1. School of Geosciences, Northeast Petroleum University, Daqing 163318, Heilongjiang, China; 

 In order to more accurately identify the fluid properties of Tongbomiao Formation reservoir of Wuerxun depression in Hailar Basin, the authors consider that the saturated pure water resistivity (R0 ) is influ? enced by lithology, physical properties and pore structure through experimental analysis of parent rock composition, pore structure and core experiments. In this paper, the value of R0 is predicted by neural network to eliminate the influence of lithology and pore structure on resistivity value, and comparison of receiver operating characteristic curves (ROC) and self?validation accuracy to optimize the combination of logging curves with high relevance for reservoir fluid properties identification: interval transit time (DT24), natural gamma (GR), density ( ZDEN), neutron (CNC), deep investigate double lateral resistivity (RLLD), and R0 . The results of three machine learning methods, namely, decision tree, support vector machine and neural network, were combined with the test well sec? tions to derive a machine learning architecture method for fluid properties identification in Tongbomiao Formation reservoir of Wuerxun depression in Hailar Basin, the BP neural network with three hidden layers and activation function of Relu has an accuracy of 92. 5% for identifying fluid properties.