Non-parametric Regression for Short-term Traffic Flow Forecasting Based on Cluster Analysis

Yang LIU,Shoufeng MA
DOI: https://doi.org/10.3963/j.issn.1674-4861.2013.02.007
2013-01-01
Abstract:The majority of the non-parametric regression prediction algorithms do not distinguish among the historical traffic flow data by putting all the data into one model database for analysis.In this way,all the historical flow data are regarded as a time series,and the differences among the traffic flow distribution on different dates are neglected.On the basis of the characteristics of traffic flow,this paper puts forward a non-parametric regression method for short-term traffic flow forecasting based on cluster analysis.This method first classifies traffic flow into different traffic patterns according to the characteristics of flow distribution using cluster analysis.Then,it chooses the database to match the flow of the moment to be predicted as the sample database and applies non-parametric regression into prediction.The example shows that the prediction accuracy is higher than traditional non-parametric regression methods.
What problem does this paper attempt to address?