Capability of self-organizing map neural network in geophysical log data classification: Case study from the CCSD-MH

Well log interpretation is one of the prime sources of information for deep lithology in drilling research. Because of the complex geological features of the crystalline metamorphic rocks, more complex nonlinear functional behaviors exist for well log interpretation purposes. Hence, establishing a prediction technology that can accurately interpret/classify well log data in terms of lithology is of major significance. This study, for the first time, explores the application of self-organizing map neural network (SOM) in the classification of metamorphic rocks from Chinese Continental Scientific Drilling Main Hole (CCSD-MH) log data. For this purpose, a total of 33,326 data points derived from resistivity, P-wave velocity, bulk density, photoelectric absorption capture cross section, gamma ray, potassium content and neutron logs were used as an input pattern to a SOM to classify lithology in five categories: orthogneiss, paragneiss, eclogite, amphibolite and ultramafic rocks. Comparison of SOM results to those of feed-forward neural network (FFNN) was also carried out. The cross-validation method was used to investigate the robustness of the two neural networks in terms of classification accuracy in the context of lithology clustering tasks by sampling rotation. Statistical tests such as student paired samples t-test was carried out to guide in classification decision of the CCSD-MH data. The results of this study have proven that SOM appears to be comparable to FFNN in classifying lithology using geophysical log data from crystalline rocks. This proposed SOM approach can serve as practical alternative technology to be used in drilling research.