Simulation Analysis With Association-Rule Mining Plus High-Dimensional Visualization

Fig. 1: Dimensional-stacking image from 243 reservoir cases from the SAIGUP data set. L=low, M=mid, and H=high.

Computational advances in reservoir simulation have made possible the simulation of thousands of reservoir cases in a practical time frame. This opens a new avenue to reservoir-simulation studies, enabling exhaustive exploration of subsurface uncertainty and development/depletion options. However, analyzing the results of a large number of simulation cases remains challenging. This paper presents a new method that enables the efficient analysis of massive reservoir-simulation results by discovering interesting patterns of relationships among variables in large data sets. The method uses association-rule mining together with high-dimensional visualization.


Ensemble-based approaches for reservoir modeling and simulation have been investigated for decades. The majority of the methods are designed to explore a high-dimensional space spanned by uncertainty parameters and find a set of reservoir models that reproduce historical production performance. Once an ensemble of history-matched models is obtained, the effect of subsurface uncertainty on production forecast is evaluated quantitatively by simulating flow performance on individual members of the model ensemble. Exploration of the parameter space during the model calibration is conducted with various stochastic algorithms. Major research efforts on these approaches are devoted to achieving efficient sampling from the parameter space and obtaining predictive capability to capture the range of uncertainty related to geological uncertainty remaining after data assimilation. However, little attention has been paid to qualitative analysis of the effect of geological features on simulated production performance.

One obvious reason for the under­utilization of the model ensemble for understanding reservoir sensitivity is the lack of efficient methods to visualize simulation results in such a way that interaction among multiple uncertainty parameters and simulated production response can be revealed rapidly. This paper presents a novel methodology that serves this purpose by coupling a well-known data-mining algorithm, called ­association-rule mining, with an existing high-­dimensional visualization method called dimensional stacking.

This article, written by Special Publications Editor Adam Wilson, contains highlights of paper SPE 174774, “Using Association Rule Mining and High-Dimensional Visualization To Explore the Impact of Geological Features on Dynamic Flow Behavior,” by Satomi Suzuki, SPE, and Dave Stern, SPE, ExxonMobil Upstream Research Company, and Tom Manzocchi, SPE, University College Dublin, prepared for the 2015 SPE Annual Technical Conference and Exhibition, Houston, 28–30 September. The paper has not been peer reviewed.
This article is reserved for SPE members and JPT subscribers.
If you would like to continue reading,
please Sign In, JOIN SPE or Subscribe to JPT

Simulation Analysis With Association-Rule Mining Plus High-Dimensional Visualization

26 June 2016

Volume: 68 | Issue: 7


Don't miss the latest content delivered to your email box weekly. Sign up for the JPT newsletter.