Congestion Prediction Based On Classifiers Combination Technology

Congestion prediction is an important part of traffic early warning. Reliable congestion prediction has significant impact on improving level of traffic services. Therefore, it is practical and valuable to using congestion prediction model for making an exact and reliable estimation that whether it would be congested or not. At present, the approach to predict congestion is remain in the stage of predicting some traffic unilateral factors, then judging whether it would be congested or not through congestion discriminance. Some researchers have developed the traffic prediction models using neural networks. But the neural networks are instable: perturbing learning set can cause significant changes in their structure and accuracy. The instability affects the reliability of neural networks model. Further more, it is difficult to improve accuracy of model based on single network by adjusting the parameters and learning algorithm. Focus on these problem, we study on developing prediction model which use classifiers combination method and combining many traffic factors to judging whether the current state is the mode which give occasion for congestion. Finally some valuable conclusion and model have been drawn. At the same time this paper is also a beneficial trying on the application of classifiers combination technology in the traffic prediction field. Main work and results are as follows: (1) Based on domestic and international references in this field, the development of the technology correlative to congestion prediction and intelligent transport system has been introduced simply. (2) The principle and methods of classifiers combination have been analyzed. Introduce to approaches for producing component classifiers: Bagging and Boosting, which are popular with researchers. And their advantages and disadvantages have been analyzed. (3) Course of data processing has been presented detailed. We select back propagation neural networks as the meta-learning algorithm of multiple classifier system and select average-based method and voting-based method as combing rules of component classifiers. Final, the integrated congestion prediction system is expounded. (4) For the first time, holiday, period of time, exceptive occurrence and weather status have been taken into account for congestion prediction. Through a large number of computer digital simulation experiment, this paper discusses the predict model based on Bagging and AdaBoost algorithm with different parameter and combing rules. By comparison to the simulation results, some meaningful conclusion has been given. And confirming the classifiers combination method applied to congestion prediction. Final, the result of experiments shows that the performance of model based on Bagging is comparatively superior. These simulation results that have practical value of projects are the research results of this paper.