一种基于车辆轨迹数据的深度学习异常检测方法

Deep learning anomaly detection method based on vehicle trajectory data

随着定位技术和基础设施的不断发展,车辆轨迹数据的采集变得更加容易,进而为异常轨迹的准确检测提供了更多的数据资源,成为轨迹数据挖掘中的重要研究方向之一。文章提出了一种基于深度学习的异常车辆轨迹检测方法,通过结合VAE,GRU和WGAN,实现了特征提取和轨迹异常检测的任务,同时构建了基于VAE结构的GRU特征学习模型,学习初始数据的近似分布,创新性地提出GRUWGAN模型,可实现轨迹数据异常特征提取,并对真实数据进行异常检测。实验结果表明,GRUWGAN模型在准确率、召回率和F1指标上均优于传统算法,可以有效地应用于车辆轨迹数据的特征提取和异常检测任务中。

With the continuous development of positioning technology and infrastructure,the collection of vehicle trajectory data has become easier,which in turn provides more data resources for the accurate detection of abnormal trajectories,and has become one of the important research directions in trajectory data mining.This paper proposes a deep learning-based abnormal vehicle trajectory detection method.By combining VAE,GRU and WGAN,the tasks of feature extraction and trajectory anomaly detection are realized.At the same time,a GRU feature learning model based on VAE structure is constructed to learn the approximate distribution of initial data.The GRUWGAN model is innovatively proposed,which can realize abnormal feature extraction of trajectory data and anomaly detection of real data.The experimental results show that the GRUWGAN model is superior to traditional algorithms in accuracy,recall and F1 indicators,and can be effectively applied to feature extraction and anomaly detection tasks of vehicle trajectory data.