Abstract
In this paper, we propose a suboptimal moving horizon estimator for nonlinear systems. For the stability analysis we transfer the "feasibility-implies-stability/robustness" paradigm from model predictive control to the context of moving horizon estimation in the following sense: Using a suitably defined, feasible candidate solution based on the trajectory of an auxiliary observer, robust stability of the proposed suboptimal estimator is inherited independently of the horizon length and even if no optimization is performed.
Introduction
State estimation is the process of inferring the internal state of a system from noisy external measurements. Moving Horizon Estimation (MHE) is a powerful tool for this task, as it optimizes over a window of past data while respecting physical constraints. However, MHE requires solving a complex nonlinear optimization problem at every time step. In real-time engineering, computational limits often prevent reaching a fully optimal solution, which can lead to instability in standard MHE formulations. This paper addresses how to guarantee that an estimator remains stable even when the optimizer is stopped early.
Methodology
The proposed framework bridges the gap between theory and real-time practice by adopting the "feasibility-implies-stability" paradigm. The approach consists of three key elements:
- The Auxiliary Observer: A computationally inexpensive, robustly stable observer (such as a nonlinear Luenberger observer) runs alongside the MHE.
- The Candidate Solution: This observer provides a "warm start" candidate solution for the MHE optimization. Because the observer is already stable, this candidate acts as a guaranteed safety net.
- The Cost Decrease Condition: Rather than requiring full optimality, the MHE solver only needs to find a feasible solution with a cost no higher than the candidate solution provided by the observer.
By satisfying these conditions, the MHE inherits the stability of the auxiliary observer. The authors use the property of exponentially input/output-to-state stability (e-IOSS) to prove that the estimation error remains bounded and converges toward zero for any horizon length ($N \ge 1$).
Key Results
The theory was validated using a numerical simulation of a chemical batch reactor ($2A \leftrightarrow B$ reaction), a standard benchmark in process control. The performance of the suboptimal MHE was tested by varying the number of solver iterations.
As illustrated in the simulation, the estimator remains stable even with zero iterations ($i=0$), where it follows the auxiliary observer. With just two iterations ($i=2$), the estimate improves drastically, and with five iterations ($i=5$), the results are virtually indistinguishable from a fully converged MHE. This proves that the optimizer can be used to improve accuracy as a "bonus" without being a prerequisite for stability.
Conclusion
This framework allows for the implementation of MHE in high-speed applications where computational resources are limited. By ensuring that stability is inherited from a simple auxiliary observer, engineers can use any number of optimization iterations to improve performance without risking estimator divergence. Future research will focus on relaxing the global stability requirements for the auxiliary observer to accommodate a wider range of complex nonlinear systems.
Summary updated