基于深度学习的车联边缘网络交通事故风险预测算法研究

Research on Traffic Accident Risk Prediction Algorithm of Edge Internet of Vehicles Based on Deep Learning

针对传统交通事故风险预测算法无法自动判别数据特征,且模型表达能力差等问题。该文提出一种基于深度学习的车联边缘网络交通事故风险预测算法,该算法首先针对车载自组织网络中采集的大量交通数据,采用边缘服务器中建立的卷积神经网络自主提取多维特征,经归一化、去均值等预处理后,再将得到的新变量输入卷积层、采样层进行训练,最后根据全连接层输出的判别值,得到模拟预测交通事故发生的风险性。仿真结果表明,该算法被验证能够预测交通事故发生的风险性,较传统的机器学习算法BP神经网络、逻辑回归具有更低的损失与更高的预测准确度。

For the problem that the traditional traffic accident risk prediction algorithm can not automatically discriminate data features,and the model expression ability is poor,a traffic accident risk prediction algorithm based on deep learning is proposed.The algorithm firstly extracts multi-dimensional features by using the convolutional neural network established in the edge server for a large amount of traffic data collected in the edge network of vehicles.After normalization,de-equalization and other pre-processing,the new variables are input into the convolutional layer and the pooling layer for training.Finally,based on the output discrimination value of the fully connected layer,the risk of traffic accidents can be predicted by simulation.The simulation results show that the algorithm is validated to predict the risk of traffic accidents,and has lower loss and higher prediction accuracy than the traditional machine learning BP neural network algorithm and Logical Regression algorithm.