一种基于C-V2X的BSM异常数据校正的两阶段学习策略

A two-phase learning strategy for BSM anomaly data correction basd on C-V2X

在蜂窝车联网(cellular vehicle-to-everything,C-V2X)这一通信技术框架下,车辆基本安全消息(basic safety message,BSM)的准确性对于确保道路交通安全是至关重要的。然而,BSM数据易受到传感器故障或环境干扰等非恶意因素的影响,导致数据异常并可能误导驾驶决策。针对此问题,提出了一种两阶段训练方法以校正BSM中的异常数据。第一阶段,通过无监督混合生成式模型学习正常BSM数据的行为模式与分布特征,并引入内存模块在特征空间中构建细粒度的原型存储库,增强模型对正常行为模式多样性的理解。第二阶段,基于第一阶段获得的网络参数,采用自监督学习策略进行数据校正。结果表明,该方法表现出了良好的校正能力,并显著减少了BSM数据的误差。

Under the cellular vehicle-to-everything(C-V2X)communication technology framework,the accuracy of basic safety messages(BSM)is crucial for ensuring road traffic safety.However,BSM data is susceptible to non-malicious factors such as sensor faults or environmental disturbances,leading to data anomalies that may misguide driving decisions.In response to this issue,two-phase learning strategy for correcting anomalies in BSM was proposed.In the first phase,an unsupervised hybrid generative model was used to learn the behavior patterns and distribution characteristics of normal BSM data and a memory module was introduced to construct a fine-grained prototype repository in the feature space for enhancing the model’s understanding of the diversity of normal behavior patterns.In the second phase,based on the network parameters obtained in the first phase,a self-supervised learning strategy was employed for data correction.Results show that the proposed solution exhibits good correc‐tion capability and significantly reduces the error in BSM.