融合VMD和GRU模型的城市道路行程时间预测方法

城市道路交通环境复杂多变，城市道路行程时间具有较强的非线性与非稳定性，为提高城市道路行程时间的预测精度，提出了基于变分模态分解(variational mode decomposition，VMD)与门控循环单元(gated recurrent unit，GRU)相结合的组合预测模型。与传统分解算法相比，VMD拥有非递归求解和自主选择模态个数的优点。首先利用变分模态分解算法将原始行程时间序列分解为若干时间子序列，降低原始序列的非平稳性；然后对每个时间子序列建立GRU预测模型；最后将各个预测结果进行融合，得到行程时间序列预测的最终结果。实验结果表明，变分模态分解与门控循环单元结合的组合模型预测结果要比对照组的单一模型预测结果精准度高，均方根误差(root mean squared Error，RMSE)及下降约3.99～4.37，平均绝对误差(mean absolute error，MAE)下降约3.02～3.35；在组合预测模型中，门控循环单元(GRU)预测效果要比长短期记忆(long short-term memory，LSTM)预测效果表现更佳，均方根误差(root mean squared error，RMSE)下降0.34，平均绝对误差(mean absolute error，MAE)下降0.22。

An urban road travel time prediction method based on VMD and GRU model

The urban road traffic environment is complex and changeable, and the travel time of urban roads has strong nonlinearity and instability. In order to improve the prediction accuracy of urban road travel time, a combined prediction model based on the combination of variational mode decomposition (VMD) and gated recurrent unit (GRU) is proposed. Compared with the traditional decomposition algorithm, VMD has the advantages of non-recursive solution and autonomous selection of the number of modes. Firstly, the original travel time series is decomposed into several time sub series using the variational modal decomposition algorithm to reduce the non-stationary of the original series; Then the GRU prediction model is established for each time subsequence; Finally, each prediction result is fused to obtain the final result of travel time series prediction. The experimental results show that the prediction result of the combined model of the variational mode decomposition and the gated cycle unit is more accurate than that of the single model in the control group, with the root mean square error (RMSE) and mean absolute error (MAE) decreasing by 3.99~4.37 and 3.02~3.35 respectively; In the combined prediction model, the prediction effect of gated recurrent unit (GRU) is better than that of long short term memory (LSTM), with the root mean square error (RMSE) decreasing by 0.34 and the mean absolute error (MAE) decreasing by 0.22.