Stations’ irregular demand evolutions exhibit different patterns under unplanned disruptions given the disruption types, operation management, and passenger travel choices. Existing studies mainly focus on irregular demand detection, but lack a deep analysis of its patterns that are essential in guiding targeted countermeasure development, such as train timetable adjustment and emergency evacuation planning. Daily demand fluctuations and imbalanced disruption types make clustering and analyzing irregular demand patterns a challenging task. To address the problem, this paper presents a robust demand decomposition and clustering model for irregular demand pattern analysis under unplanned disruptions. It consists of, an irregular demand detection model, i.e., a robust principal component analysis (RPCA)-based model, and an irregular demand pattern clustering model, i.e., a robust discretization-based clustering model. The RPCA-based model is used to identify irregular demand by decomposing observed demand into regular and irregular demand. The robust discretization-based clustering model, which considers the non-sparse issue and the imbalanced issue of irregular demand sequences (caused by the daily demand fluctuations and imbalanced disruption types, respectively), is developed with a customized discrete input module, distance metric module, and Hierarchical Density-Based Spatial Clustering of Applications with Noise (H-DBSCAN) module, to identify irregular demand patterns. We validate the proposed model using synthetic and real-world data from the Hong Kong Mass Transit Railway system. Compared with other baseline models, the results demonstrate that the proposed model can effectively identify irregular demand patterns rather than wrongly mixing them or classifying them as outliers. The factors that lead to each identified entry/exit irregular demand pattern are analyzed at a station level and illustrated through binary tree-based architectures, which enable operators to predict irregular demand patterns under unplanned disruptions.