The use of a cell filter for state estimation in closed-loop NMPC of low dimensional systems

Combining variants of the Kalman filter and moving horizon estimation (MHE) with nonlinear MPC has been studied before. The MHE is appealing due to its ability to impose constraints and demonstrated superiority over extended Kalman filter. However, nonlinear MPC based on MHE requires solutions to two back to back nonlinear programs. In this paper we propose to use the cell filter (CF) to provide state feedback to the MPC regulator. The cell filter is a piecewise constant approximation of the conditional probability density of the states, whose temporal evolution is modeled by an aggregate Markov chain. Since the CF is based on discretized state, input and output spaces, the curse of dimensionality limits its application to low dimensional and constrained systems. In this paper we present simulation examples of closed-loop MPC for a nonlinear reactor and agricultural pest control based on state feedback from both CF and MHE.