State estimation of NonLinear Descriptor Systems using Particle Swarm Optimization based Extended Kalman Filter

Descriptor systems/Differential algebraic equation (DAE) naturally arise in dynamic systems and provides a better understanding of the system on basis of physical behaviour. Such form of system representation is different from Ordinary Differential Equations (ODEs) or conventional state space models on account of issues such as index, consistent initialization and requirement of smooth inputs. The methods for descriptor systems cannot be directly borrowed from other forms of system representation and hence be dealt with explicitly. The problem of state estimation for nonlinear descriptor systems (NLDS) has received considerable attention lately [1], [2], [3], [4], [5]. The proposed approach for state estimation of NLDS is based on a systematic study reported in [2]. The Extended Kalman Filter for descriptor systems so demonstrated in [2] is improved upon by implementation of an evolutionary algorithm. The proposed Particle Swarm Optimization based Extended Kalman Filter (PSO EKF) algorithm specifically differs from the former in terms of its approach to ensure consistent initial conditions. In order to incorporate an evolutionary algorithm within state estimation framework for descriptor systems, a novel approach for swarm initialization is put forth. The proposed algorithm is demonstrated on a NLDS of galvanostatic charge process and compared with DAE EKF of [2].