基于注意力机制和残差网络的地铁客流量预测方法研究

Research on subway passenger flow prediction method based on attention mechanism and residual network

研究基于注意力机制和残差网络的地铁客流量预测方法,达到提高客流量预测精度的目标,为城市交通智能管理提供数据保证。在地铁客流量预测问题描述的基础上,构建基于注意力机制和残差网络的客流量预测模型,将历史客流量数据划分成近邻、日周期、周周期模式时间序列数据,将其与客流量外部影响因素数据作为模型输入,分别利用结合ResNet34残差模块和分割注意力机制模块的ST-SANet网络,以及LSTM网络捕捉其更深层次多尺度信息特征,利用全连接层完成各部分输出特征的融合拼接,经过激活函数处理后,输出地铁客流量预测结果。实验结果表明:该方法可实现地铁客流量预测,学习率参数为1×10^(-4)时,地铁客流量预测损失最低;预测周期设定为15 min时,预测曲线与实际客流曲线贴合度最高。

Research on subway passenger flow prediction methods based on attention mechanism and residual network,aiming to improve the accuracy of passenger flow prediction and provide data guarantee for urban traffic intelligent management.On the basis of describing the problem of subway passenger flow prediction,a passenger flow prediction model based on attention mechanism and residual network is constructed.Historical passenger flow data is divided into time series data of nearest neighbor,daily cycle,and weekly cycle modes,which are used as model inputs along with external influencing factors of passenger flow,The ST-SANet network and LSTM network,which combine the ResNet34 residual module and the segmentation attention mechanism module,are used to capture their deeper multi-scale information features.The fully connected layer is used to fuse and concatenate the output features of each part.After processing with the activation function,the subway passenger flow prediction results are output.The experimental results show that this method can achieve subway passenger flow prediction with a learning rate parameter of 1×At 10 to 4 hours,the predicted loss of subway passenger flow is the lowest;When the prediction period is set to 15 minutes,the predicted curve has the highest fit with the actual passenger flow curve.