CMStatistics 2021: Start Registration
View Submission - CMStatistics
B0764
Title: Data-driven identification of dynamical systems Authors:  Michelle Carey - Univerity College Dublin (Ireland) [presenting]
James Ramsay - McGill University (Canada)
Abstract: Dynamical systems facilitate a causal explanation for the drivers and impediments of a process. But do they describe the behaviour of observed data? And how can we quantify the models' parameters that cannot be measured directly? These two questions are addressed by estimating the solution and the parameters of a linear dynamical system from incomplete and noisy observations of the processes. This methodology builds on the parameter cascading approach, where a linear combination of basis functions approximates the implicitly defined solution of the dynamical system. Then the systems' parameters are estimated so that this approximating solution adheres to the data. By taking advantage of the linearity of the system, we have simplified the parameter cascading estimation procedure, and by developing a new iterative scheme, we achieve fast and stable computation. We illustrate our approach by obtaining linear dynamical systems that represent real data from medicine, climatology and biomechanics.