Dealing with Ill Conditioning in Recursive Parameter Estimation for a Synchronous Generator

This paper presents how to deal with ill-conditioning in recursive parameter identification for a synchronous generator using subset selection, the Extended Kalman Filter (EKF), and the Iterated Extended Kalman Filter (IEKF). The problem of ill conditioning of the parameter estimation is caused by system inputs that are not rich enough to fully excite all system modes. We present how the quality of parameter estimated for ill-conditioned parameter estimation problems is significantly affected by noise and we show how by proper modifications to the EKF we still extract useful parameter estimates from low quality data. The methodology is based on the subset selection method, where a subset of parameters is fixed to prior values to reduce the ill-conditioning of the estimation problem. Simulation studies using a linearized model of a synchronous generator are presented to illustrate the concepts being studied in this work.