Abstract: In view of the contradiction between the complexity and the prediction accuracy of short-term traffic flow forecasting model, an ensemble learning XGBoost (eXtreme Gradient Boosting) model was proposed to predict traffic flow, making full use of its advantages of high prediction accuracy and fast calculation speed for high-dimensional characteristic data. Firstly, the outliers of the original data were processed by median filtering. Then, a forecasting model was established based on XGBoost model, which used the method of cross-validation to determine the optimal value of the super-parameter and obtain the importance of each feature by predicting the test set. Finally, the prediction results of the model were compared with those of other short-term traffic flow prediction methods. The results show that using median filter to reduce noise and making full use of traffic flow data of adjacent sections can significantly improve the prediction accuracy of the model. The prediction accuracy of XGBoost model is 96.6%. In comparison with the other short-term traffic flow forecasting models, the proposed model can more fully utilize the temporal characteristics and spatial correlation of traffic flow.

Key words: traffic engineering; intelligent transportation; short-term traffic flow forecasting; XGBoost; spatial-temporal correlativity; ensemble learning

摘要： 针对短时交通流预测模型复杂度与预测精度的矛盾，提出基于集成学习的XGBoost（eXtreme gradient boosting）模型预测交通流，充分利用其对高维特征数据预测精度高以及计算速度快的优势。首先对原始数据的异常值进行中值滤波处理；然后基于XGBoost模型建立预测模型，利用交叉验证的方法确定最优的超参数取值，对测试集进行预测得到各个特征的重要度；最终将模型预测结果与其他短时交通流预测方法的预测结果进行比较。结果表明：中值滤波降噪处理和充分利用相邻断面的交通流数据均对模型预测精度有显著提升，XGBoost模型的预测精度高达96.6%，相比其他短时交通流预测模型更能充分利用交通流的时间特性和空间相关性。

关键词: 交通工程；智能交通；短时交通流预测；XGBoost；时空相关性；集成学习