“Sharing Private” Multi-task Learning for Petrophysical Parameters Prediction with Logs

• DOI: 10.3997/2214-4609.202112714
• Publication Date: 2022-08-25T09:53:03Z
• Authors: R. Shao; L. Xiao; G. Liao
• Keywords: Petrophysics; Information gain

Summary The existing study single predict neural network were used, that is, for a one petrophysical parameter can be predicted, such as porosity (POR) or water saturation (SW) with set of logging data. When the tasks are related to each other, underlying information extracted from input features model has certain commonality. In this case, multi-task extract higher quality help multi-dimensional output information, so obtain better performance and produce an effect similar "information gain". We propose "sharing private" machine learning method prediction logs, which improve efficiency, simplify process reduce mean absolute error compared network. Using data 64 wells in Chinese oilfields, it is found that, single-task model, significantly permeability saturation.