Research And Application Of Single Intersection Signal Timing Based On Support Vector Machine

In urban intelligent traffic system, signal control at junctions is an important method of traffic management of which theories and applications is constantly progressing. Traffic flow is a high dimensional, real-time, non-linear, fluctuant and random process while at the same time greatness of data, data noise and other serious problems pose great challenge to the control of traffic flow signals. To elevate the vehicular traffic rate at junctions, this paper proposes a control algorithm for signal timing based on short-term prediction of traffic flow at single-junctions.Signal control at junctions is usually based on forecasts on vehicle arrival rate, and short-term forecasts can be achieved by discovering the law of time series and constructing machine learning model. Support Vector Machine is a novel way of machine learning based on statistical learning theory and the VC dimension theory and with structural risk minimization as its principle. By converting the problem into a high dimensional feature space through nonlinear transformation and minimizing structural risk, it is capable of dealing with non-linear, high dimensional and other features of traffic flow, compensating for the lack of priori knowledge about it, improve the accuracy of short-term traffic flow prediction at junctions and ensure the efficiency of signal control system.A comprehensive research is made in this paper about key technologies in traffic signal control including traffic data acquisition and preprocessing, short-term traffic forecasts and signal timing at single junctions. Among them, the short-term traffic flow forecasting and signal timing is the most critical as they determine the feasibility of the signal control system. In this paper:researches firstly identify the traffic flow data according to probability and statistics, and then repair and normalize the traffic flow data according to time series so as to reduce the influence of problems like data noise and loss upon the model; three strategies are proposed for time series segmentation including multiple fuzzy c-means clustering, fuzzy c-means clustering with the inflection point combined with segmentation, and segmentation fusion on the basis of fuzzy c-means clustering. Correspondently, the time series of traffic flow is divided into three sections as peak, flat peak and low peak so as to reduce load and improve the operating efficiency of the model; V-support vector machine regression institutions are applied to eStablish a short-term traffic flow forecasting model, and ways of choosing parameters are proposed to construct parameter optimization algorithm according to genetic algorithm so that the precision of the model’s forecast can be improved; according to the traffic forecasting model, the signal light timing algorithm based on minimum traffic delay is designed so as to generate traffic signal timing plan and consequently reduce the average time for traffic delays in single intersection; the control system of single intersection traffic signal is designed, which includes system and structural design, function module, and network topology design.In the end of this article, the LS-SVMlab platform is used to simulate the urban expressway road traffic flow data, and experimental results is achieved to demonstrate the effectiveness of the prediction model. Moreover, the predicted results of the model is applied in signal timing program, and the specific timing data is then obtained to reveal the value of prediction models.