Talk by Ran Chen
© 2024 EPFL
On Thursdyay, 18/1/24, Ran Chen will give the talk "Advancing Closed-Loop Identification in Sparse Linear Systems:
Insights from Controller Design through Fractional Parameterization and Sparsity Invariance". The seminar will start at 11h00 in ME C2 405.
Abstract:
System identification, the technique of building reliable system models based on data, has been widely used in engineering. In the realm of modern cyber-physical systems (CPS), we encounter challenges due to their expansive sparse structures, and the fact that the identification data are primarily sourced from closed-loop operations. This situation necessitates the creation of innovative methods designed for closed-loop identification in CPS, utilizing the understanding of their sparsity structures to improve the identification.
To address such problems, this presentation is divided into two sections. We first introduce a novel identification approach, namely the dual input-output parameterization (D-IOP). This method is utilized for identifying LTI systems using closed-loop data, and is inspired by the recent advancements in input-output parameterization (IOP) used for stabilizing controller design. Analogously, D-IOP characterizes the unknown system using closed-loop transfer functions and ensures the closed-loop stability of the system once identified.
In the second part, we explore how a common closed-loop identification technique, known as the dual-Youla parameterization (D-YP), is adapted from identifying LTI SISO systems to MIMO systems with sparsity structures. This adaptation is mindful of the system's inherent sparsity pattern and is achieved by integrating the Sparsity Invariance (SI) Theorem, which introduces a set of linear equality constraints reflecting the sparsity pattern. Provided the knowledge of the stabilizing controller, the adapted method is proven to offer a consistent estimate of the plant, ensuring its closed-loop stability and desired sparsity structure. Eventually, we will demonstrate the effectiveness of the proposed algorithms by identifying example systems directly from closed-loop noisy data.
Bio:
Ran Chen received his B.Sc. degree in Aerospace Engineering (Cum Laude) at TU Delft in 2021, since when he has been pursuing a Master's degree in Robotics, Systems and Control at ETH Zürich. He is currently working on an internship at the R&D group of the division of System Drives, ABB Switzerland. He is interested in the fields of modeling, simulation, system identification and control theories, with a particular fascination for their applications on large-scale or distributed systems.