HUN-REN's SZTAKI researchers have been awarded a grant from the US Air Force Office of Scientific Research (AFOSR). Their proposal entitled "Combining analytic and data-driven machine-learning based methods for efficient adaptive dynamic modelling" was prepared by Roland Tóth, Tamás Péni and Bálint Vanek, leading researchers at HUN-REN SZTAKI, all three researchers actively contribute to the research carried out at ARNL.
The aim of their research is to develop new theory and new algorithmic tools to solve data-driven modelling problems for general systems, including also vehicles, where there is insufficient prior knowledge of the system dynamics, and therefore an automated modelling process is needed that can adapt the model – even online – based on the measured data from the system.
Specifically, the research addresses the case where and when the model deviates significantly from a given, physically well-known baseline model of the system dynamics, either due to manufacturing tolerances, inaccurate knowledge of complex aspects of the dynamics, or sudden changes in the dynamics during operation.
In order to ensure the fast adaptability of the basic system model, the ARNL researchers aim at an efficient fusion of the prior knowledge of the dynamics with the measured system data.
Therefore, they propose new model-extension methods based on state-of-the-art approaches to machine learning-based system identification: Gaussian regression methods and deep learning methods based on state-space artificial neural network (SS-ANN) models.
Building upon the extended model structures, a novel control design methodology is to be developed that can be employed to design efficient, optimal, feedback and prediction-based controllers for fast real-time operations on the extended models. The ARNL researchers aim to create an adaptive learning algorithm that can be derived from the fusion of efficient model learning and well-established controller design methodology to ensure the devices’ continuous operation with proper safety and performance guarantees.