基于XGBoost的短时交通流预测模型

为提高路段短时交通流的预测精度,选取路段平均旅行时间作为预测指标,建立了一种基于极端样度上升(extrem gradient boosting,XGBoost)的短时交通流预测模型。首先通过对交通流数据的分析,在考虑交通流时空特性的基础上,分别构建目标路段时间序列训练集、测试集以及时空序列训练集、测试集,然后基于XGBoost模型以及构建的训练样本集建立时间序列预测模型以及时空序列预测模型,并利用训练好的模型进行预测,最后将模型预测结果与线性回归模型、神经网络模型预测结果进行比较。实验结果表明:基于XGBoost的短时交通流预测模型能够对路段未来时段平均旅行时间进行比较准确的预测,其中时间序列预测模型均方根误差为5. 32,时空序列预测模型均方根误差为4. 82,均低于线性回归模型和神经网络模型,且相比于仅考虑时间因素的短时交通流预测模型,同时考虑时空因素的预测模型得到的误差更低,预测效果更好。

Research on Short-term Traffic Flow Prediction Model Based on XGBoost

In order to improve the prediction accuracy of short-term traffic flow, the average travel time of the road segment is selected as the forecasting index, and a short-term traffic flow forecasting model based on XGBoost is established. Through the analysis of traffic flow data and based on the space-time characteristics of traffic flow, the target segment time series training set, test set and spatio-temporal training set and test set are constructed respectively, and then, based on the XGBoost model and the constructed training sample set, the time series prediction model and the spatiotemporal sequence prediction model are established, and the trained model is used for prediction. Finally, the model prediction results are compared with the linear regression model and the neural network model prediction results. The experimental results show that the short-term traffic flow prediction model based on XGBoost can predict the average travel time of the road segment in the future. The root mean square error of the time series prediction model is 5.32, and the root mean square error of the space-time sequence prediction model is 4.82, both are lower than the linear regression model and the neural network model. And compared with the short-term traffic flow prediction model considering only the time factor, the model considering the space-time factor has lower error and better prediction effect.