Multistep prediction of dynamic uncertainty under limited data

Engineering systems are growing in complexity, requiring increasingly intelligent and flexible methods to account for and predict uncertainties in service. This paper presents a framework for dynamic uncertainty prediction under limited data (UPLD). Spatial geometry is incorporated with LSTM networks to enable real-time multistep prediction of quantitative and qualitative uncertainty over time. Validation is achieved through two case studies. Results demonstrate robust prediction of trends in limited and dynamic uncertainty data with parallel determination of geometric symmetry at each time unit. Future work is recommended to explore alternative network architectures suited to limited data scenarios.