基于梯度提升回归树的城市道路行程时间预测

为了提高行程时间的预测精度，在考虑时间序列相关性的同时，分析相邻路段的空间相关性对于行程时间的影响，并提出基于梯度提升回归树模型的城市道路行程时间预测方法.对车牌识别设备获取的实际数据进行预处理，并提出相应的补全算法以解决数据缺失问题，建立完整的历史数据集.通过分析各影响因素与行程时间的相关性，构建特征向量.为了能更好地理解模型，通过梯度提升回归树模型输出各变量对于预测结果的重要度.利用实际数据对模型进行评估，预测行程时间的平均绝对误差百分比，约为10.0%.与SVM、ARIMA等方法相比，所提方法具有较高的精度.
Abstract: A new method based on gradient boosting regression tress was proposed in order to improve the prediction accuracy of travel time, considering the correlation of the time series, and also took into account spatial correlation. First, massive data collected by license plate recognition equipment was preprocessed, and the missing data problem was solved by corresponding data completion algorithm. Then the travel time historical data set was established. By analyzing the correlation between the different influence factors and travel time, the feature vector was established. Moreover, in order to better understand the model, the importance of each feature vector was proposed by gradient boosting regression tress model. Finally, the actual data was used to evaluate the model, and the average absolute error percentage of the travel time is about 10.0%. Compared with SVM, ARIMA and other methods, the proposed method has higher accuracy.