State estimation and prediction using clustered particle filters
Abstract
Particle filtering is an essential tool for the estimation and prediction of complex systems including non-Gaussian features. A class of particle filters, clustered particle filters, is introduced for high-dimensional dynamical systems such as geophysical systems. The proposed method uses relatively few particles compared with the standard particle filter and captures the non-Gaussian features of the true signal, which are typical in complex nonlinear systems. The method is also robust for the difficult regime of high-quality sparse and infrequent observations and does not show any filter divergence in our tests. In the clustered particle filter, coarse-grained localization is implemented through the clustering of state variables and particles are adjusted to stabilize the filter.
Document Details
- Document Type
- Pub Defense Publication
- Publication Date
- Dec 05, 2016
- Source ID
- 10.1073/pnas.1617398113
Entities
People
- Andrew J. Majda
- Yoonsang Lee
Organizations
- New York University
- Office of Naval Research