Non-Parametric Identification of Linear Parameter-Varying Spatially-Interconnected Systems Using an LS-SVM Approach

This paper considers a general approach for the identification of partial differential equation-governed spatially-distributed systems. Spatial discretization virtually divides a system into spatially-interconnected subsystems, which allows to define the identification problem at the subsystem level. Here we focus on such a distributed identification of spatially-interconnected systems with temporal/spatial varying properties, whose dynamics can be captured by temporal/spatial linear parameter- varying (LPV) models. Inaccurate selection of the functional dependencies of the model parameters on scheduling variables may lead to bias in the identified models. Hence, we propose a non-parametric identification approach via a least-squares support vector machine (LS-SVM)-'non-parametric' estimation is in the sense that the model dependence on the scheduling variables is not explicitly parametrized. The performance of the proposed approach is evaluated on an Euler-Bernoulli beam with varying thickness.