On Improving the Efficiency of Bayesian Stochastic Subspace Identification

The recent development of a Bayesian stochastic subspace identification (SSI) algorithm for OMA has provided a new systematic and principled way of recovering posterior distributions over desired modal characteristics in an operational setting. Despite their many advantages, there is often a reluctance to adopt Bayesian methodologies in engineering practice because of their higher computational requirements. In the case of Bayesian SSI, this problem is even more relevant given the inherent speed of the traditional SSI algorithm. This has highlighted the need for a computationally efficient implementation of the Bayesian SSI algorithm, required to make Bayesian SSI a more competitive choice when considering multiple OMA approaches. This paper presents a novel solution, based on stochastic variational inference, and develops upon existing methods to speed up the Bayesian SSI algorithm. This method is evaluated using a simulated case study and subsequently compared to that of classical SSI and the current Bayesian SSI implementation.