A Robust Discrete State Approximation to the Optimal Nonlinear Filter for a Diffusion.

摘要

A robust computable approximation to the nonlinear filtering problem for a diffusion model is treated, where the system and data models are given by dx = f(x)dt + sigma(x)dz, dy = g(x)dt + dw. The approximation (with approximation parameter h) is robust in the sense that it is locally Lipschitz continuous in the data y(dot) (sup norm) uniformly in h and, as h approaches 0, it converges to the optimal filter for the diffusion. (Author)