Traffic Flow Time Series Periodicity Based on Recurrence Quantitative Analysis

In order to improve traffic flow prediction accuracy, recurrence plot and recurrence quantitative analysis were introduced to analyze the traffic flow time series periodicity. Further more, the prediction methods BPNN(back propagation neural network) and K-NN(nearest neighbor) were employed to predict the short-term traffic flows with different periodicity. The result of an empirical study indicates that the traffic flow time series periodicity differs with the length of statistical time interval and the time period in a day. The traffic flow time series with a statistical time interval of 5 min show a good real-time performance and a strong periodicity. The periodicity has a positive correlation with the prediction accuracy of short-term traffic flow: The traffic flow in night time has a weak periodicity for which the prediction accuracy is 87.41%, while the traffic flow in day time has a strong periodicity for which the prediction accuracy is 92.16%.