为精确评估城市群多层客运网络的鲁棒性，提出一种耦合过载失效与修正渗流理论的级联失效模型。首先，通过Space-L构建包括公路、铁路等子网与换乘关系的多层耦合网络，并通过融合站点容量与拓扑属性的方法为节点加权；其次，在负载-容量模型基础上，引入综合考虑全局、个体和局部3个维度的渗流失效判据，用以识别传统过载模型难以捕捉的结构性失效节点；再次，构建网络连通效率指标，以级联失效后各连通分量内节点数量占初始网络节点总数的比例量化网络服务功能的保持程度；最后，以呼包鄂乌城市群为对象，模拟在不同攻击强度（P=20%、P=70%）、策略（度中心性、PageRank、最大连通子图及加权修正介数攻击）及负载（低、中、高）下的改进后级联失效模型的鲁棒性。研究显示：低强度攻击主要引发过载失效，高强度攻击则转为渗流失效；4种攻击策略下，改进后的级联失效模型预测网络崩溃进程分别提前了47.4%, 45.5%, 50%, 59.5%；结构均衡的长途汽车网络对渗流失效敏感，而轴辐式的铁路网络对过载失效敏感。结果表明，该模型通过双重失效机理，能更精准评估网络鲁棒性，为城市群各层客运网络的差异化防护提供重要的决策支持。

To accurately evaluate the robustness of multi-layer passenger transport networks in urban agglomerations, this study proposes a cascading failure model that couples overload failure with a modified percolation theory. First, a multi-layer coupled network incorporating sub-networks such as highways and railways, along with transfer relationships, is constructed using the Space-L method. Nodes are weighted by integrating station capacity with topological attributes. Second, based on the load-capacity model, a percolation failure criterion that comprehensively considers global, individual, and local dimensions is introduced to identify structural failure nodes that are difficult to capture using traditional overload-based models. Then, a network connectivity efficiency metric is constructed to quantify the degree to which network service functions are maintained, based on the proportion of nodes in each connected component after cascading failure to the total number of nodes in the initial network. Finally, the improved cascading failure model is simulated using the Hohhot-Baotou-Ordos-Ulanqab urban agglomeration as a case study, analyzing robustness under different attack intensities (P=20%, P=70%), strategies (degree centrality, PageRank, giant connected component, and weighted modified betweenness attacks), and load levels (low, medium, high). The research reveals that low-intensity attacks primarily trigger overload failures, whereas high-intensity attacks shift toward percolation failures. Under the four attack strategies, the improved cascading failure model predicts the network collapse process to occur 47.4%, 45.5%, 50%, and 59.5% earlier, respectively. Furthermore, the structurally balanced long-distance bus network is sensitive to percolation failure, while the hub-and-spoke railway network is sensitive to overload failure. The results indicate that by incorporating dual failure mechanisms, this model can more accurately assess network robustness, providing crucial decision-making support for the differentiated protection of various passenger transport layers within urban agglomerations.