Accelerating simulation for the multiple-point statistics algorithm using vector quantization.

Area of Science:

• Geostatistics
• Computational Geoscience
• Machine Learning Applications

Background:

• Multiple-point statistics (MPS) is crucial for simulating categorical variables using training images.
• Current MPS methods face challenges with large databases and long retrieval times for complex patterns.
• State-of-the-art MPS requires improvements in simulation speed and quality.

Purpose of the Study:

• To develop an accelerated MPS algorithm using vector quantization (VQ) for faster and higher-quality simulations.
• To address limitations in database size and retrieval time associated with complex patterns in MPS.
• To enhance the simulation of geological variables, including nonstationary and multifacies scenarios.

Main Methods:

• Proposed Vector Quantization for Multiple-point Statistics (VQ-MPS) algorithm.
• Developed a variable representation for applying VQ to categorical variables.
• Utilized tree-structured VQ for database compression and stationary simulations.
• Implemented transformed templates and classified VQ to handle nonstationarity.

Main Results:

• VQ-MPS demonstrated significantly improved performance in computational time, pattern reproduction, and spatial uncertainty compared to existing MPS programs.
• The method successfully validated on a 2D stationary channelized reservoir image.
• Further tests confirmed VQ-MPS capability in multifacies, nonstationary, and 3D simulations using 2D training images.

Conclusions:

• VQ-MPS offers a substantial advancement over traditional MPS methods for geological simulations.
• The algorithm effectively handles complex patterns, nonstationarity, and multi-dimensional data.
• VQ-MPS provides a robust and efficient solution for geostatistical simulation challenges.