A Distributed Short-Term Traffic Forecasting System Based on Non-Parametric Regression Approach

Abstract

As a significant part of intelligent traffic systems, the short-term traffic forecasting system undertakes the important task of providing basic data for traffic control and route guidance. In view of the real-time requirements and accuracy of short-term traffic forecasting systems, this study designs a data-driven, short-term traffic forecasting system based on a non-parametric regression model. First, this paper adopts distributed architecture, which assigns forecasting tasks of different roads to several host computers. Second, a short-term traffic forecasting algorithm based on non-parametric regression is improved in terms of feature of application, promoting search efficiency, and continuously self-adaptive. Finally, according to the characteristics of different databases, it optimizes storage structure, raises input/output (I/O) efficiency, and improves the system performance. The experiment results show that this system has higher prediction speed and reliability and can handle the prediction tasks of large-scale transportation network systems. At the same time, this system can optimize and adjust itself in the process. It also has low dependence of initial setting and better adaptation.