Abstract
Many facets of geology involve processing large sets of data, and recognizing patterns. Still, geology largely remains a qualitative study where professionals spend endless man-hours pouring through reports to build models for predictive analysis. Machine Learning is a subfield of analytics and computer science which is dedicated to using pattern recognition to build models. In traditional data analytics an algorithm is applied to data which then yields a result, but Machine Learning compares large sets of data with results from analysis to create an algorithm and models which can then be used for predictive analysis.
In this paper we will examine large sets of curve data from the Norwegian Continental Shelf, and compare it with the classifications for the stratigraphy and geological structures from the Norwegian Petroleum Directorate to build models using different Machine Learning strategies.
During the analysis the data and result sets are divided into three parts, where half is to build the model, and fourth is used as preliminary test set to improve the algorithm and the final fourth to test the final algorithm. The efficacy of the algorithm is determined by how close to the results of the final test set the model can classify the layers.
The results of this analysis can not only be used to classify stratigraphy for new wellbores with data, but also identify anomalies and errors in classification in the database. In addition, the algorithm can show which sensor data correlate and contribute to the classification of sedimentary layers.
