Design And Implementation Of Short-term Traffic Flow Prediction Algorithm Based On Cloud Platform

Short-term traffic flow prediction as an important component of intelligent traffic system is always one of the research hotspots. The prediction accuracy and calculation efficiency are the two key factors that influence the traffic control, guide and management effect. However, high prediction accuracy and excellent calculation efficiency are contradictory which cannot be achieved simultaneously. In general, high prediction accuracy algorithms usually have more complicate logic, or have to process large scale of data., which results in low efficiency and even influence practicality. In order to optimize the practicality of short-term traffic flow prediction, this paper combines the advantage of cloud platform in big data storage and large scale of data real-time processing, and proposed a MapReduce-based short-term traffic flow prediction method. The MapReduce-based method can effectively better the calculation efficiency while prediction accuracy is not lowered.This paper presents the design and implementation of MapReduce-based short-term traffic flow prediction method. Two popular prediction accuracy algorithms are selected and improved to adjust to the requirements of MapReduce programming model:one is GA-BP (Generic algorithm optimized Back-Propagation Neural Network) algorithm which has fairly prediction accuracy but demand large scale of data processing in the early network weights training stage and time-consuming; the other is K-nearest neighbor method which has very high prediction accuracy but has to search the whole history database and has low calculation efficiency.Simulation numeric results show that the MapReduce-based short-term traffic flow prediction algorithm can solve the problem of GA-BP’s time-consuming early network weights training as well as the problem of low calculation efficiency history database searching. Therefore, the MapReduce-based short-term traffic flow prediction method can effectively optimize the algorithm practicality.