In the introduction, Mike Economides makes a bold claim for data mining in oil and gas. For Economides, mining of the huge datasets generated by exploration and production is a new way of ‘matching real with predicted data’—along with traditional numerical analysis and numerical simulation. Oil and Gas Data Mining* deals mainly with three techniques—neural networks (NN) , self organizing maps (SOM) and genetic algorithms. The presentation is at a suitable level for non specialists to get a grasp of the basic concepts and is written in a clear style with a straightforward presentation of the math.
The treatment of mainstream data mining—with SQL and OLAP is a bit skimpy—a shame in view of the plethora of tools which are available for slicing and dicing hypercubes of data. But the book scores with excellent case studies. The first covers the use of multiple linear regression, SOM and NN to establish a model of permeability over a large middle east oilfield. The second compares techniques for selecting wells to stimulate on the Wattenburg field in Colorado. Five different artificial intelligence methods were compared with a type curve analysis. Intriguingly, the study shows that different methodologies threw up different stimulation candidates.
A third way?
The book is a great introduction to a variety of new techniques, but is data mining really a ‘third way’ for analysis? This reviewer sees these techniques as contributing to a bewildering armory of methods from analysis through simulation to statistics. Most approaches will likely combine elements of some or all of these into the estimation process.
* Data Mining Applications in the Petroleum Industry. Zangl and Hannerer, Round Oak 2003. ISBN 0-9677248-1-3.
This article originally appeared in Oil IT Journal 2004 Issue # 2.
For more information or to comment on this topic email here.