使用AVL数据的公交到站时间CEEMD-LSTM预测模型

准确的公交到站时间预测具有重要意义，但现实公交运行受突发路况影响，运行速度具有非平稳性，本文结合时序特征处理技术和深度学习，建立一种使用AVL数据预测公交到站时间的互补集合经验模态分解-长短期记忆神经网络模型。模型收集公交自动车辆定位数据，经预处理后引入互补集合经验模态分解平稳化公交运行速度，再借助Adam参数寻优后的长短期记忆神经网络对福州市303路公交某日早高峰公交到站时间进行预测。结果表明：优化的公交到站时间预测模型平均绝对误差比单一模型低了1.69min，预测精度高于长短期记忆神经网络模型和经验模态分解的到站时间预测模型，可有效地为安装车载自动车辆定位系统的公交线路预测公交到站时间提供参考。

A CEEMD-LSTM prediction model for bus arrival time used AVL data

Accurate bus arrival time prediction is important, but actual bus operation is affected by sudden road conditions. In view of the non-stationary running speed of bus, this paper proposed an optimized CEEMD-LSTM model for bus arrival time prediction combined with time series feature processing technology and deep learning by AVL data. The model collects bus AVL data, then bus running speed was stabilized by CEEMD after preprocessing. Next, with the help of Adam-LSTM predicted the bus arrival time of Fuzhou 303 bus in morning peak of one day. Experimental results show that the MAE of optimized LSTM neural network model was lower than single LSTM model,1.69 minutes. It can be concluded that CEEMD optimized improves the prediction accuracy and can provide a reference for bus arrival time prediction.