(346u) Gsvd Carleman State Estimation of Nonlinear Discrete-Time Systems with Gross Errors

Reliable state estimation of evolving process dynamics in sensitive systems that experience noise and gross errors is corner stone to optimal decision-making and control of real industrial processes. Nonlinear systems present significant challenges in computation complexities resulting from stability issues. Here, the nonlinear system dynamic constraints are captured in a bilinear Carleman linearization of nonlinear dynamics. Gross error detection is performed using a data reconciliation of system measurements from constrained model expectations. The Generalized Singular Value Decomposition (GSVD) of process observations is used to identify gross error term and estimate the driving state variables. The proposed formulation enables robust analytical computation of state estimates across gross error interferences of outliers, bias and drift. An illustrative example with nonlinear isothermal CSTR is used to demonstrate that the proposed formulation reduces computational efforts.