16 April 2026 in IEEE Access
PJPedro Ribeiro Mendes Júnior
Pedro Ribeiro Mendes Júnior
+11
Causal discovery is fundamental for understanding the underlying structure of complex systems, integrating domain knowledge with statistical methods, machine learning models, and observational and interventional data. While originally rooted in the social and health sciences, modern causal inference techniques have expanded to address challenges in dynamic systems across domains such as earth and environmental sciences, healthcare, and oil and gas engineering. To address the challenges posed by nonstationarity and temporal variability in time-series data, we propose a novel methodology for causal discovery in multi-scenario environments. The proposed approach simultaneously uncovers temporal dependencies and cause–effect relationships, enabling a more comprehensive interpretation of complex system dynamics. Although global causal analysis can provide valuable insights within fixed temporal windows, its limitations over extended periods motivate the need for scenario-aware analysis. By combining automatic change-point detection with human-in-the-loop intervention, our method identifies distinct operational scenarios and their associated causal structures. Furthermore, we introduce two classes of causal temporal relationships: Invariant Causal Temporal Relationships (ICTRs), which remain consistent across all scenarios, and Variant Causal Temporal Relationships (VCTRs), which occur only under specific conditions or during particular time intervals. The effectiveness of the proposed methodology is demonstrated through extensive experiments on synthetic data, benchmark datasets, and real-world oil field production data, highlighting its ability to capture both stable and scenario-dependent causal dynamics.