Study On The Multi-steps Prediction Method Of Short-time Traffic Parameters

With the rapid development of urbanization, and the sharp increase of carownership, the city's traffic congestion is more and more serious. It also increases thatenvironmental pollution, energy waste, traffic accidents, increase in travel time, andthe decline in environmental quality, which caused by traffic congestion. Thoseseriously impact on the sustainable development of cities and people's normal life andwork.Dynamic traffic management is an effective measure to solve the increasinglyserious traffic congestion problems, which is based on dynamic traffic data. Thedynamic traffic data include the current data of the short-term traffic parameters andforecast data of the future period of time. The predicted effect of short-term trafficparameter data has an important influence on the predictability and reliability ofdynamic traffic management.In order to further improve the predictability of dynamic traffic management, inthis paper, we studied the multi-step prediction methods for short-term trafficparameters basing on measured data obtained by the vehicle detector, the researchachievements mainly include the following three aspects.Firstly, we proposed a multi-step prediction method based on multiple timescales step extrapolation for short-term traffic parameters.Based on the existing step extrapolation method for multi-step prediction andcombination forecasting method, and according to the mechanism of synthesis anddecomposition of different time scales short-term traffic parameters, we design amulti-step prediction method based on MTS step extrapolation for short-term trafficparameters, and using the measured data of the induction coil in a large urban freewayto verify and comparative analysis the algorithm. The results show that the proposedmulti-step prediction method change the previous thinking way of the multi-stepprediction of the short-term traffic parameters, which do not need to make the steppredictive value as the input to the next step prediction, so it not only reduce the erroraccumulation problem cased by the cycle step extrapolation, but also can real-timetracking the prediction error, which make the combination of model have a goodresult.Secondly, we designed a multi-step prediction method for short-term trafficparameters based on k-nearest neighbor estimate.Against to limited real-time online applications of the multi-step predictionmethod based on the multi-time scales step extrapolation, depending on the method ofk-nearest neighbor estimate, we designed a multi-step prediction method. Over mode definition,mode matching and mode estimation, multi-step prediction of short-termtraffic parameters can be a one-time. Based on the measured data of induction coil ofa large urban freeway, the results of comparative analysis show that this method betterthan the comparison method in predicting efficiency and predicting effectiveness.Finally, a dynamic predictability analysis for short-term traffic data serials basedon BP neural network was designed.The inherent volatility of short term traffic flow characteristics, making thesequence of short-time traffic parameter data have different predictability at differenttime point, which manifest in different predictable steps. In order to improve thedynamic predictability analysis results of short-term traffic, the characteristicsassociated indexes of traffic data serials predictability is designed, and DPT isestablished based on BP neural networks, and the performance of the method wasverified by the measured data of a large urban freeway induction coil.It is a useful exploration of the multi-step prediction method that the multi-stepprediction methods and experimental findings in this paper,and it opened up newideas for the multi-step prediction of short-term traffic parameters. At the same time,it provides more valuable information infrastructure and related technical support formore effective implementation of dynamic traffic management.