关键词: 自适应, 局部加权学习, 交通流预测

Abstract: The authors propose an adaptive neighborhood adjusting mechanism based on certain domain knowledge, namely three phrase traffic theory. The training dataset is divided into three subsets, in which corresponding neighborhood sizes are generated. In the estimation process, the authors make predictions on data at different traffic state with respective locally weighted learning models whose kernel bandwidths are different. The final prediction is a combination of all the previous predictions. Experiments based on real traffic data prove that the proposed method can improvethe forecasting accuracy.

Key words: adaptive, locally weighted learning, traffic flow forecasting