Short Term Traffic Volume Prediction Based On Markov Model

Short term traffic flow prediction plays a significant role in urban traffic volume control andguidance system. The quality of prediction influences the effect of urban traffic flow control andguidance system directly. Hence, short term traffic flow prediction is very essential to intelligenttransportation system (ITS).This paper, aims at improving the accuracy of the prediction scheme, builds the Markov modelfirst. Considering the non-linear and non-stationary characteristic of the traffic data time series, thispaper proposes Markov-BP neural network model and Wavelet-Markov-BP neural network model,respectively.The main contents of this paper list as below:1. Since the data gained from inductive loop detector has errors, redundancy and other qualityproblems, this paper utilizes outlier mining technology to detect abnormal data and makes use of‘adjacent time period data average’ method to repair abnormal data, in order to resolve data qualityproblems. Then, an improved wavelet denoising method is proposed. The traffic flow data afternoise reduction processing can reflects the characteristic of the traffic flow better.2. Considering the non-linear characteristic of the traffic data time series, this paper proposesMarkov-BP neural network compositional model. By using the error correction thought anddynamic rolling prediction scheme, this compositional model can achieve better results incomparison with Markov model.3. Considering the non-stationary of the traffic data time series, this paper proposesWavelet-Markov-BP neural network compositional model (W-M-BPNN). It uses multi-resolutionanalysis to decompose the raw traffic data time series. Therefore, the time series will bedecomposed on different scales and those decomposed time series can present more properties oftraffic flow. The experiment results show that the three models can achieve acceptable accuracy, bycontrast, the W-M-BPNN model is superior to the other two models.