Fluid Identification Method for Low Resistivity Reservoir Logging Based on Gated Recurrent Unit Network
The research block is characterized by extensive development of low-resistance oil formations,with little difference in resistivity between the oil and water formations,making logging fluid identification more difficult.In order to effectively identify the oil layer of low resistivity,synthetic minority oversampling technique(Smote)was adopted to oversample a few types of samples such as oil-water homogeneous layer,oil layer,etc.to equalize the dataset,and the gated recurrent unit(GRU)network model was utilized for fluid identification in low resistance reservoirs.Eight logging curve data such as natural gamma-ray(GR),deep resistivity(RD),density(DEN)were identified as input training models through correlation analysis and applied to the medium actual data,and GRU was compared with traditional recurrent neural network(RNN)and other three machine learning algorithms.The results show that the sequential data model is better than the traditional machine learning model for fluid recognition,and interpretation agreement rate of the Smote-GRU-based fluid recognition model reaches 89.5％,compared to 81.1％of traditional RNN,which achieves better application results.It is also confirmed through controlled experiments that the Smote algorithm improves the recognition rate of the classifier for minority class samples.The proposed method can provide a reference for fluid identification in oil layers of low resistivity with sample imbalance.
low resistivity oil layersidentification of fluidunbalanced samplesgated recurrent unit(GRU)
万方数据
维普
