TY - GEN
T1 - Research on Road Roughness Identification Based on LSTM-KAN Neural Network
AU - Cong, Shengze
AU - Wu, Zhicheng
AU - Zhang, Yuxiao
AU - Yang, Lin
N1 - Publisher Copyright:
© 2025 Author(s).
PY - 2026/1/5
Y1 - 2026/1/5
N2 - The power spectrum of road unevenness is an important input signal of the automobile vibration system, and the accurate power spectral density is of great significance to the driving smoothness of the automobile. In this paper, the long short-term memory plus the Kolmogorov-Arnold Network (LSTM-KAN) neural network is applied to the identification of road surface unevenness based on the time domain response of vertical acceleration of the body. Based on the time-domain data of vertical acceleration obtained in the ADAMS/Car Ride random road input smoothness simulation test, the dataset is established for training LSTM-KAN and LSTM neural networks and road surface unevenness recognition tests. The comparison results of the pavement unevenness identification test between LSTM-KAN and LSTM show that the recognition accuracy of LSTM-KAN is increased by 2.44% and recall increased by 1.93%compared with LSTM. Therefore, LSTM-KAN neural network has significant advantages over traditional algorithms for pavement unevenness identification.
AB - The power spectrum of road unevenness is an important input signal of the automobile vibration system, and the accurate power spectral density is of great significance to the driving smoothness of the automobile. In this paper, the long short-term memory plus the Kolmogorov-Arnold Network (LSTM-KAN) neural network is applied to the identification of road surface unevenness based on the time domain response of vertical acceleration of the body. Based on the time-domain data of vertical acceleration obtained in the ADAMS/Car Ride random road input smoothness simulation test, the dataset is established for training LSTM-KAN and LSTM neural networks and road surface unevenness recognition tests. The comparison results of the pavement unevenness identification test between LSTM-KAN and LSTM show that the recognition accuracy of LSTM-KAN is increased by 2.44% and recall increased by 1.93%compared with LSTM. Therefore, LSTM-KAN neural network has significant advantages over traditional algorithms for pavement unevenness identification.
KW - driving smoothness
KW - neural network algorithm
KW - road surface unevenness
KW - root mean square error
UR - https://www.scopus.com/pages/publications/105027429385
U2 - 10.1145/3783779.3783795
DO - 10.1145/3783779.3783795
M3 - Conference contribution
AN - SCOPUS:105027429385
T3 - Proceedings of 2025 3rd International Conference on Mathematics and Machine Learning, ICMML 2025
SP - 91
EP - 97
BT - Proceedings of 2025 3rd International Conference on Mathematics and Machine Learning, ICMML 2025
PB - Association for Computing Machinery, Inc
T2 - 2025 3rd International Conference on Mathematics and Machine Learning, ICMML 2025
Y2 - 14 November 2025 through 16 November 2025
ER -