Rock identification plays a critical role in oil and gas exploration and development, particularly in reservoir evaluation and the optimization of development plans. However, traditional well logging interpretation methods often face challenges in accurately extracting key features and effectively classifying complex reservoir conditions, such as low-permeability and highly heterogeneous reservoirs. To address these challenges, a deep learning model was proposed that integrates CNN(convolutional neural networks), LSTM(long short-term memory) networks, and attention mechanisms. This model combines local feature extraction, temporal dependency modeling, and feature weighting to achieve accurate classification of complex reservoir lithology and efficient prediction of reservoir parameters. The dataset utilized in this study was based on well logging data from gas fields drilled between 2023 and 2024, with GR(gamma ray), DT(acoustic time difference), and RT(resistivity) selected as input features. After performing data cleaning, missing value imputation, wavelet denoising, and standardization, a high-quality dataset was generated. Lithology classification results were used as the output of the model. Through model training and validation, an overall accuracy of 94.89% is achieved for lithology classification tasks, demonstrating that the model is capable of handling the nonlinear characteristics of complex reservoirs. Furthermore, it provides a scientific foundation for reservoir lithology identification and gas reservoir development through dynamic reservoir analysis.