State space partitioning based on constrained spectral clustering for block particle filtering

To overcome the curse of dimensionality particle filter (PF), block PF (BPF) partitions state space into several blocks smaller dimension so that correction and resampling steps can be performed independently on each block. Despite its potential performance, this approach has a practical limitation. BPF requires an additional input compared to classical PF: partition from which are defined. A poor choice will not offer expected performance gain. In paper, we formulate partitioning problem as clustering propose data-driven method based constrained spectral (CSC) automatically provide appropriate partition. We design generalized contains two new steps: (i) estimation vector correlation matrix particles, (ii) CSC using estimate determine blocks. The proposed succeeds in providing online restricted size grouping most correlated variables. This allows escape by reducing variance filtering distribution while limiting level bias. Since our relies particles already necessary generate part BPF, computation overhead is limited.