采用多任务学习预测短时公交客流

Predict Short Term Bus Passenger Flow via Multi-Task Learning

现有公交线路短时客流预测方法主要依赖于单条线路的信息,忽略了多条线路之间的相关关系对客流预测的提升效果。针对这一问题,提出一种基于相关度分析和多任务学习的公交客流预测方法。利用灰色关联分析和皮尔逊相关系数获取公交线路之间的关联度系数,计算当前线路的相关线路集合;将相关线路的客流预测作为当前线路客流预测的辅助任务,建立基于门控循环单元(GRU)神经网络的多任务深度学习模型来预测客流。实验结果表明,该多任务学习模型在预测精度方面优于传统的时间序列预测模型以及仅考虑单条线路信息的神经网络预测模型。

Most of the existing methods for short-term bus passenger flow prediction only rely on the information of a single bus line,which may ignore the impact of the correlation of multiple lines on the improvement of prediction accuracy.Thus,a prediction method based on correlation analysis and multi-task learning is proposed.Firstly,the correlation coefficient of multiple bus lines is obtained by using grey correlation analysis and Pearson correlation coefficient.Then,taking the passenger flow prediction of relevant bus lines as the auxiliary task of the passenger flow prediction of the current line,a multi-task deep learning model based on gate recurrent unit(GRU)neural network is established to predict the passenger flow.Experimental results show that the multi-task deep learning model performs better on the prediction accuracy compared against the traditional time series prediction model and the neural network model that considers a single bus line only.