Unsupervised machine learning-based multi-attributes fusion dim spot subtle sandstone reservoirs identification utilizing isolation forest

Abstract

Subtle sandstone reservoirs are difficult to identify due to their weak seismic responses. Here, we propose to identify subtle sandstone reservoirs by an unsupervised machine learning-based multi-attribute fusion scheme using prestack seismic data. The proposed scheme carries out seismic attenuation gradient analysis and prestack simultaneous inversion to obtain the attributes that are sensitive to subtle channel sands, and uses them as the selected multiple attributes, and further employs a state-of-the-art unsupervised machine learning algorithm, called isolation forest, for the multi-attribute anomaly detection and analysis to identify subtle sandstone reservoir. To the best of our knowledge, this is the first time to introduce the isolation forest unsupervised anomaly detection algorithm in the reservoir identification. Prestack simultaneous inversion can use multi-angle and multi-scale information as constraints, and the attenuation gradient reflects the body response of the reservoir. For the field seismic data from a subtle channel sandstone reservoir in the western Sichuan basin, China, we found that the proposed scheme has good application effect in identifying subtle reservoirs. The application example demonstrates that the identified results are highly consistent with the actual development results, illustrating the feasibility and effectiveness of this scheme on the characterization for dim spot subtle sandstone reservoirs. This study is hoped to be useful as an aid for reservoir identification for dim spot subtle sandstone reservoirs, as well as to provide a new technical idea and method for reservoir characterization.