CEAS EuroGNC 2022
|
A new hybrid model approach coupling a physical model and an artificial neural network through joint estimation
|
| Franz A. R. Enkelmann |
Research Associate, Technical University of Darmstadt, Flight Systems and Automatic Control, 64287, Darmstadt, Germany. | | Saleh H. Krüger |
Research Associate, Technical University of Darmstadt, Flight Systems and Automatic Control, 64287, Darmstadt, Germany. |
|
| Abstract:
The hybrid model approach presented in this paper is characterized by coupling a physical model and an artificial neural network, which are identified through joint estimation. By using the physical parameters as the interface between the physical model and the artificial neural network, the hybrid model structure combines both approaches directly. Joint estimation using a modified iterated Unscented Kalman Filter (UKF) ensures parallel updating of the dynamic, parameter and artificial neural network weight states. This approach represents an innovation in the context of the literature, in which hybrid model approaches are often cascaded, with separate identification. When using the hybrid model approach for complex modeling problems, especially for aerospace applications, high nonlinearity, demanding requirements for a robust filter and stability issues can occur. To handle nonlinearity the UKF is chosen. To achieve the required robustness and stability, a modification is derived that separates noisy state and covariance estimation. Testing the hybrid model approach with the modified UKF in a simulation environment on a simplified oscillating problem with time-variant parameters shows convincing results. Both the time-variant and constant parameters can be estimated and predicted with sufficient accuracy. The stability is confirmed by the use of the newly introduced Unscented Wiener Filter.
|
| Keywords: hybrid model; coupling; physical model; artificial neural network; joint estimation; unscented Kalman filter |
 View PDF CEAS-GNC-2022-049 |
| Franz A. R. Enkelmann, Saleh H. Krüger: A new hybrid model approach coupling a physical model and an artificial neural network through joint estimation. Proceedings of the 2022 CEAS EuroGNC conference. Berlin, Germany. May 2022. CEAS-GNC-2022-049.
|
| BibTeX entry:
@Incollection{CEAS-GNC-2022-049,
authors = {Enkelmann, Franz A. R. and Krüger, Saleh H.},
title = {A new hybrid model approach coupling a physical model and an artificial neural network through joint estimation},
booktitle = {Proceedings of the 2022 {CEAS EuroGNC} conference},
address = {Berlin, Germany},
month = may,
year = {2022},
note = {CEAS-GNC-2022-049}
}
|