Learning-based data-enabled moving horizon estimation with application to membrane-based biological wastewater treatment process

Abstract: In this paper, we propose a data-enabled moving horizon estimation (MHE) approach for nonlinear systems. While the approach is formulated by leveraging Koopman theory, its implementation does not require explicit Koopman modeling. Lifting functions are learned from the state and input data of the original nonlinear system to project the system trajectories into the lifted space, where the resulting trajectories implicitly describe the Koopman representation for the original nonlinear system. A convex data-enabled MHE formulation is developed to provide real-time state estimates of the Koopman representation, from which the states of the nonlinear system can be reconstructed. Sufficient conditions are derived to ensure the stability of the estimation error. The effectiveness of the proposed method is illustrated using a membrane-based biological water treatment process.