The development of low resistivity oil layers in the research block is extensive, and the difference in resistivity between oil layers and water layers is not significant, making it difficult to identify logging fluids. This paper uses Smote Oversampling algorithm to Oversampling a few samples of oil and water in the same layer, oil layers and so on to balance the data set. And uses gated recurrent unit (GRU) network model for fluid identification of low resistivity oil layers. Through correlation analysis, eight logging curve data such as GR, RD and DEN are determined as input training models, which are applied in actual data. GRU is compared with traditional RNN and other three machine learning algorithms. The results show that the fluid identification performance of sequential data models is better than that of traditional machine learning models and the coincidence rate of the fluid recognition model based on Smote GRU reaches 89.5% (compared to 81.1% of traditional RNN), achieving good application results. Through comparative experiments, it was also confirmed that the Smote algorithm improved the classifier"s recognition rate for minority class samples. The method proposed in this article can provide reference for fluid identification of low resistivity oil layers with imbalanced samples.