Road Grade Determination Based on Improved NARX Neural Network
Haoyun Jin, Hongliang Zhou, Zhen He, Haifeng Liu · 2024
This paper proposes an improved method considering the significant impact of road unevenness on vehicle performance and safety, as well as the problems of model complexity and insufficient generalization ability of existing road unevenness identification algorithms. First, simulated white noise is used to generate road irregularity data, combined with an improved NARX (Nonlinear autoregressive with external input) neural network and dropout technology to reduce model complexity and improve generalization capabilities. NARX neural networks excel in processing time series data, while dropout technology prevents overfitting by randomly discarding neurons, thereby enhancing the model's generalization ability. Following simulation verification, this method demonstrates higher accuracy and improved generalization in identifying newly generated road unevenness data, thereby enhancing the model's adaptability across diverse road conditions. Lastly, the road roughness estimation results were utilized to determine the road grade, validating the reliability and effectiveness of this method in practical engineering applications.