Research On The Traffic Information Forecasting Methods Based On GM-LSSVM

Intelligent Transportation System (ITS) is developed nearly a dozen years which isfocused on building integrated transportation system and its success will inevitably have afundamental change in the way of the current traffic works. Currently, the complicated trafficsituation is a headache problem. In order to improve traffic conditions, improving urban roadtraffic information prediction accuracy is one of the main tasks. At the same time, improvingtraffic information prediction accuracy is an important subject in intelligent transportationsystem. However, in recent years, many scholars have focused on reliability and accuracy offorecast in traffic forecast theoretical studies.Grey model (GM) generates new data through accumulated generating operation (AGO),to some extent weakens the randomicity of the original data and makes it easier to find out thedata’s transformation rule. Therefore, the grey model is widely applied in dealing with theproblems of “small sample, poor information”. But recently, some researches show that thelarger error and poorer accuracy of the results appeared in the grey prediction method.Support Vector Machine (SVM) depends on statistical learning and it is a powerfulclassification and prediction scheme. It can achieve the best compromise between theforecasting accuracy and complexity according to its own characteristics. It can better solvethe small samples, nonlinear, high dimension, local minimum point and some other practicalproblems and well describe the nonlinear relationship between the traffic information data andthe influence factors, so it is suitable to be applied to predict traffic information. Suykens et al.proposed the Least Squares Support Vector Machine (LSSVM) on the basis of SVM,transformed the quadratic programming problem of SVM into solving the system of linearequations, reduced the complexity and improved the speed.In this paper, according to that the traffic information has the characteristics ofnonlinearity, complexity and uncertainty, in view of advantages of grey model and leastsquares support vector machine, the Grey Model and Least Square Support Vector Machine Algorithm (GM-LSSVM) is established by combining the grey model and LSSVM to obtain acombined mathematics method. Firstly the method of residual correction and backgroundvalue correction was adopted to improve grey model and cross validation method was used tofind the optimal parameters of LSSVM. Then, by using the combined forecasting model,traffic flow, speed, and occupancy from a road in Changchun city were predicted. And thesimulation analysis used the grey model and LSSVM model for comparison. Finally, thesimulation results are analyzed and summarized.The experimental results show that predicted results which are obtained by proposedmethod in this paper is very close to the actual result, the prediction accuracy is high and thecombined model is superior to the several other contrast model, so the combination forecastmethod can be widely used in related engineering field. Forecasting for accurate real-timetraffic information not only can improve the performance of the various subsystems of ITS,but also can promote the urban traffic management level and solve worsening urban trafficcongestion problem. To sum up, the study of theoretical advantage and better performanceshow that the GM-LSSVM model has good development potential and it is effective andfeasible to be applied to forecast traffic information.