ABSTRACT: Short-term traffic flow prediction based on multi-model combination under Stacking framework is proposed in this paper, associated with frontier theory research of artificial intelligence. Firstly, the mechanism of Stacking ensemble learning is introduced. The XGBoost algorithm constructed by tree model and the deep learning algorithm represented by long and short memory network (LSTM) are presented. Then considering the difference of training principles and the correlation coefficient of models, the Stacking based traffic forecasting model embedded various machine learning algorithms is proposed to utilize their diversified strength. In this Stacking framework, the XGBoost algorithm, the LSTM algorithm, the Support vector machine and k-Nearest Neighbor are chose as base-learner, and the XGBoost algorithm is chose as meta-learner. Finally, the effectiveness of the algorithm is verified by actual traffic flow data. The use case shows that the forecasting results are more accurate when each of base-learner has lower correlation coefficient in Stacking. The results indicated the Stacking ensemble learning based on multi-model has better prediction performance compared with the traditional single model. Key words: traffic flow prediction; multi-model combination; Stacking ensemble learning; machine learning Author: CHEN Yong Affiliation: Key Laboratory of Road and Traffic Engineering of Ministry of Education, Tongji University, No.4800 Cao’an Road, Shanghai 201804, China

CICTP