Extended Kalman Filter Based Nonlinear Model Predictive Control

This paper formulates a nonlinear model predictive control algorithm based on successive linearization. The extended Kalman filter (EKF) technique is used to develop multi-step prediction of future states. The prediction is shown to be optimal under an affine approximation of the discrete state / measurement equations (obtained by integrating the nonlinear ODE model) made at each sampling time. Connections with previously available successive linearization based MPC techniques by Garcia (NLQDMC, 1984) and Gattu & Zafiriou (1992) are made. Potential benefits and shortcomings of the proposed technique are discussed using a bilinear control problem of paper machine.