Research On Urban Road Condition Analysis And Prediction Method Based On Track Data

At present,China is in the stage of rapid urbanization.The urbanization process is speeding up with the rapid growth of motor vehicles.Serious traffic jams generally occur in major cities.Traffic congestion leads to the increase of traffic delays,the decrease of vehicle speed and the waste of time for people.Traffic congestion has become a bottleneck restricting China's economic and social development.The key to alleviate traffic congestion is to predict the traffic jam and use the predicted result to plan the travel route reasonably.Considering the problem of urban road traffic congestion,a new model,MIXDL model(Mixed Deep Learning Model),was proposed in order to improved the traffic.The model based on Deep Learning and Ensemble Learning used four basic learner with different parameter which based on DNN and LSTM.Firstly,this paper preprocesses a large number of Beijing taxi track data,including data cleaning,feature value selection and specific calculation of traffic flow parameters such as traffic volume and average speed,so as to extract the traffic flow data of the early peak.By referring to the calculation standard of congestion assessment provided by Beijing transportation department,the formula for calculating the traffic congestion index is defined,and a traffic congestion assessment model is established to evaluate the congestion of selected sections in the morning rush hour.Secondly,after calculating the congestion index,in order to get a better training prediction model,this paper selects the characteristic variables that may affect traffic congestion from three aspects,and combines the POI number,residential area,subway entrance and other basic data near the road network with the track data through map matching method,and constructs the traffic flow characteristic vector as the input of the prediction model.Finally,this paper compares the proposed traffic congestion prediction model and other common traffic congestion prediction models,compares the experimental results with the real congestion index,and calculates MAE value.Experimental results show that,with the same epochs and batch-size,the traffic congestion prediction model has better prediction accuracy than other traffic congestion prediction models,and the average absolute error between the predicted value and the actual value is maintained at about 7.5.According to the road classification method of traffic congestion index in the traffic congestion evaluation model proposed in chapter 3,the predicted value and the actual value are divided and compared.The results show that the prediction accuracy of traffic jam grade is more than 78%.