New paper published in Mechanical Systems and Signal Processing

In our work entitled "Emulating the dynamics of complex systems using autoregressive models on manifolds (mNARX)," we present a novel surrogate modeling approach. The mNARX algorithm efficiently approximates the responses of complex dynamical systems influenced by time-varying exogenous inputs. By incrementally constructing a problem-specific input manifold, mNARX decomposes the problem into more manageable subproblems, allowing effective scaling to complex nonlinear systems. Its compatibility with dimensionality reduction techniques makes it particularly suitable for dealing with high-dimensional problems, which are typically difficult to handle.

Since prior knowledge about the system being modeled is used during the manifold construction, mNARX is particularly well-suited for domains rich in such prior knowledge, such as civil and mechanical engineering applications. We therefore demonstrate the efficiency of mNARX by applying it to a coupled spring-mass system and an onshore wind turbine simulator. In these cases, the mNARX surrogate accurately emulates the true system responses over extended time periods at very low computational costs.

For more information, please follow this external pagelinkcall_made or the publication and this link for the associated report on our internal archive.