Lithology recognition and porosity prediction from well logs based on Convolutional Neural Networks and sliding window

Abstract

Predicting the lithology and porosity of borehole rocks based on wireline logging data holds significant importance. The sampling interval of the logs is relatively small, so the log values within a specific range above and below the target depth contain effective information about the borehole rock at the target depth. This paper proposes a method that combines a deep sliding window with a convolutional neural network. In this approach, multiple logging curves within the sliding window serve as inputs, and the convolutional neural network extracts valuable information from these logging curves. Subsequently, the borehole lithology and porosity at the window center are predicted based on the extracted information. As the window slides vertically, it enables the rapid prediction of lithology and porosity for the entire wellbore. Based on the practical application in an oil field in the east of China, it was determined that the optimal length of the sliding window is 1.125 m. The accuracy rate of the proposed convolutional network model for lithology prediction can exceed 94.4 %, and the accuracy rate for porosity prediction is 94.9 %. The prediction speed is notably fast, making it applicable with precision to lithology or porosity predictions in numerous oil fields and new wells.