03FORCE Larsen Machined learned well lithology prediction from a disparate well log dataset and impe скачать в хорошем качестве

03FORCE Larsen Machined learned well lithology prediction from a disparate well log dataset and impe

Presentation: http://www.force.org/en/Seminars/Arch... Title: Machined learned well lithology prediction from a disparate well log dataset and imperfect training data This paper explores various machine learning technologies to automate lithology cruve prediction from well logs. The database of 220 wells ranging from 2 to 40 years in age and a contractor aquired lithology data training set that has been generated during many years illustrate the typical pitfalls of using subsurface data for machine learning . In most cases the subsurface datasets are not very clean and consistent which make the application off the shelf machine learning workflows less efficient. Classic data analysis and data cleanup needs to be performed in order to improve the machine prediction performance. After rigerous data cleaning and qc the machine learned algorithm for lithology prediction outperforms the quality of the initial training dataset and can be used qc the contractor database as well as classifying wells where no lithology data has been available so far. Eirik Larsen Earth Analytics