Page 173 - Intelligent Digital Oil And Gas Fields

132 Intelligent Digital Oil and Gas Fields

Objective

Correct Correct Correct
pressures saturations flowrates

Modify fluid Modify well
Modify Modify properties properties
permeability porosity
(PVT) (IP,skin)
Increase Decrease Increase Decrease Increase

Primary corrective layer Secondary corrective layer Direction layer
Fig. 4.12 Examples of a hierarchical learning exercise composed of intelligent agents to
perform autonomous history matching. (Modified from Zangl, G., Al-Kinani, A.,
Stundner, M., 2011. Holistic Workflow for Autonomous History Matching using Intelligent
Agents: A Conceptual Approach. SPE-143842-MS, https://doi.org/10.2118/143842-MS.)

Kohonen self-organizing map (SOM) (Fig. 4.13); it is referred to as a map
because it assumes a topological structure among its cluster units and effec-

tively maps cluster weights w ij1 , w ij2 …, w ijn to input data vector x 1 , …, x n
and generates output data vector y 1 , …, y n .
Roy et al. (2013) are also using SOMs for supervised and unsupervised
multi-attribute facies analysis in seismic stratigraphy, and Zangl and

Input vector
X=(x , x , ... , x )
n
2
1
w =(w , w , ... , w )
ijn
ij
ij1
ij2
Weights vector

Output vector
Y=(y , y , ... , y )
n
2
1
Fig. 4.13 A rendering of Kohonen self-organizing model architecture. (Modified from
Magomedov, B., 2006. Self-Organizing Feature Maps (Kohonen Maps), https://www.
codeproject.com/Articles/16273/Self-Organizing-Feature-Maps-Kohonen-maps.)