An Algorithm for Traffic Flow Prediction Based on Improved SARIMA and GA: Fid-Bau Portal

An Algorithm for Traffic Flow Prediction Based on Improved SARIMA and GA

Luo, Xianglong / Niu, Liyao / Zhang, Shengrui

Abstract The traffic flow prediction plays a key role in modern Intelligent Transportation Systems (ITS). Although great achievements have been made in traffic flow prediction, it is still a challenge to improve the prediction accuracy and reduce the operation time simultaneously. In this paper, we proposed a hybrid prediction methodology combined with improved seasonal autoregressive integrated moving average (ISARIMA) model and multi-input autoregressive (AR) model by genetic algorithm (GA) optimization. Since traffic flow data has strong spatio-temporal correlation with neighboring stations, GA is used to select those stations which are highly correlated with the prediction station. The ISARIMA model is used to predict the traffic flow in test station at first. A multiinput AR model with traffic flow data in optimal selected stations is built to predict the traffic flow in test station as well. The final prediction result can be gained by combining with the results of ISARIMA and multi-input AR model. The test results from traffic data provided by TDRL at UMD Data Center demonstrate that proposed algorithm has almost the same prediction accuracy with artificial neural networks (ANNS). However, its operation time is almost the same with SARIMA model. It is proved to be an effective method to perform traffic flow prediction.