Abstract: To address practical challenges in geological drilling, including pronounced inter-well distribution shifts, stringent privacy constraints, and the lack of manual annotations, this paper proposes an intelligent monitoring method based on federated dictionary learning. Centered on sparse dictionary representations, the proposed approach integrates an event-driven heuristic scoring and alignment mechanism to enable interpretable discrimination and abnormal pattern recognition without requiring human-labeled data. Meanwhile, under a data-locality constraint where raw data never leave local sites, the method incorporates collaborative multi-well dictionary training and a sample-size–weighted server-side aggregation scheme to enhance robustness and cross-well generalization across heterogeneous drilling environments. Experiments are conducted on field logging data collected from multiple real-world drilling projects, where a multi-well federated dataset is constructed for evaluation. The results demonstrate that the proposed method achieves superior monitoring performance under multi-well settings, with an average separability score of 4.018 and an average weakly supervised precision of 0.804, indicating stable identification of typical abnormal events in the absence of annotations. These findings substantiate the effectiveness of the event labeling mechanism in improving model interpretability and cross-well generalization, and provide a feasible and effective technical pathway for distributed intelligent monitoring in complex geological environments.