Details
| Original language | English |
|---|---|
| Pages (from-to) | 5717-5722 |
| Number of pages | 6 |
| Journal | IFAC-PapersOnLine |
| Volume | 56 |
| Issue number | 2 |
| Early online date | 22 Nov 2023 |
| Publication status | Published - 2023 |
| Event | IFAC World Congress 2023 - Pacific Convention Plaza Yokohama, Yokohama, Japan Duration: 9 Jul 2023 → 14 Jul 2023 |
Abstract
For state and parameter estimation in vehicles, Kalman filters, especially nonlinear extensions like the extended Kalman filter (EKF) and unscented Kalman filter (UKF), are very common. However, the estimation accuracy is highly dependent on the quality of the model used in the process update of the Kalman filter. Model errors can result from non-modeled dynamics that are either unknown or very difficult to describe. In recent years data-driven approaches for state estimation are the subject of research with promising results in estimation accuracy and reduced implementation effort. In this work, both a model-based method with an UKF and a data-driven approach based on recurrent neural networks (RNN) are implemented and combined to two hybrid methods for the application of state and parameter estimation in a truck-semitrailer for three different loading conditions. Hybrid estimation architectures promise to combine the advantages of model-based and data-driven methods to achieve better estimation accuracy than their standalone components. To the best knowledge of the authors, this work is the first to extend the field of hybrid state estimation to semitrailers estimating the truck steering angle, articulation angle, and the trailer's lateral and vertical tire forces. Four estimation architectures (an UKF, one purely data-driven method, and two hybrid methods) are optimized and compared to each other regarding estimation accuracy. The UKF is optimized with a particle swarm optimization (PSO) while the hyperparameters of the data-driven method are tuned with the asynchronous successive halving algorithm (ASHA) to result in a fair comparison. All methods are developed and compared based on an experimental data set from a test vehicle.
Keywords
- Hybrid Estimation, Kalman Filtering, Recurrent Neural Networks, State and Input Estimation, Truck-Semitrailer Combination
ASJC Scopus subject areas
- Engineering(all)
- Control and Systems Engineering
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
In: IFAC-PapersOnLine, Vol. 56, No. 2, 2023, p. 5717-5722.
Research output: Contribution to journal › Conference article › Research › peer review
}
TY - JOUR
T1 - Hybrid State Estimation in a Semitrailer for Different Loading Conditions
AU - Ehlers, Simon Friedrich Gerhard
AU - Kortmann, Karl-Philipp
AU - Kobler, Jan Philipp
N1 - Publisher Copyright: Copyright © 2023 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)
PY - 2023
Y1 - 2023
N2 - For state and parameter estimation in vehicles, Kalman filters, especially nonlinear extensions like the extended Kalman filter (EKF) and unscented Kalman filter (UKF), are very common. However, the estimation accuracy is highly dependent on the quality of the model used in the process update of the Kalman filter. Model errors can result from non-modeled dynamics that are either unknown or very difficult to describe. In recent years data-driven approaches for state estimation are the subject of research with promising results in estimation accuracy and reduced implementation effort. In this work, both a model-based method with an UKF and a data-driven approach based on recurrent neural networks (RNN) are implemented and combined to two hybrid methods for the application of state and parameter estimation in a truck-semitrailer for three different loading conditions. Hybrid estimation architectures promise to combine the advantages of model-based and data-driven methods to achieve better estimation accuracy than their standalone components. To the best knowledge of the authors, this work is the first to extend the field of hybrid state estimation to semitrailers estimating the truck steering angle, articulation angle, and the trailer's lateral and vertical tire forces. Four estimation architectures (an UKF, one purely data-driven method, and two hybrid methods) are optimized and compared to each other regarding estimation accuracy. The UKF is optimized with a particle swarm optimization (PSO) while the hyperparameters of the data-driven method are tuned with the asynchronous successive halving algorithm (ASHA) to result in a fair comparison. All methods are developed and compared based on an experimental data set from a test vehicle.
AB - For state and parameter estimation in vehicles, Kalman filters, especially nonlinear extensions like the extended Kalman filter (EKF) and unscented Kalman filter (UKF), are very common. However, the estimation accuracy is highly dependent on the quality of the model used in the process update of the Kalman filter. Model errors can result from non-modeled dynamics that are either unknown or very difficult to describe. In recent years data-driven approaches for state estimation are the subject of research with promising results in estimation accuracy and reduced implementation effort. In this work, both a model-based method with an UKF and a data-driven approach based on recurrent neural networks (RNN) are implemented and combined to two hybrid methods for the application of state and parameter estimation in a truck-semitrailer for three different loading conditions. Hybrid estimation architectures promise to combine the advantages of model-based and data-driven methods to achieve better estimation accuracy than their standalone components. To the best knowledge of the authors, this work is the first to extend the field of hybrid state estimation to semitrailers estimating the truck steering angle, articulation angle, and the trailer's lateral and vertical tire forces. Four estimation architectures (an UKF, one purely data-driven method, and two hybrid methods) are optimized and compared to each other regarding estimation accuracy. The UKF is optimized with a particle swarm optimization (PSO) while the hyperparameters of the data-driven method are tuned with the asynchronous successive halving algorithm (ASHA) to result in a fair comparison. All methods are developed and compared based on an experimental data set from a test vehicle.
KW - Hybrid Estimation
KW - Kalman Filtering
KW - Recurrent Neural Networks
KW - State and Input Estimation
KW - Truck-Semitrailer Combination
UR - http://www.scopus.com/inward/record.url?scp=85184960151&partnerID=8YFLogxK
U2 - 10.1016/j.ifacol.2023.10.518
DO - 10.1016/j.ifacol.2023.10.518
M3 - Conference article
VL - 56
SP - 5717
EP - 5722
JO - IFAC-PapersOnLine
JF - IFAC-PapersOnLine
SN - 2405-8963
IS - 2
T2 - IFAC World Congress 2023
Y2 - 9 July 2023 through 14 July 2023
ER -