Abstract

This study introduces AttriFusionAI, an AIdriven approach, designed to automate the selection of optimal seismic attribute combinations for performing the unsupervised model in identifying and classifying geological facies, based on seismic data calibrated with well. Recognizing the variability in seismic attribute effectiveness across different geological settings, the approach offers a robust, adaptive solution to enhance facies prediction accuracy in diverse subsurface environments. The method is demonstrated using a 3D seismic volume integrated with well data from a gas-bearing deep offshore reservoir in the Brazil Basin. AttriFusionAI processes 15 seismic attributes derived from five categories- Instantaneous, Elastic, Amplitude, Frequency, and Hybrid (Amplitude & Frequency). The methodology leverages a multi-step analytical framework incorporating Statistical correlation, Principal Component Analysis (PCA), and SHAP (SHapley Additive Explanations) to evaluate the contribution and relevance of each attribute. This enables the generation of composite, fieldspecific attributes that best represent the geological variability of the studied reservoir. Unlike traditional approaches that rely on static or generalized attribute selection, AttriFusionAI dynamically adapts to the structural and depositional complexity of each reservoir, thereby minimizing interpretational bias. The algorithm supports asynchronous analysis across multiple formations and fields, ensuring scalability and computational efficiency. The selected and fused attributes form the basis for unsupervised facies classification, which can be directly integrated into static reservoir modelling workflows.