Regression Committee Machine and Petrophysical Model Jointly Driven Parameters Prediction From Wireline Logs in Tight Sandstone Reservoirs

Abstract

Tight sandstone reservoirs are characterized by low porosity, extra-low permeability, and diverse mineral compositions for which previously established log interpretation methods are not suitable. Therefore, it is necessary to explore new log interpretation methods. The committee machine is a recently developed composite expert network. The regression committee machine (RCM) is constructed by combining a backpropagation neural network (BPNN), extreme learning machine (ELM), wavelet neural network (WNN), and using two different weight calculators for decision-making. Petrophysical models for tight sandstone reservoirs are integrated together with the RCM. The RCM and petrophysical model hybrid intelligent system for log interpretation is developed. The data process flow and specific implementation method of the log interpretation are designed. The tenfold cross-validation method is used to train and optimize the parameters of the RCM in the hybrid intelligent system. The hybrid intelligent system is applied to the Chang 8 tight sandstone of Yanchang Formation in the Ordos Basin, China. The case studies show that the predicted porosity, permeability, and water saturation by the RCM with petrophysical models are all consistent with the core measurement. The data and models jointly driven intelligent system is more accurate than petrophysical models and individual expert networks. This study effectively improves the formation evaluation for tight sandstone reservoirs.