Revisiting gray box model learning and Kalman filtering with subspace based model identification
In numerous industrial projects, engineers face the crucial task of extracting physical parameters and signals from real-world data. When employing gray box state space model learning, a key challenge lies in generating reliable initial estimates to ensure convergence towards precise parameter values. Similarly, in utilizing Kalman filtering for signal reconstruction, a common obstacle is selecting reliable covariance matrices to attain accurate estimates. This presentation delves into how subspace-based model identification can effectively address these challenges. These solutions, relying solely on linear algebra, center on determining the similarity transformation between various state space realizations. Through both simulated and real data demonstrations, we illustrate the efficacy of these approaches.
Praktische info:
Leertraject
georganiseerd door:
Lesgever/spreker
Main research topics: model learning, learning from data, system identification, estimation theory, state space model, gray box model, linear parameter varying model, linear fractional representation, subspace-based methods, numerical optimization
Main applications: electrical engineering, aeronautics, heat transfer, flexible, cable-driven manipulators, vehicle tire/road interactions and image processing.
Back to the Roots Seminar Series
The ERC research project "Back to the roots of data-driven dynamical system identification", led by Prof. Dr. Bart De Moor (KU Leuven, ESAT-STADIUS), focuses on system identification, where mathematical models are derived from observed data generated by systems such as medical monitoring, electricity consumption and industrial processes. Utilizing optimization algorithms, one seeks to identify the best model in a chosen model class. This methodology finds widespread application across thousands of use cases within the AI community. However, there is no guarantee that optimization algorithms will find the best model. Present-day optimization practices are heuristic in nature, yielding results that may not be reproducible and consequently difficult to interpret.
The main objective of the Back to the Roots project is to develop a theoretical framework that combines model classes and optimization algorithms, enabling the calculation of the optimal model within the specified model class with 100% certainty.
