The detection of anomalies in dam deformation is paramount for evaluating structural integrity and facilitating early warnings, representing a critical aspect of dam health monitoring (DHM). Conventional data-driven methods for dam anomaly detection depend extensively on historical data; however, obtaining annotated data is both expensive and labor-intensive. Consequently, methodologies that leverage unlabeled or semi-labeled data are increasingly gaining popularity. This paper introduces a spatiotemporal contrastive learning pretraining (STCLP) strategy designed to extract discriminative features from unlabeled datasets of dam deformation. STCLP innovatively combines spatial contrastive learning based on temporal contrastive learning to capture representations embodying both spatial and temporal characteristics. Building upon this, a novel anomaly detection method for dam deformation utilizing STCLP is proposed. This method transfers pretrained parameters to targeted downstream classification tasks and leverages prior knowledge for enhanced fine-tuning. For validation, an arch dam serves as the case study. The results reveal that the proposed method demonstrates excellent performance, surpassing other benchmark models.

To overcome the challenges of limited data, domain distribution differences, and the need for retraining in unsupervised learning methods for cross-scenario anomaly detection in dams, this study introduces a novel approach; the Temporal Displacement Subdomain Adaptation Network (TDSAN) combines temporal convolutional networks with subdomain adaption. This study presents the first application of subdomain adaptation for cross-scenario anomaly detection in dams, addressing distribution shifts across varying operational conditions. The proposed method comprises three key components: a feature extraction network leveraging temporal convolutional layers to capture long-term displacement patterns, a classifier network with fully connected layers to distinguish between normal and anomalous behaviors, and a domain alignment module that uses Local Maximum Mean Discrepancy (LMMD) to align feature distributions between the source and target domains, thereby enhancing the method’s robustness. The approach was validated using data from gravity and arch dams in a specific canyon region in China. The results show that the proposed method demonstrates high classification accuracy and stability in both same-domain and cross-domain scenarios. Compared to other state-of-the-art methods, the proposed approach demonstrates superior classification accuracy and more reliable risk control. This makes it particularly well-suited for cross-domain applications, which are prevalent in real-world engineering scenarios, thereby significantly enhancing its practical applicability.