Predictive Analytics for Elastic Log Reconstruction: A Critical Driver for CCS Studies Available to Purchase

CCS (Carbon Capture and Storage) feasibility studies need a base of robust inputs to proper characterize the geomechanical settings of the candidate area. The aim is to optimize the CCS activities and to avoid any possible damage due to partial knowledge of the zone destined to the injection. In this respect, elastic logs, in particular compressional and shear sonic and density, are the mainstay for properly driving several studies focused on this storage activity. The proposed paper deals with a real case study, where CCS application has been planned for a brown field. The background is challenging because, in an old mature field, the main criticality is represented by the available log datasets, often incomplete and ineffective to provide a robust characterization. In particular, the lack of some fundamental elastic information makes the subsequent studies very challenging to be representative of the real geomechanical background. In this picture, with a very limited acquired log dataset, Machine Learning (ML) techniques allow to address the issue by means of the elastic log triplet reconstruction. This approach consists in choosing the most representative wells with complete datasets, which represent the core of the learning phase. A mandatory step is the Quality Check (QC) of the available logs, in order to further assess the reliability of derived reconstructed logs. In fact, a detailed QC on the complete log dataset is performed. Then, the implemented ML methodology takes advantage of a dedicated algorithm that learns through experience how the available input logs (compressional sonic and deep resistivity) are related to the output logs (shear sonic and density), objective of the reconstruction. In detail, a subset has been used as training set for the learning phase. Then, the algorithm is used to predict the elastic responses in a given validation set. In the end, a test dataset provides an unbiased evaluation of the complete approach. Finally, the uncertainty of the reconstructed logs has been also assessed through a data-driven step. The successful outcomes coming from the validation and test analyses allow to propagate the model prediction to all the wells of interest in the field. This final propagation provides a complete elastic log dataset that has been used for the geomechanical studies at well and at reservoir scale, fundamental for the purposes of the CCS project.