Semi-Supervised Clustering of Seismic Data for Oil Prospectivity

Traditional oil prediction is notoriously difficult due to the convoluted relationship between seismic parameters and actual oil volume. To address this, CGI developed a breakthrough semi-supervised clustering method based on a constrained Graph Laplacian, together with partners from Chevron Corporation. This proprietary technology allows for the processing of massive 3D seismic datasets—containing millions of data points—with extremely limited labeled borehole information. By treating each point in a 3D volume as a data point, our algorithm creates a closed-form solution that significantly reduces computational burden while increasing predictive accuracy.

CGI applied this method to the SEAM Life of Field synthetic dataset and an offshore West African Field (WAF). At the SEAM site, our technology successfully mapped central oil deposits across complex fault blocks with a low average prediction error. In the WAF project, CGI’s approach identified heterogeneous oil pockets and intricate stratigraphic channeling systems that are often missed by traditional inversion. These results validated new potential drilling locations, proving that CGI’s machine learning solutions can turn sparse well data into reliable, field-wide prospectivity maps.

‍