Study On Method Of Signal Control Time Division Based On Intersection Flow Prediction Data

With the acceleration of the urbanization process,the consumption level of residents in China is increasing day by day.The car is not a luxury,but a daily means of transportation.City is an important part of road network,road intersection is the bottleneck part of the urban road traffic,most of the traffic jam phenomenon also occurred near the intersection,the control of the division of time is directly related to the relevant line of the play,and even the entire road network traffic function for mobility and traffic capacity of urban road,the road network capacity and traffic safety have great influence.Therefore,practice has proved that adjusting the timing scheme and channelization design of existing intersections is the fastest.This paper has carried out an in-depth study on how to optimize and improve the signal control period of intersections based on the existing signal control facilities and effectively alleviate the congestion at intersections.In this paper,a signal control time division model is established,and the following work is mainly carried out:1)In order to improve the prediction accuracy of short-term traffic flow at intersections,a short-term traffic flow prediction model based on GRU-RNN that was proposed based on historical traffic flow data under the theoretical framework of deep learning.First USES the random forest algorithm of traffic flow data preprocessing,repair cycle missing cases of loss of data,to ensure the integrity of the sample and accuracy,secondly the GRU-RNN helped algorithm to forecast the short-term traffic,adjusting parameters in Myeclipse development environment to adapt to the samples,by controlling the control model of memory ability with the update threshold layer,after constant iterative,memory and the characteristics of the historical data information updates,historical data will be given different weight value,has been training model will butt down predict data for model training and validation,And compared with the classical BP neural network prediction model.2)This paper was proposed based on improved SOM algorithm of signal control time division method,the predicted flow as the data input,do initial clustering with SOM first,again with PAM algorithm of quadratic clustering optimization clustering center,set different clustering number,as the use of Synchro simulation software to simulate the real environment,with the total number of parking and delay as the evaluation index,the contrast analysis of the two kinds of clustering algorithms under different clustering number period of validity of the results.