Sequential Monte Carlo (SMC), also known as particle filtering, is
a powerful simulation method to perform Bayesian inference for state
space models. It estimates system states by recursively updating samples
and weights, or the so called particles, to approximate the underlying
posterior distribution of system state, which evolves over time as new
observations arrive sequentially.
However, with a finite particle set, as constrained by available
computing resources, the practical performance of SMC could be sensitive
to the specified prior. Other restrictions with applying SMC include
handling constraints and parameter estimation. Regular SMC cannot handle
constraints that are common in practical systems. Parameter estimation
in dynamic models has been a challenging problem
since system state estimation itself is already a tough one.
In our research, we introduced a numerical smoothing method, moving
horizon estimation, to get smoothed prior information for the start of
SMC simulation. For parameter estimation, we developed a moving window
EM algorithm to get the MLE of model parameters and simultaneous state
estimation.
Meet the speaker in Room 212 Cockins Hall at 4:30
p.m. Refreshments will be served.
