To improve the energy efficiency and reduce emissions of motor vehicles, this study tests and compares five machine learning algorithms in conjunction with three sets of feature indicators to establish an assessment model for the ecological nature of lane-changing behavior. The model combining the Extreme Gradient Boosting (XGBoost) algorithm and the Trend Feature Symbolic Aggregate Approximation (TFSAX) feature metrics set performs well. The effectiveness of the TFSAX feature metrics set in capturing factors influencing vehicle fuel consumption and driving behavior sequence features was also verified. Furthermore, it was concluded that the specific value of pedal pressing depth is not the primary factor contributing to differences in fuel consumption levels; rather, the magnitude of its trend largely determines fuel consumption levels. Therefore, the model we have developed has important applications in assessing the ecological aspects of lane-changing behavior on urban roads.

中文翻译：

---

变道行为中的生态驾驶策略使用：驾驶员如何降低油耗？


为了提高机动车的能源效率和减少排放，本研究结合三组特征指标对五种机器学习算法进行了测试和比较，以建立变道行为生态本质的评估模型。结合了极端梯度提升 （XGBoost） 算法和趋势特征符号聚合近似 （TFSAX） 特征指标集的模型表现良好。还验证了 TFSAX 特征指标集在捕获影响车辆油耗和驾驶行为序列特征的因素方面的有效性。此外，得出的结论是，踏板踩下深度的具体值并不是导致油耗水平差异的主要因素;相反，其趋势的幅度在很大程度上决定了油耗水平。因此，我们开发的模型在评估城市道路变道行为的生态方面具有重要的应用。