Moving Horizon Estimation for Discrete-Time Linear Time-Invariant Systems Using Transfer Learning

Junyao Xie, Biao Huang

Abstract: In this article, we propose a novel moving horizon estimation method for discrete-time linear systems through transfer learning. Most moving horizon estimator designs require data from the considered systems of interest. However, practical processes might suffer from data availability issues, especially in a new or early operating environment. Motivated by the idea of transfer learning, this manuscript proposes a moving horizon estimator design using data from a similar but different system (i.e., source system) instead of the considered system (i.e., target system). Based on the data from the source system, we propose a novel moving horizon state estimation method for the target system and provide convergence and stability analyses. The state estimation error is upper bounded by a time-dependent sequence that is related to three types of similarities/differences between target and source systems, including initial conditions, disturbance levels, and model parameters. The effectiveness of the proposed approach is demonstrated through a numerical example.