随着油气勘探技术的不断进步，测井岩性识别技术成为储层预测中的关键手段。然而，传统的测井岩性智能识别方法严重依赖于人工经验和专业知识，不仅成本高且效率低下，而且在处理大规模数据中的复杂非线性关系时表现不佳。因此，基于深度学习的测井岩性智能识别方法的研究是一个十分重要的课题。通过利用残差网络（ResNet）结合卷积神经网络（convolutional neural network, CNN）的架构，实现测井数据的智能化岩性识别。具体而言，使用随机森林算法进行特征重要性评估，选取相关性较强的特征属性，构建了ResNet-CNN模型，并且用混淆矩阵方法评估模型性能。实验结果显示，本文中构建的ResNet-CNN模型的准确率达到93.5%，明显优于其他两种方法。因此，基于ResNet-CNN的方法在测井岩性智能识别中展现出很高的应用价值。

With the continuous advancement of oil and gas exploration technologies, well logging lithology identification has been established as a key tool in reservoir prediction. However, traditional intelligent lithology identification methods in well logging are found to be heavily dependent on human experience and professional knowledge, which are not only considered costly and inefficient, but are also observed to underperform when handling complex nonlinear relationships in large-scale data. Therefore, the investigation of deep learning-based well logging lithology identification methods has been recognized as a research topic of significant importance. To achieve intelligent lithology recognition of well logging data, a ResNet architecture integrated with CNN (convolutional neural network) was implemented. Feature importance was evaluated using the Random Forest algorithm, and strongly correlated features were selected to construct the ResNet-CNN model. The performance of the model was systematically evaluated using confusion matrices. Experimental results demonstrate that an accuracy rate of 93.5% was achieved by the ResNet-CNN model, which is found to significantly outperform two other comparative methods. It is therefore concluded that the ResNet-CNN-based methodology demonstrates substantial application value in intelligent lithology identification of well logging data.