Edge Intelligent Deployment Optimization Based On Traffic Flow Prediction

With the widespread application of 5G technology,and the rapid development of edge computing and on-vehicle applications,Internet of vehicles（Io V）has become one of the most promising scenarios at present.However,the rapid growth of on-vehicle applications has brought great pressure to the edge network,and the mobility of vehicles has exacerbated the complexity of resource allocation in the edge network.Therefore,how to allocate realtime and efficient resources in the Internet of vehicles is the key to improve the application environment of the Internet of vehicles.This paper proposes a solution to this problem called “EICF-TFP”（Edge Intelligent Coordination Framework Based on Traffic Flow Prediction）.This scheme deploys artificial intelligence algorithms in edge nodes and uses edge computing to predict and allocate resources in the Internet of vehicles.Firstly,a traffic flow prediction algorithm based on deep learning is proposed.The algorithm uses the grey relational analysis method to analyze the correlation between lanes,and selects the traffic flow data of the remaining lanes that have a greater correlation with the predicted lane as input,and then uses the Convolution-Gated Recurrent Unit（ConvGRU）extracts the spatiotemporal features of the traffic flow and realizes fast and accurate prediction of the traffic flow.In addition,the relationship between lane traffic flow prediction,vehicle resource requests prediction and vehicle mobility prediction is explored.It is proposed to predict vehicle resource requests and vehicle mobility by predicting the traffic flow in the lane.The results of numerous comparative experiments on the real traffic flow dataset in Qingdao show that the prediction model proposed in this topic has smaller error and faster convergence speed than traditional deep learning models,and can accurately predict the traffic flow in a short time.Secondly,using the correlation between lanes,a migration algorithm of associated lane model is proposed to migrate and deploy the deep learning model,to reduce the number of training deep learning models and reduce resource consumption.Experiments show that the model migration algorithm can save about 60% ～ 70% of the number of model training and greatly reduce the resource consumption.Then,based on the prediction,an adaptive time batching and matching algorithm is designed by using deep reinforcement learning.The algorithm can dynamically adjust the frequency of resource allocation by edge nodes and improve the success rate of resource allocation.The experimental simulation shows that compared with the fixed frequency resource allocation of edge nodes,the adaptive batching and matching algorithm can increase the revenue of edge nodes by at least 50%.