Scholarlink

<p>Accurate and rapid recognition of road roughness is crucial for vehicle driving safety and the adaptive control of intelligent suspension systems. However, existing neural network models for road roughness recognition often suffer from input feature redundancy, leading to low recognition efficiency and a tendency to converge to local optima. To address these issues, this paper proposes a CNN-GRU model integrated with random forest feature selection and Bayesian optimization. The model combines the strengths of CNN in local spatial feature extraction and GRU in capturing long-term temporal dependencies. Random forest is employed to select key vehicle response features, while Bayesian optimization is utilized to determine the optimal hyperparameters of the network. Simulation results demonstrate that the proposed model outperforms the gated recurrent unit (GRU), long short-term memory (LSTM), back propagation (BP) neural network, and nonlinear autoregressive with exogenous inputs (NARX) network. Compared with these four models, the proposed model reduces the root mean square error (RMSE) by 57.14%, 51.61%, 95.34%, and 91.43% and improves the Pearson correlation coefficient (R) by 7.15%, 4.81%, 16.2%, and 11.2%, respectively. Additionally, real-vehicle test results verify that the model exhibits high prediction accuracy for road roughness under different pavement types and vehicle speeds, with the inference time stable within the range of 51–96&#xa0;ms. This confirms that the proposed model demonstrates excellent accuracy, robustness, and real-time performance in real road environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Road Roughness Recognition Based on Random Forest Feature Selection and Bayesian Optimization with CNN-GRU Network

  • 期刊论文

  • 发表日期：

    2026年5月18日

查看全文
  • Deli Yan,
  • Yuhao Zhang,
  • Shaohua Li,
  • Zekun Yang,
  • Jianwei Li

摘要

Accurate and rapid recognition of road roughness is crucial for vehicle driving safety and the adaptive control of intelligent suspension systems. However, existing neural network models for road roughness recognition often suffer from input feature redundancy, leading to low recognition efficiency and a tendency to converge to local optima. To address these issues, this paper proposes a CNN-GRU model integrated with random forest feature selection and Bayesian optimization. The model combines the strengths of CNN in local spatial feature extraction and GRU in capturing long-term temporal dependencies. Random forest is employed to select key vehicle response features, while Bayesian optimization is utilized to determine the optimal hyperparameters of the network. Simulation results demonstrate that the proposed model outperforms the gated recurrent unit (GRU), long short-term memory (LSTM), back propagation (BP) neural network, and nonlinear autoregressive with exogenous inputs (NARX) network. Compared with these four models, the proposed model reduces the root mean square error (RMSE) by 57.14%, 51.61%, 95.34%, and 91.43% and improves the Pearson correlation coefficient (R) by 7.15%, 4.81%, 16.2%, and 11.2%, respectively. Additionally, real-vehicle test results verify that the model exhibits high prediction accuracy for road roughness under different pavement types and vehicle speeds, with the inference time stable within the range of 51–96 ms. This confirms that the proposed model demonstrates excellent accuracy, robustness, and real-time performance in real road environments.