A recent arXiv paper introduces GraphNARX, a graph-based nonlinear autoregressive exogenous model for structural dynamics. The model is explicitly built to account for the geometry of structures, a factor that is often simplified or ignored in conventional time-series models.

The abstract describes the model as one that represents structural geometry and models dynamic behavior as a nonlinear autoregressive process with exogenous inputs. By leveraging graph structures, the approach aims to capture spatial relationships that are inherent to physical structures.

The paper's abstract does not provide details on the model's architecture, training, or validation results. It focuses on the conceptual foundation, positioning GraphNARX as a step toward geometry-aware dynamic modeling for structural engineering applications.