Porosity prediction from pre-stack seismic data via committee machine with optimized parameters

Abstract

Prediction of porosity from the seismic data via geophysical methods when limited number of wells are available is a challenging task that has high uncertainties. This study aims to construct a hybrid data-driven predictive model to establish a quantitative correlation between seismic pre-stack (SPS) data and the porosity. First, three intelligent models that are optimized by bat-inspired algorithm (BA): optimized neural network (ONN), optimized fuzzy inference system (OFIS), and optimized support vector regression (OSVR) are constructed for relating porosity to the SPS data. Then, to benefit from all individual optimized models, a final hybrid model was built via committee machine (CM) where single models are combined with a proper weight to predict porosity in the reservoir space. This approach is examined on the SPS data from an oil field in the Persian Gulf with a single exploratory well where input parameters (Vp, Vs, and ρ) to the AI models are derived from a two-parameter inversion method. We found that the coefficient of determination, root mean square error, average absolute relative error, and symmetric mean absolute percentage error for the CM are 0.923615, 0.015793, 0.132280, and 0.061310, respectively. Moreover, based on four statistical indexes that are calculated for each model, CM outperformed its individual elements followed by the OSRV. A comprehensive analysis of the results confirms that CM with the OM elements is a superior approach for computing porosity from the SPS in the well and then throughout the entire reservoir volume. This strategy can aid petroleum engineers to have a better forecast of porosity population in the reservoir static model immediately following the data that is obtained from the first exploratory well. Ultimately, successful implementation of this approach will promptly delineate sweet spots that can replace uncertain and complicated conventional geophysical methods.