Soap-Particle Clustering for Non-linear System Identification
S.J. Thus · Research Repository (Delft University of Technology) · 2009
The Soap Particle Clustering algorithm for non-linear system identification is a novel clustering algorithm that can decompose the operating region of a dynamic system into multiple affine operating premises. The clustering algorithm performs comparable to Gustafson-Kessel fuzzy clustering for system identification, but without the need for the number of clusters and their shape a priori. The algorithm is inspired by the way soap particles move to the interface of grease and water. The artificial soap particles will move to the boundary of the hull of the measurement data cloud, thereby creating a description of the shape of the data cloud. Within this shape we extract the largest convex shapes, which become the clusters, and these in turn define the affine subsystems.