A new well log interpretation model based on Emergent Self-organizing Maps

When the training samples of well log data for Kohonen Self-Organizing Maps(KSOM) are large and high dimensional, the adjacent clusters may be overlap in a common region. In the paper, a new model of clustering analysis and recognition for well log data is proposed with Ultsch Emergent Self-organizing Maps(ESOM) of neural network. This method can overcome the weakness of KSOM and optimize the result of clustering by using component map, U-Matrix and P-Matrix to visually compare and analysis the clusters on boundless toroid topology grids. This model is trained by the data clustering and visualization for key wells´ data in oilfield block. The results show that this new model has good application prospects for well log interpretation using the trained pattern classifier.