Multiple-clustering ARMAX-based predictor and its application to freeway traffic flow prediction

摘要: An adaptive predictor for a linear discrete time- varying stochastic system is proposed in this paper order to forecast freeway traffic flow at specific location over one- hour horizon. Historical sensor data first clustered by the K-means method obtain representative pattern of sensor. For each cluster and using clusters centroid as exogenous input, time-varying output subsequently modeled an ARMAX process, identified real time recursive least squares (RLS) with forgetting factor algorithm. Based on model, D-step ahead optimal generated its associated estimated error prediction variance calculated. The estimate that produces smallest selected sampling instant generate output. technique applied empirical vehicle detector station (VDS) both mainline on-ramp locations horizon one hour. Results indicate often offers superior flexibility overall performance compared either only historical or real-time normal commute days when unusual incidents occur. I. INTRODUCTION Stochastic statistical learning techniques are gaining increased attention managing large complex systems. One such network equipped inductive loop detectors. These detectors provide data, which gathered management centers (TMC) control traffic. main challenge TMC reduce congestion, since it leads wasted time, air pollution, waste gasoline, reduced driver safety. Congestion can be recurrent non-recurrent. Recurrent congestion occurs regularly during rush hours, peak morning evening some instances both. Non-recurrent caused special events, road accidents. To TMCs rely accurate information near-future predictions evaluate potential strategies. A Decision Support System (DSS) currently being developed Connected Corridors program University California PATH program. With use