钻井液流变性是钻井液流动和变形的特性，对于携带与悬浮岩屑、提高钻进速度至关重要，准确掌握钻井液流变参数是保证井眼清洁与高效钻进的前提。提出一种基于卷积神经网络(Convolutional Neural Network，CNN)的钻井液流变参数智能识别方法，通过磁力搅拌产生稳定的钻井液流动图像，利用多种数据增强方法增加图像数量并建立数据库，增强模型的鲁棒性和泛化能力。优化AlexNet卷积神经网络算法，构建钻井液流变参数识别模型。将数据库划分为训练集：验证集：测试集=7：2：1，对训练集进行迭代训练并通过验证集调整参数获得最佳拟合模型。此外，运用混淆矩阵、卷积核可视化技术和类激活技术(Gradient-weighted Class Activation Mapping，Grad-CAM)对模型进行多方位评估。结果表明：(1)钻井液流变参数识别模型对钻井液塑性黏度测试的宏精确率为95.2%，宏召回率为94.7%，宏F₁值为0.95。(2)对钻井液表观黏度测试的宏精确率为91.6%，宏召回率为91.5%，宏F₁值为0.90。(3)利用卷积核可视化技术和Grad-CAM对特征提取进行可视化处理，发现钻井液波纹形状和大小会影响模型流变参数识别准确度。(4)室内测试结果表明，该模型的测试误差为±2 mPa·s，在设计允许范围以内，具有较高的识别精确度和稳定性。所提出的钻井液流变参数实时智能识别方法可为安全、快速和准确地进行钻井液流变性测试提供智能化技术思路。

The rheology of drilling fluid, which characterizes its flow and deformation, is vital for transporting and suspending rock cuttings as well as for enhancing the drilling rate. Precise control of drilling fluid rheological parameters is essential to ensure borehole cleanliness and efficient drilling. This paper proposes an intelligent identification method for drilling fluid rheological parameters based on Convolutional Neural Networks (CNNs). The method employs magnetic stirring to generate stable images of drilling fluid flow, uses various data augmentation methods to increase the number of images and create a database, thereby enhancing the model’s robustness and generalization capabilities. The AlexNet CNN algorithm is optimized to construct a model for identifying the rheological parameters of drilling fluids. The database is divided into a training set, validation set, and test set in a 7:2:1 ratio. Additionally, the model is evaluated through multiple approaches, including the confusion matrix, convolution kernel visualization technique, and Gradient-weighted Class Activation Mapping (Grad-CAM). The results indicate that: (1) The model achieves a macro precision of 95.2%, macro recall of 94.7%, and a macro F1 score of 0.95 for the plastic viscosity test of drilling fluids. (2) For the test of the apparent viscosity of drilling fluids, it achieves a macro precision of 91.6%, macro recall of 91.5%, and a macro F1 score of 0.91. (3) The utilization of convolution kernel visualization and Grad-CAM for feature extraction visualization reveals that the shape and size of drilling fluid ripples influence the accuracy of rheological parameter identification; (4) Indoor testing results demonstrate that the model has a test error of ±2 mPa·s within the allowable design range, indicating high prediction precision and stability. The proposed real-time intelligent identification method for drilling fluid rheological parameters can provide an intelligent technical approach for the safe, rapid, and accurate testing of drilling fluid rheology.