Fluid Identification Using Conditional Variational Autoencoder and Hierarchical Time Series Classification Leveraging Logging Data

Reservoir fluid identification is a critical aspect of oil and gas geophysical exploration. Accurate fluid identification directly impacts the interpretation of subsurface geological conditions, reduces exploration risks, and provides essential guidance for formulating oil and gas development strategies. Therefore, reliable and precise fluid identification is indispensable across different stages of oil and gas exploration and production. This study proposes a hierarchical classification method based on conditional Variational Autoencoder (cVAE) and time series forest (TSF) algorithms to address reservoir fluid identification under complex geological conditions. The main contributions of this work are as follows: (i) the cVAE is used to pre-process the logging data to suppress local high-frequency disturbances and isolated anomalies that may exist in the logging curves, thereby improving the quality of the input data; and (ii) hierarchical classification strategy is utilized to perform the fluid identification task in two steps. The first step involves a top-level classification to distinguish the gas bearing layer from the non-gas layer. The second step refines this classification into subcategories, including the gas layer (GL), gas–water layer (GW), gas-bearing water layer (GBW), water layer (WL), and non-reservoir layer (DW). This can fully address the challenges of imbalanced datasets and improve the recognition accuracy of minority classes. Additionally, integrating the TSF algorithm within the hierarchical classification framework effectively captures the sequential characteristics of well logging data, improving the model’s ability to recognize complex geological patterns. A real-world application in a block of the Yinggehai Basin in the South China Sea demonstrated the superior performance of the proposed model. Experimental results show that the method achieves an accuracy of over 84% in all four wells, enabling accurate and reliable reservoir fluid classification.