The detection algorithm of anomalous traffic congestion based on massive historical data

Abstract

Traffic congestion can be classified into recurrent congestion and anomalous congestion. Anomalous congestion is probably caused by emergencies, which refer to events that go against historical normal states. Therefore, anomalous congestion can be detected by analyzing the differences between massive historical traffic data and real-time data. It is of great significance to actively detect and quickly deal with anomalous congestion. However, processing massive historical data in standalone mode cannot complete the whole process within the tolerable time. In view of existing problems in the detection of anomalous congestion, a distributed mining algorithm was proposed. The real-time data perceived by microwave radar and massive historical data sets were adopted to define the quantitative expression of traffic abnormal degree, meanwhile consideration was given to technical links such as data cleaning, cumulative effect of time and update of historical data. Results of the experiment show that the proposed algorithm has satisfactory identification effects.