International Symposium on Nonlinear Theory and its Applications
On Comparing Nonlinear Filtering Algorithms
PDF download (118.7KB)
In this paper we consider the performance of filtering algorithms, which means algorithms to retrieve the underlying state of a nonlinear system in a causal way. Since for nonlinear systems the optimal filter is computationally prohibitively expensive in general, one faces a trade?off between desired filtering accuracy and computational complexity. In order to ascertain whether a certain more powerful filtering algorithm is worth the computational effort, it is neccessary to carefully evaluate the potential benefits. We show that skill scores are a useful concept for this. We compare the performance of a number of filtering algorithms applied to a simple nonlinear system. We also discuss the perfomance in relation to the required computational resources. In general, higher accuracy requires more cpu-power.