Akademik Veri Yönetim Sistemi
Automatica, cilt.189, 2026 (SCI-Expanded, Scopus)
In this paper, we present an online method for the identification of jump Markov linear systems (JMLSs). State inference is performed using a computationally tractable multi-model filter, while a recursive expectation maximization (EM) algorithm is derived to obtain the maximum likelihood estimates (MLEs) of the unknown parameters. Unlike existing online EM methods that rely on particle filters, our method utilizes analytical expressions that exploit the conditionally linear structure of JMLSs, enabling the identification of higher-dimensional models. We evaluate the performance of the method by identifying transfer functions, noise parameters, and transition probability matrices in simulated systems and compare it to state-of-the-art batch and particle filter-based online EM methods. The simulation results demonstrate that the suggested method requires significantly lower computational resources compared to the batch EM and particle filter-based online EM methods. In contrast to the batch methods, the proposed scheme is suitable for online system identification problems, which are frequently encountered in various real-time applications.