Original Formation Resistivity Reconstruction and Water-Flooded Zone Identification Using a CNN-BiLSTM-Attention Network

Reliable identification and assessment of water-flooded intervals are crucial for enhancing oil recovery during the middle and late stages of field development. Original formation resistivity is an important parameter characterizing the initial electrical properties of a reservoir and provides a critical basis for water-flooded zone identification and remaining-oil evaluation. To overcome the limited accuracy and robustness of conventional inversion approaches in complex reservoirs, this study develops a CNN-BiLSTM-Attention framework for reconstructing original formation resistivity. CNN captures local variations in the logging responses and BiLSTM models’ bidirectional dependencies along the depth sequence, and the attention module highlights informative representations, thereby strengthening nonlinear characterization and model robustness. The Yuejin II Oilfield in the western Qaidam Basin was selected as the study area, and conventional logging data were used for model development and independent assessment. The proposed framework achieved better reconstruction performance than the comparison models. Based on the reconstructed resistivity, a resistivity attenuation ratio M was introduced and integrated with conventional logging parameters to distinguish different water-flooding levels. The integrated M+GR+AC classification scheme reached 93.6%. These results demonstrate that combining deep-learning-based resistivity reconstruction with conventional logging parameters can improve the reliability of water-flooded zone identification in complex heterogeneous reservoirs and provide technical support for remaining-oil evaluation and refined development of high water-cut oilfields.