Research on Traffic Accident Prediction Model Based on Convolutional Neural Networks in VANET

Aiming at the problem that the traditional machine learning traffic accident prediction algorithm can't automatically discriminate the data features, it requires a lot of artificial feature engineering prediction models and the machine learning algorithm has poor expression ability. The paper proposes a traffic accident prediction algorithm based on Convolutional Neural Networks (CNNs). The deep learning algorithm can extract autonomous features from a large amount of data collected in Vehicular Ad-hoc Network (VANET). Different convolution kernels are responsible for extracting different features and the obtained new variables are input into the established training model in the edge computing server for training and testing. Simulating and predicting the possibility of traffic accidents and then formulate traffic accident risk warning standards, and transmit alarm messages to vehicle units in real time. The driver can avoid the traffic accident by adjusting the driving state in time. The simulation results show that the prediction model is validated to predict the possibility of vehicle accidents. Compared with the traditional Back Propagation (BP) Neural Network, CNNs has lower loss and higher prediction accuracy, which not only provides a theoretical basis for vehicle safety assisted driving, but also provides guidance for anti-collision and optimization of intelligent vehicle path planning.