A NOVEL RULE-BASED TRAFFIC STATE FORECASTING APPROACH FOR LARGE SCALE ROAD NETWORKS

摘要

As collection of traffic data becomes less costly and more commonplace, large scale traffic flowforecasting is becoming increasingly needed. This paper proposed a novel rule-based traffic stateforecasting approach which is based on K-nearest neighbor (KNN) nonparametric regressionmodel, called rule-based KNN model (RKNN). Rules were extracted from the historical datausing Rough Set Theory, which assists to find the near neighbors. Traffic impact factors, such asweather and time-of-day information were incorporated automatically into the rules. Everyhistorical record is labeled with a rule. With the current traffic flow state and the traffic flowimpact data, the nearest neighbors can be found quickly from the historical data records coveredby the corresponding rule. An additional methodology was proposed to keep the historical dataand the rules up to date. A case study on an interstate freeway I-395 in Virginia is performed toevaluate the performance of RKNN. The results show that the proposed approach can decreasethe MAPE (Mean Absolute Percentage Error) by 26.86%. Moreover, the calculation time of theproposed algorithm reduced by 65.69% compared with traditional KNN algorithms, whichconsequently indicates the effectiveness of the algorithm for large urban road networks.