Study On Forecasting Model And Algorithm For Urban Intersections Short-term Traffic Flow

City traffic problem has already been upgrated to the biggest constraints for thesustainable debelopment of city. Intellingent transportation system (ITS) is the importantmethod to solve this problem. The real-time accurate forecast traffic information is the basisand the key of dynamic route guidance system. However, intersection is the throat of roadnetwork traffic capacity, the locations of traffic jams and accidents, so the prection of theintersections’ traffic flow appears incereasingly important. At present, the traffic flowguidance control time span become shorter, strengthen the randomness, chaos, nonlinear anduncertainty of traffic flow. Lead to early detector intersectinons of short-term traffic flowprediction model usually cannot reflect the uncertainty and the nonlinear of traiic flow, alsocannot overcome the influence of random disturbance factors on the traffic flow, forecasteffect is not ideal; In most cities the number of detector intersection is amounted to less thanone over ten of the whole intersection, so that non detector intersection traffic flowinformation is difficult to obtain, all of these have brought greater difficulty to traffic controland guidance, also makes real-time and accurate short-term traffic flow forecasting of the twokinds of intersections is becoming more urgent.In this paper, firstly, analysed and studied the domestic and foreign scholars havepublished the current situation, the future development trend and existing problems about thetwo kinds of intersections’ short-term traffic flow forecasting. Summarized three problems tostudy and solve in the article.Secondly, using recursive figure and Lyapunov index to analyse the predictability andchaos analysis of east import of Fei Jia Ying intersection’s traffic flow data. On this basis,research the intersection of short-term traffic flow’s time and spatial dependence. First, on thetime dimension, weekly similarity of intersections’ short-term traffic flow traffic is studied byusing similar and fluctuation coefficient weeks comparability study to determine the workingdays, rest days, the weather (such as sunny day, rainy days) time factors’ the important role atthe process of traffic flow prediction. Second, on the space dimension, the flow’s mutualinfluence among import and export of intersection prediction section, around the intersectionand road are determined, and adjacency space dependence method and the adjacent sectionsare quantitatived to get the influence extent of space. Last, through the above study wasproposed based on multidimensional space and time parameters of short-term traffic flowprediction model and framework, lay a solid foundation for the detector intersection ofshort-term traffic flow prediction.Then, in view of the detector intersection traffic flow prediction from collocationpatterns of the combination model and the selection of the single model’s weight parameters was improved and optimized. According to the advantages and disadvantages of the singlemodel, selecting three architectural modules and transform it to be able to usemultidimensional space and time factors; Due to the prediction error is the random error,based on the feedback mechanism of the delta variable weighting method is presented by useof good properties of normal distribution, namely using each sub module’s prediction errorweighted average method.Higher value weight can be given greater for the higher precision.According to the feedback about the traffic state’s changing the error relationship at eachperiod the relationship between error and then changing traffic state feedback, thus weight canbe adjusted timely updates, won’t cause too much prediction deviation, so as to establish thecombination forecast model based on the spatial and temporal correlation status, and in part ofintersections of An-Ning region road network as an example, verify the feasibility of themodel and algorithm.Finally, two kinds of thoughts for non detector intersections’ traffic flow prediction areanalysed, this article from the latter perspective, set up with the connection between nondetector intersections with detector intersections, while using detector intersections traffic topredict. First, this paper introduced the commonly used several kind of classification methodand the concept of probability neural network PNN based on bayes minimum risk criteria.And applied them for the first time to classify non detector intersections and detectorintersections, forecasting method based on the probability of PNN neural network intersectionclassification model is proposed. Second, introduced prediction method after classifition,which has a linear regression and nonlinear fitting and raised two classes of nonlinear fittingof the BP neural network and that is optimized by genetic-GA. Set up a dynamic contactbetween detector with non detector intersections. Non detector intersections realizeclassification of time and space. While short-term traffic flow’s prediction is also realized,and in part of intersections of An-Ning region road network as an example, verify thefeasibility of the model and algorithm.