Earth, Atmosphere & Environment Sciences
Title : | Delineation of subsurface geological features, prediction of missing logs prediction and conditioning of vintage sonic logs by machine learning perspectives |
Area of research : | Earth, Atmosphere & Environment Sciences |
Focus area : | Geology |
Principal Investigator : | Dr. Bappa Mukherjee, Wadia Institute of Himalayan Geology, Dehradun, Uttarakhand |
Timeline Start Year : | 2024 |
Timeline End Year : | 2027 |
Contact info : | bappa.ism@gmail.com |
Details
Executive Summary : | This project aims to develop new approaches based on machine learning algorithms and apply to address challenges like delimitation of subsurface geologic features from 3D surface seismic data, prediction of missing logs at some depth intervals due to certain constraints in data acquisition, and conditioning of vintage log data in bringing out new information that could not be delineated earlier. This type of research is very useful to the E&P industry for sustainable development and for de-risking hydrocarbon exploration. Machine learning tools such as deep learning Convolutional Neural Networks (CNN), Long short Term Memory (LsTM) Networks, as well as shallow learning Artificial Neural Networks (ANN) and K-Nearest Neighbour (KNN), would be utilised to the industry-standard data analysis and interpretation against the tedious and subjective traditional way of data analysis and interpretation. The CNN U-Net will be utilised to predict the subsurface fault network and facies classification from seismic data by avoiding the attribute analysis-based seismic interpretation. In geologically complex areas, attribute analysis may pose several artifacts, which in turn lead contamination to interpret the outcome. The LsTM (special kind of recurrent neural network) and bidirectional-LsTM (Bi-LsTM) will be utilised to tackle the challenges of E&P industry such as the prediction of missing logs and conditioning of vintage sonic logs. Additionally, by implementing the Bi-LsTM along with the KNN and ANN techniques, the data-driven patterns between the wireline logs and core-derived lithologies will be extracted for the prediction of lithologies for the un-cored sections at the wells. The KNN and ANN will be used to predict the lithologies under a particular stratigraphic unit. Whereas, the Bi-LsTM will be implemented to predict the lithologies associated with the multiple stratigraphic units, simultaneously. This project is thus aims for an integrated study of seismic, wireline logs, and core information in order to delineate the subsurface geological structures (faults networks and facies), derive log responses at missing data interval, condition the vintage sonic logs with a view to make prediction of lithologies at the uncored depth interval using AI or ML-based approaches. The feasibility of the proposed research will be tested on the data procured from hydrocarbon industry in the Assam-Arakan basin, which is tectonically active, geologically complex and prospective for higher hydrocarbons (Category-I basin). |
Total Budget (INR): | 28,01,720 |
Organizations involved