Application and comparison of RBF and BP neural networks for lithology identification of Permian volcanic rocks in the Shunbei area of the Tarim Basin in China

Due to the complexity of the sedimentary environment and the heterogeneities and similarities between logging data, lithology identification is difficult. Taking the Permian in Shunbei area of Tarim Basin as an example, based on a complete understanding of the target reservoir characteristics, used the crossplot method, BP (backpropagation) neural network method and RBF (radial basis function) neural network method to identify three types of volcanic rocks: tuff, andesite, and basalt. At the same time, the crossplot method was used to select four logging curves that are sensitive to lithofacies changes as important indicators for identifying volcanic rocks, such as the natural gamma ray (GR), compensated density (DEN), compensated neutron (CNL) and spontaneous potential (SP) logs. Then, the sensitive curves were preprocessed by standardization, and suitable learning samples were selected. Two types of neural network prediction models were established, and the mapping relationship between the lithology and logging curves was used to identify the lithofacies of the key wells in the study area. Finally, by comparing the recognition results of the three methods, it was found that the RBF network not only achieved higher accuracy in the prediction results but also had fewer learning iterations than the BP network could more accurately identify volcanic rocks, and has certain popularization and application values, while the crossplot method had the worst recognition results.