Abstract: Short-term passenger flow forecast between stations is an important basis for high-speed railway operation and management. Firstly, the samples and tag sets were obtained based on the extraction of the characteristics of the original passenger flow data; and then, the parameters of the neural network model were pre-trained based on the stacked auto-encoder algorithm; finally, the neural network prediction model was constructed. Taking Yuwan high-speed railway as an example, the data from November 2016 to October 2018 were used for verification. The results show that the prediction error of the proposed model is 12.08%, which improves the accuracy by 12.12%, 1.12%, 6.9% and 19.12% respectively, compared with the other four commonly used prediction models. The proposed model is suitable for short-term passenger flow prediction.

Key words: traffic and transportation engineering, high-speed railway, passenger flow forecast, feature extraction, stacked auto-encoder, neural network

摘要： 站间短期客流预测是高速铁路运营管理的重要依据。首先在提取原始客流数据特征的基础上得到样本和标签集，然后基于栈式自编码算法预训练神经网络模型参数，最后构建神经网络预测模型。以渝万高铁为例，采用2016年11月到2018年10月数据进行验证，结果表明:提出的模型预测误差为12.08%，与其它4种常用预测模型相比精度分别提高12.12%、1.12%、6.9%和19.12%，模型适用于短期客流预测。

关键词: 交通运输工程, 高速铁路, 客流预测, 特征提取, 栈式自编码, 神经网络