Evolving Finite State Transducers to Interpret Deepwater Reservoir Depositional Environments

Abstract

Predicting oil recovery efficiency of deepwater reservoirs is a challenging task. One approach to characterize and predict the producibility of a reservoir is by analyzing its depositional information. In a deposition-based stratigraphic interpretation framework, one critical step is the identification and labeling of the stratigraphic components in the reservoir according to their depositional information. This interpretation process is labor intensive and can produce different results depending on the stratigrapher who performs the analysis. To relieve stratigrapher's workload and to produce more consistent results, this research developed a methodology to automate this process using various computational intelligent techniques. Using a well log data set, we demonstrated that the developed methodology and the designed workflow can produce finite state transducer models that interpret deepwater reservoir depositional environments adequately.

Computational intelligence for deepwater reservoir depositional environments interpretation

“…This model accurately describes the composition of channel-axis deposition: massive sandstone without shale (see Section 2). The other 4 rules in the best team are given in Figures 13,14,15 and 16. Note that the symbols in the classification rules are shorthand for the symbol thickness.…”

“…This is a huge search space and stochastic search methods such as evolutionary algorithms are good candidates to locate a decent solution within a reasonable time frame. We therefore implemented a GA in Java to train the two FST tables [14].…”