This thesis describes how I have researched and developed new methods for the prediction of rock physical properties using genetic algorithms and fuzzy logic (GAFL).  These techniques are improvements on conventional methods providing two original but dissimilar tools for formation evaluation and reservoir characterisation. The premise behind the use of fuzzy logic in this context is that a reservoir can be broken down into several lithotypes, each having characteristic statistical distributions for electrical log values.  Fuzzy logic attempt to uncover the relationships between these distributions.  Genetic algorithms use a feedback technique that assumes a continuous functional relationship between the electrical log values and rock properties, generating and testing equations that fit predicted and observed responses.  Complex non-linear equations are “evolved” until the best fit is obtained.  Genetic algorithms provide the functional form of the equation as well as the constant parameters of the relationship. I have modified conventional GAFL techniques so that they can be more precisely calibrated and applied to geoscience problems more successfully.  This research analysed the characteristics of large data sets from several North Sea and Middle Eastern fields, and led to the design of software that automatically calibrates GAFL in a way that is less sensitive to noise and data outliers.  I describe the applications of these new techniques to permeability, litho-facies, porosity and shear velocity prediction;  the repair of poor electrical logs and the modelling of shaly sand equations. Permeability governs the movement of fluids through reservoir rocks and is therefore a critical input into reservoir models.  Permeability estimation is extremely challenging, as it is difficult to measure directly using current sub-surface logging technology.  GAFL was applied to predict permeability in the Northern North Sea oil fields.  The newly developed software provides an important and visual indication of the uncertainty associated with the predicted permeabilities.

No description available.