Research On Short-term Traffic Flow Forecasting And Signal Control In Urban Arterial Road

Complexity,randomness and chaos are three important characteristics of urban traffic system which is a huge system.In the process of coordinating arterial traffic signals,the frequent change of coordinated timing scheme will cause the disorder of traffic flow.Turbulence will seriously affect the normal operation of vehicles,greatly reducing the operational efficiency.In this case,the economic benefits of changing the coordinated timing scheme of the vehicle are not sufficient to compensate for the economic losses due to turbulence.At the same time,the long optimization time interval cannot reflect the real-time performance of the optimization of traffic signal timing parameters.In view of this characteristic of the urban transportation system,the method of changing the timing scheme every 10-15 minutes is usually adopted.In order to ensure the accuracy of the prediction,the upper limit of the time interval used for short-term traffic flow prediction is usually set to 15 minutes.Therefore,on the basis of fully considering the nonlinear characteristics of the traffic system,this paper uses genetic algorithms to optimize the SVM model and uses the Long ShortTerm Memory(LSTM)as short-term prediction model of road traffic flow data.On this basis,the idea of comprehensive research is introduced,combined with Bayesian theory,and the prediction results are integrated.Take the predicted traffic volume of the arterial road within 15 minutes as input,adjust the phase difference and green signal ratio of the intersection,and combine the real-time traffic flow detection data to optimize and update the signal timing diagram during this period.So as to realize the collaborative optimization of traffic flow prediction and traffic signal.Firstly,the research on traffic flow prediction and signal control at home and abroad is summarized and reviewed,and the research content and structure of this paper are clarified.Preprocessing the ultrasound data to solve data quality problems caused by data loss,anomalies and errors.At the same time,the historical data of each intersection of trunk roads were collected at 15 min intervals.Secondly,a Bayesian combination prediction model based on probabilistic motivation was proposed,and the GA-SVM and the LSTM models were trained as the base predictors in the Bayesian fusion prediction model.Bayesian posterior probability was used to represent the weight of the base predictor,and the weighted combination of the base predictor was generated.On the selection of training data time range at the same time,the grey theory is introduced in the grey correlation analysis method based on entropy,time selection and prediction of traffic high correlation between the time interval as sample training set,so that they can more quickly adjust the weights of local predictor,the prediction model of MAE fell by 13.5%,MAPE fell by 14.4,RMSE fell by 48.6%.Finally,in order to forecast traffic volume at 15 minutes interval as input,with the help of a multi-objective optimization algorithm,NSGA-III intersects the phase difference of the backbone in different states,and the green letter ratio needs to adjust the signal control signal,such as in the period of the signal timing scheme optimization and update,so as to realize the collaborative optimization of prediction data based on the main signal.Taking the main road of Mingguang Road as an example,which situates in Weiyang District of Xi ’an province,VISSIM was used to simulate the plate and peak periods of the main road,and the comparison with Webster timing method was made from multiple performance indicators.The simulation results show that the traffic delays on the main road are reduced by 23.6%,the number of stops is reduced by 14.5%,the average travel speed is increased by 21.1%,while the traffic running state of the main road is improved.