Log Facies Identification via Dynamic Classifier Selection

The automatic identification of log facies using machine learning appears to be a promising solution to the challenges encountered by traditional methods because of its objectivity, efficiency, and ability to manage high-dimensional data. However, selecting an appropriate model for the available data without any a priori information is a complicated task. Even with a well-trained model, the prediction process must handle samples from multiple classes, and this indiscriminate handling of samples complicates the identification problem. In this study, we proposed a novel log facies identification method based on dynamic classifier selection to achieve more intelligent identification. We first developed a pool of classifiers with multiple learning algorithms and then chose only one trained model from this pool with a dynamic selection scheme for different test patterns. We applied this method to a set of open-source data from the Panoma field in southwest Kansas and found that the dynamic selection method achieves higher log facies classification compliance than a single model that has not undergone dynamic selection. The results indicate that dynamic selection is a promising approach for model selection in intelligent interpretation.