基于EMD-GRU的高速公路行程时间组合预测模型

考虑到高速公路行程时间影响因素繁多且行程时间序列非线性、非平稳特征显著，设计了基于经验模态分解和GRU神经网络的高速公路行程时间组合预测模型.首先，利用高速公路收费数据中车辆进出高速公路的时间信息获取路段行程时间序列；然后，利用经验模态分解算法，将复杂的行程时间序列分解为若干时间尺度不同、相对平稳的本征模态函数分量和残差分量；接着，使用GRU神经网络对各本征模态函数分量和残差分量进行预测与集成操作.实例分析表明:经验模态分解可有效提高LSTM、GRU神经网络的预测精度；在相同参数设置的情况下，GRU神经网络的预测精度优于LSTM神经网络.

A Combined Predicting Model for Expressway Travel Time Based on EMD-GRU

In view of the variety of influential factors on the expressway travel time and the significance of nonlinear and non-stationary characteristics of the travel time series, a combined expressway travel time predicting model was designed based on the empirical mode decomposition and the GRU neural network. First, the time information of vehicles entering and exiting the expressway in toll data was used to obtain the travel time series of the road segments; then, the empirical mode decomposition algorithm was applied to decompose the complex travel time series into a number of relatively stable and different-time-scale eigen modal function components as well as residual components; then, the GRU neural network was used to predict and integrate the intrinsic modal function components and residual components. The example analysis shows that, the empirical mode decomposition can effectively improve the prediction accuracy of the LSTM and the GRU neural networks; under the same parameter settings, the prediction accuracy of the GRU neural network is better than that of the LSTM neural network.