Overcoming Observability Issues in State Estimation of Switched Converters: The Average Kalman Filter

In power electronics, components such as switches operate in a piecewise manner. Piecewise linear models are used to approximate these nonlinearities, breaking the system into segments where the dynamics can be treated as linear. A configuration-based model requires an observer with a matching structure: the configuration-based Kalman filter. A critical requirement for the correct implementation of a Kalman filter is the observability of the model being observed. The Kalman filter cannot operate correctly when applied to an unobservable model, resulting in failed or unreliable state estimation. The proposed solution is a Kalman filter of an zeroorder averaged affine model. The proposed Kalman filter is compared to the configuration-based Kalman filter on a boost converter in order to evaluate the overall performances.