Scalable Risk-Averse Well-Placement Optimization Using QKP and Randomized SVD

This repository contains the implementation for the paper "Scalable Risk-Averse Well-Placement Optimization Using Quadratic Knapsack Problem and Randomized Singular-Value Decomposition" (SPE-J-0325-0021).

Overview

A scalable framework for risk-averse well placement optimization that:

• Formulates well placement as a Quadratic Knapsack Problem (QKP) with mean-variance objectives
• Uses Gaussian Process (GP) regression to model spatial correlations in well productivity
• Employs Randomized SVD for efficient low-rank approximation of covariance matrices
• Achieves significant computational speedup while maintaining solution quality

Key Features

• Scalable optimization: Handles large candidate well sets (1000+ locations)
• Risk-aware decisions: Balances expected NPV against portfolio variance
• Efficient computation: RSVD approximation provides 10-100x speedup
• Flexible framework: Supports multiple GP kernels and solver configurations

Installation

1. Clone the repository:

   git clone https://github.com/rfarell/qkp-well-placement.git
   cd qkp-well-placement

2. Install dependencies:

   pip install -r requirements.txt

3. Ensure you have Gurobi installed and licensed (required for the QKP solver).

Quick Start

# Generate scenarios and run optimization
./run.sh

# Run with custom experiment name
./run.sh -n my_experiment

# Skip scenario generation (use existing)
./run.sh --skip-scenarios

Configuration

Key parameters in config/config.yaml:

• N_PUDS: Candidate well locations (e.g., [100, 500, 1000])
• LAMBDA_VALUES: Risk aversion parameters [0-1]
• RSVD_R: Low-rank approximation ranks
• VARIOGRAMS: GP kernel types (rbf, exponential, matern)

Results

The framework produces:

• Optimal well placement solutions for different risk preferences
• Performance comparisons between exact and RSVD-approximated solutions
• Computational speedup analysis
• Solution quality metrics (objective values, variance reduction)

Citation

If you use this code in your research, please cite:

@article{farell2025scalable,
  title={Scalable Risk-Averse Well-Placement Optimization Using Quadratic Knapsack Problem and Randomized Singular-Value Decomposition},
  author={Farell, R. and Bickel, J. E. and Bajaj, C.},
  journal={SPE Journal},
  year={2025},
  note={Paper Number: SPE-J-0325-0021 (Submitted)}
}

Authors

• Ryan Farell - University of Texas at Austin ([email protected])
• J. Eric Bickel - University of Texas at Austin
• Chandrajit Bajaj - University of Texas at Austin

License

This project is licensed under the MIT License.