INTRODUCTION: FIELD APPLICATIONS OF INTELLIGENT COMPUTING TECHNIQUES
P.M. Wong1 and M. Nikravesh2
1 School of Petroleum Engineering, University of New South Wales, Sydney, NSW 2052, Australia.
2 Berkeley Initiative in Soft Computing, Computer Science Division, Department of EECS, University of California at Berkeley, CA 94720, USA.
"As complexity increases precise statements lose meaning and meaningful statements lose precision" L.A. Zadeh (1921-)
The natural complexities of petroleum reservoir systems continue to provide a challenge to geoscientists. The absence of reliable data often leads to an inadequate understanding of reservoir behaviour and consequently to poor performance predictions. Although this is an ongoing problem and one which may be difficult to resolve without additional data and/or investment, it is important to pursue the best possible solutions using whatever data is readily available. Data integration, and risk and uncertainty assessment, have become the major issues in reservoir characterization and improved oil recovery.
In past decades, classical data processing tools and physical models were adequate for the solution of relatively "simple" geological problems. However because of the uncertainties which are inherent in geological data, the challenge we now face now is not just to predict the presence of hydrocarbons, but rather to quantify the confidence of reservoir predictions. We are increasingly being faced with more and more complex problems, and reliance on current technologies based on conventional methodologies is becoming less satisfactory.
Intelligent (or "soft") computing consists of a suite of emerging technologies in data and knowledge processing. These include neural networks, fuzzy logic, evolutionary computing and advanced statistical tools. The formulation and application of intelligent techniques have increased exponentially over the past few years. Unlike conventional (or "hard") computing, intelligent techniques are tolerant of imprecision, uncertainty and partial truth. They are also tractable, robust, efficient and inexpensive.
Intelligent techniques are bound to play a key role in earth sciences in the future (Zadeh and Aminzadeh, 1995; Aminzadeh, 1999; Tamhane et al., 2000). This is mainly due to the fact that physical models cannot describe geological and physical phenomena accurately but rely on the interpretation of data. Many conventional modeling techniques rely purely on data and occasionally on indirect knowledge which can sometimes be unrelated to reservoir sedimentology and depositional characteristics. Thus:
Contouring without data is possible in computer simulation;
Contouring with data on its own is a bit better;
Contouring with data but without geological knowledge is unrealistic; and
Contouring with data conformed to geological knowledge is acceptable.
The current role of reservoir geology in numerical reservoir modeling is diminishing (Tamhane et al., 1997; Wong et al., 1999). However it is "...most dangerous to think that, having developed wonderfully sophisticated technologies, they are going to do the whole job for us" as noted by Stoneley (1997). Intelligent techniques aim to combine data (of varying qualities) and knowledge (from multiple scenarios) in an effective manner. In reservoir characterization, these techniques can be used to integrate multivariate non-linear information (data and knowledge). They can then be applied, for example, to feature extraction from seismic attributes, comparison of seismic characters, fracture identification, biostratigraphic modeling and petrophysical evaluation using well logs.
There are a good number of technical papers, special issues and books on geoscience applications of intelligent techniques (e.g. Nikravesh et al., 2001; Wong et al., 2001). Unlike many of these publications, the thematic series of papers in this issue of the JPG focus on wider oilfield applications of intelligent computing techniques. As editors, our objectives were to present the state-of-the-art to the wider petroleum-geological community and to show how these technologies can provide improved results in actual field studies. The specific problems, fields/basins and technologies addressed in this issue are listed in Table 1.
Below, we first briefly describe the intelligent techniques -- neural networks and fuzzy logic -- used by the contributors, and summarise their major findings (the techniques used by Russell et al. will not be mentioned). We then attempt to predict some future directions for work in this area.