Velocity Inversion for Sandstone Reservoir Based on Extreme Learning Machine Neural Network

Abstract

This study focuses on reservoir velocity inversion from seismic data using extreme learning machine in Yanqi gas field. The analysis data consist of target logs from wells which tie the 3-D seismic volume. The specific aim of this work is to build reliable non-linear network models for velocity inversion and then predict velocity logs from seismic data for the whole survey. We first calculate five types of seismic attributes and extract them at well locations. Then pair them with the velocity logs as the training data for the ELM network. Appropriate network parameters are determined and non-linear ELM model predictor is set up for the research. We then carry out the velocity inversion for the whole survey. The results show that the ELM network has good performance, and two favorable areas are suggested for further research.