Time Series Pattern Discovery Techniques for Well-to-Well Log Correlation

Abstract

Well log data is one of the most abundant data sources for subsurface characterization since almost all drilled wells are logged. Subsurface characterization requires correlated well log information. Processes to help perform well-to-well log correlation have been developed in this study based on the two approaches: pattern matching and pattern discovery. Pattern matching focuses on finding a known pattern of interest in uncorrelated wells. The use of Euclidean, Hamming, and Levenshtein distances in similarity measurement, Piecewise Aggregate Approximation (PAA), and Symbolic Aggregate approXimation (SAX) data representation helps create reduced-resolution while maintains the main characteristics of the signals. Multi-resolution analysis provides the most probable match from different data resolutions obtained by different discretization levels. The results show that multi-resolution analysis can identify the most probable match from the most frequent data window promoted. The results also show that Levenshtein distance is far superior to Hamming and Euclidean distances in finding the best match even when local variations are present. Pattern discovery focuses on finding repeating patterns without any prior knowledge of patterns that might exist in the well logs. MK (Motif Kymatology) is used as the base algorithm in this study. For the data sets tried in this study, the proposed algorithm can successfully identify repeating patterns. A modified step is incorporated in order to help increase stability of the final correlated well sections. In summary, pattern matching and pattern discovery can be formed as an integrated workflow to support well log correlation task and subsurface characterization.