基于深度学习算法的城市轨道交通客流短时预测

Short-term Passenger Flow Prediction of Urban Rail Transit Based on Deep Learning Algorithm

文中基于某地铁自动售检票数据,分析得到城市轨道交通短时客流时序数据具有非平稳性、非线性和非周期性波动等特征.借助深度学习算法,分别构建了基于LM-BP和LSTM神经网络的短时客流预测模型.选取苏州地铁中央花园站为例,以5 min为时间段,采用某一周的客流时序数据作为训练数据,预测未来一天内短时客流量的变化趋势,并对所提预测方法进行验证.结果表明:相较于参数模型中常用的三次指数平滑模型,所提方法的预测准确性和稳定性改进了50%左右,预测精度得到了大幅提升,LSTM模型具有更好的预测精度,而LM-BP模型则在计算效率方面更具优势.

Based on the AFC data of a subway, the short-term passenger flow time series data of urban rail transit was analyzed and found to be non-stationary, nonlinear and non-periodic. With the help of deep learning algorithm, short-term passenger flow forecasting models based on LM-BP and LSTM neural networks were constructed respectively. Taking Suzhou Metro Central Garden Station as an example, the passenger flow time series data of a certain week was used as training data for 5 minutes to predict the change trend of short-term passenger flow in the next day, and the proposed forecasting method was verified. The results show that, compared with the cubic exponential smoothing model commonly used in parametric models, the prediction accuracy and stability of the proposed method are improved by about 50%, and the prediction accuracy is greatly improved. LSTM model has better prediction accuracy, while LM-BP model has more advantages in computational efficiency.