Bayesian dynamic graphical models combine graphical models with Bayesian state space models. They have been developed to model and forecast multivariate time series. The models represent any conditional independences related to causality which may exist across the multivariate time series by a graph, which then breaks the multivariate problem into simpler sub-problems. As a consequence, even though Bayesian dynamic graphical models can model what can be highly complex multivariate relationships, they are computationally straightforward.
These models are currently being developed for forecasting multivariate time series of traffic flows. They are, however, potentially suitable for any application involving flows, such as electricity flows, signal flows in telecommunication networks, flows of packages over the internet, flows of goods in supply chains, and so on.
Catriona Queen leads the research on Bayesian dynamic graphical models in the school.