Seismic inversion based on principal component analysis and probabilistic neural network for prediction of porosity from post-stack seismic data

Abstract

This research delves into the utilization of Principal Component Analysis (PCA) and Probabilistic Neural Network (PNN) techniques for predicting porosity values based on seismic inversion data, which provides crucial insights into subsurface properties essential for reservoir characterization. The study integrates Principal Component Analysis (PCA) with Probabilistic Neural Networks (PNN), to streamline the interpretation of seismic data by condensing numerous attributes into a smaller set with a probabilistic classification approach. This integration provides a novel methodology for processing and interpreting complex seismic data. Additionally, model-based inversion is conducted to obtain extra attributes (inverted impedance 7000–18000 m/s*g/cc) for use in training the PNN model. Subsequently, the PNN model is utilized to make accurate porosity predictions by leveraging the reduced-dimensional seismic features. The study aims to compare PNN performance with single attributes against PNN with Principal Component(PC), highlighting their differences and similarities for the data from the Blackfoot field in Canada.Initially, various single attributes are generated and then subjected to PCA techniques to convert them into three principal components. Both PNN models, one with single attributes and the other with PC, are trained and validated to reveal detailed subsurface information. The inverted results show a strong correlation with well-log parameters, with an average correlation coefficient of 0.84 for PNN using single attributes and 0.78 for PNN using PC.It was observed that PNN with single attributes performed better on the training data set but Performs below expectations on validation dataset compared to PNN with PC.Furthermore, these algorithms are applied to whole seismic data to predict porosity within the inter-well region. The inverted volumes depict consistent porosity ranging from 5 to 11% across the region for PNN with PC. Furthermore, the interpretation of the inverted results for PNN with PC highlights anomalous zones characterized by low impedance and high porosity. These zones closely match well-log data and are identified as likely sand channels. The combination of PCA and PNN represents an advancement in machine learning techniques applied to geoscience. This research contributes to the growing field of machine learning in geoscience by exploring new methodologies and their applications. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024.