Research On Traffic Accident Risk Prediction Algorithm Based On Deep Learning In Edge Internet Of Vehicles

Under the background of the rapid development of modern transportation,the unbalanced development of people,vehicles and roads is becoming increasingly apparent,and the road traffic safety situation is not optimistic.For this reason,the government has taken a series of measures,such as: forbidding to change lanes at will,setting up signal lights at intersections in peak hours,etc.,but the effect of improvement is not obvious.With the development of automatic driving technology,the real-time data of map data and sensor detection combined with deep learning algorithm is applied to the unmanned vehicle,which is also convenient for the unmanned vehicle to make decisions in emergency.With the development of automatic driving technology,the real-time data of map data and sensor detection combined with deep learning algorithm is applied to the unmanned vehicle,which is also convenient for the unmanned vehicle to make decisions in emergency.This paper focuses on the traffic accident risk prediction algorithm based on deep learning in the vehicle-connected edge network.The main research contents can be generalized as follows:First of all,the traditional traffic accident risk prediction algorithm cannot automatically extract data features,and the model expression ability is poor.In this paper,a traffic accident risk prediction algorithm based on Convolutional Neural Networks(CNNs)is proposed.In this algorithm,firstly,a large number of real-time traffic data collected in VANETs are input into the convolution layers of CNN to extract multi-dimensional features.Secondly,the number of feature parameters is optimized through the pool layers of CNN.Finally,according to the evaluation index output from the full connection layer,the risk of traffic accident can be predicted by simulation.The simulation results show that the algorithm has lower prediction loss rate and higher prediction accuracy than the traditional machine learning algorithm BP neural network and logistic regression(LR).The reduction rate of loss is about 6.496% and 1.986%.The improvement rate of accuracy is about 1.220% and 2.765%.Secondly,the existing traffic accident risk prediction algorithm based on CNN has the problems of redundancy and low efficiency in full connection mode,and the deployment of server needs to be selected.In this paper,a traffic accident risk prediction algorithm based on deep convolutional random forest network in edge network of vehicles is proposed.The algorithm firstly aims at a large amount of real-time traffic data collected in edge network of vehicles,and inputting the new feature matrix of the traffic data into the convolutional layers of CNN with low structural complexity to extract multi-dimensional features.Secondly,after optimizing the number of feature parameters through pooling layers of CNN,flatten layer's output is transformed into a suitable feature vector,which can be directly input into random forest(RF)for classification prediction.Finally,according to the evaluation index of RF output,the risk of traffic accident is predicted by simulation,and the warning message is transmitted to the vehicle unit through the edge server in real time.The driver can reduce the risk of traffic accident by adjusting the driving status in time.The simulation results show that the algorithm has higher AUC,prediction accuracy and lower prediction loss rate than the traditional CNN and Ada Boost.And when dealing with massive data,the prediction performance of its model is more stable.