A Deep Learning Based Geosteering Method Assembled with "Wide-angle Eye"

Abstract

The intelligent guided drilling system adopts the precise guided drilling geological system and a new rotary steering drilling tool to achieve deep drilling intelligent cruise. It can increase the amount of oil and gas exploration and ensure safety in production. However, the geosteering problem in deep wells and ultra-deep wells is still an outstanding issue due to the hostile environment for signal transmission. In this research, an autonomous geosteering method based on deep learning model is proposed, which is able to make the strategic decision of the drill bit direction in downhole operating mode. According to the characteristics of the Logging While Drilling (LWD) data, the "Wide-angle Eye" mechanism is embedded to feel the future change of stratum ahead and give preview information to the drill bit. Consequencely, the Drilling Decision Model is designed to be a Convolutional Neural Network (ConvNet). The performance of the proposed model was validated in simulation, and the experimental results indicate that the proposed method has high accuracy and robustness, appearing an enhanced capacity to predict stratigraphic changes.