Implementing short-term traffic flow forecasting based on multipoint WPRA with MapReduce

摘要

The multipoint short-term traffic flow forecasting should deal with mass of historical traffic flow data from intersections in an area or urban. So the problem in runtime still remains to be the major obstacle for the practical and successful applications of the prediction algorithms. This problem becomes the key point to the evaluation of the data-driven methods especially for the nonparametric forecasting approach such as the weighted pattern recognition algorithm (WPRA). In order to solve this problem we use the MapReduce computing framework to implement the multipoint WPRA which is an improvement of the pattern recognition algorithm (PRA) based on the nonparametric regression method (NPR). By using MapReduce for the multipoint WPRA, the runtime successfully decreases, compared with using one computer.