最后，本文提出了一种适用于不同交通状态的组合算法用于预测行程时间。通过

Finally, this paper proposes a combined algorithm suitable for different traffic conditions to predict travel time. By introducing new variables and analyzing the data, the input parameters of the machine learning algorithm can be determined. According to the predicted traffic conditions, the travel time accuracy and reliability indicators are used to select the best algorithms for different traffic conditions. In the supersaturated state, the PSO-SVR-Kalman algorithm obtains the best results. Then from a spatial perspective, the generalization ability of the algorithm is verified, and TVF-EMD is used to decompose the travel time error sequence to optimize the prediction results.

Finally, a combination algorithm for different traffic states is proposed to predict travel time. By introducing new variables and analyzing the data, the input parameters of the machine learning algorithm can be determined. According to the predicted traffic state, the algorithm suitable for different traffic states is selected by using travel time accuracy and reliability index. In the persaturation state, the PSO-SVR-Kalman algorithm produces the best results. Then from the spatial point of view, verify the generalization ability of the algorithm, and then use TVF-EMD to decompose the travel time error sequence, so as to optimize the prediction results.

最后，本文提出了一种适用于不同交通状态的组合算法用于预测行程时间。通过引入新变量以及对数据进行分析，可以确定机器学习算法的输入参数。根据预测所得的交通状态，运用行程时间精确度和可靠性指标择优选择适用于不同交通状态的算法。在过饱和状态下，PSO-SVR-Kalman算法得到的结果最好。然后从空间角度出发，验证算法的泛化能力，再运用TVF-EMD对行程时间误差序列进行分解，从而优化预测结果。<br>