Abstract: Accurate identification and evaluation of water-flooded layers is important for enhancing oil recovery in the middle and later stages of oil field development. To address the limitations of traditional inversion methods in terms of insufficient accuracy and weak anti-interference for complex reservoirs, this paper proposes an original formation resistivity inversion model based on a fusion of convolutional neural network (CNN), bidirectional long short-term memory (BiLSTM), and attention mechanism (CNN-BiLSTM-Attention). This model utilizes CNN to extract local response features from log curves, BiLSTM to characterize sequence dependencies along the depth direction, and an attention mechanism to enhance weight allocation for effective features, thereby improving nonlinear modeling capability and inversion stability. Trained on conventional log curves from the target block, X Oilfield in the Pearl River Mouth Basin, the proposed CNN-BiLSTM-Attention model shows higher inversion accuracy than traditional models. On this basis, the resistivity attenuation rate (M) and comprehensive parameter (GPR) are introduced to construct a water-flooded layer identification chart, achieving fine classification of different water flooding levels with an accuracy of 92.71%. This model provides technical support for the evaluation and precise development of water-flooded layers in oil fields with high water cut.