard

---

🤝 Contributing

See CONTRIBUTING.md for guidelines.

---

📜 License

MIT License — see LICENSE.

---

_{Built with ❤️ for the refinery optimization research community}