Fracture characterization via AI-assisted analysis of temperature logs (2022-2024)
Investigator: Xiaoyu Yang
The LSTM fully convolutional network architecture for fracture characterization
Fracture characterization is key to the successful evaluation and development of subsurface energy resources. Fracture information is strongly related to downhole temperature distribution due to the fully coupled flow and heat transfer behavior in fractured reservoir systems. Benefiting from the high spatiotemporal resolution of the DTS measurement, downhole temperature has become one of the most commonly measured signals, and can be a good candidate for fracture distribution inversion. Unlike commonly used knowledge-based inverse methods which have strict requirements on the pre-established forward mapping and can only process cases one by one, machine learning algorithms learn the mapping relation from the data, and can automatically process a large number of cases at the same time, making the solution of inversion problems more efficient and effective. The main goal of this work is to:
1. Develop a general deep learning approach to process DTS profiles at various stages, e.g., injection, warmback, and static stages for accurate fracture identification;
2. Explore the complex mapping relation between fracture information and spatiotemporal temperature series;
3. Develop a robust deep learning approach to estimate fracture flowrate distribution by DTS profiles for the optimal design and control of the reservoir’s exploitation.