Scholarlink

<p>To enhance the accuracy of vehicle trajectory prediction and driving safety while accounting for the impact of driver-specific behaviors, this study proposes a personalized vehicle trajectory prediction method (DS-TCTM) based on driving style classification. First, driving style features are extracted through the calculation of acceleration change rate and average time headway. Subsequently, traffic flow density along vehicle routes is classified using the K-Means + + algorithm, and these features are integrated into a K-nearest neighbor-enhanced Gaussian Mixture Model (K-GMM) for comprehensive analysis, resulting in three distinct driver style categories. Finally, the identified driving style types are incorporated into a multi-level personalized trajectory prediction network architecture (TCTM) to achieve precise trajectory forecasting. Experimental results demonstrate that DS-TCTM achieves average root mean square error (RMSE) and negative log-likelihood (NLL) values below 4.46&#xa0;m and 3.89&#xa0;m, respectively, across varying prediction horizons. The model attains optimal performance after over 100 hyperparameter optimization iterations. Compared to baseline methods, DS-TCTM reduces prediction errors by 35.8%, with particularly notable accuracy improvements observed in long-term trajectory prediction scenarios. These findings indicate that DS-TCTM effectively characterizes driving style influences on trajectory patterns, significantly enhances prediction precision, and provides critical data support for vehicle collision warning systems.</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

Personalized vehicle trajectory prediction method based on driving style classification

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月27日

查看全文
  • Xiang LI,
  • Mingbao Zhang,
  • Adriano Jose Tavares,
  • Hao Xu,
  • Shijun Ma

摘要

To enhance the accuracy of vehicle trajectory prediction and driving safety while accounting for the impact of driver-specific behaviors, this study proposes a personalized vehicle trajectory prediction method (DS-TCTM) based on driving style classification. First, driving style features are extracted through the calculation of acceleration change rate and average time headway. Subsequently, traffic flow density along vehicle routes is classified using the K-Means + + algorithm, and these features are integrated into a K-nearest neighbor-enhanced Gaussian Mixture Model (K-GMM) for comprehensive analysis, resulting in three distinct driver style categories. Finally, the identified driving style types are incorporated into a multi-level personalized trajectory prediction network architecture (TCTM) to achieve precise trajectory forecasting. Experimental results demonstrate that DS-TCTM achieves average root mean square error (RMSE) and negative log-likelihood (NLL) values below 4.46 m and 3.89 m, respectively, across varying prediction horizons. The model attains optimal performance after over 100 hyperparameter optimization iterations. Compared to baseline methods, DS-TCTM reduces prediction errors by 35.8%, with particularly notable accuracy improvements observed in long-term trajectory prediction scenarios. These findings indicate that DS-TCTM effectively characterizes driving style influences on trajectory patterns, significantly enhances prediction precision, and provides critical data support for vehicle collision warning systems.