常用的围岩分级方法受限于样本数量与经验归类，通常只适用于特定类型围岩，存在不良地质条件考虑不精细、多源地质信息利用不充分等局限，其智能化程度不高，在复杂地质条件下隧道围岩分级应用的精度和效率有限。针对这些问题，基于地质勘察设计、超前地质预报、岩石物理实验、掌子面测量等技术获取多源数据，综合考虑岩石性质、不连续体、地质背景、工程施工等多种因素对围岩分级的影响，提出一种融合多源异构信息的隧道围岩智能分级方法。综合多个网络单元构建关联多源异构信息与围岩级别的混合深度神经网络模型，可针对不同来源、不同类型的输入数据选用不同的网络结构开展特征学习，并面向围岩级别开展多分类目标学习，实现围岩智能分级，具有良好的灵活性。

Limited by sample number and empirical classification, the conventional rock classification methods are usually applicable for specific types of rock only. Due to lack of elaborate consideration of unfavorable geological conditions, insufficient utilization of multisource geological data, and low intelligentization level, the accuracy and efficiency of rock classification for tunnels under complex geological conditions are limited. To address these problems, with multisource data obtained from geological survey, advanced geological forecast, rock physical experiments, tunnel face survey and other technologies, an intelligent rock classification method for tunnels is proposed in this paper based on multisource heterogeneous data fusion by comprehensively considering the influence of rock properties, discontinuities, geological settings, engineering construction and other factors on rock classification. A hybrid deep neural network model that correlates multisource heterogeneous data with rock classification is built by integrating multiple network modules. In this model, different network structures are selected to carry out feature learning according to different types of input data from different sources, and multiclass target learning is conducted to realize intelligent rock classification with good flexibility.