基于ACBiGRU模型的短时交通流量预测

Short-term traffic flow prediction based on ACBiGRU model

针对传统的短时交通流预测方法只关注交通流的时间特征而未考虑空间特征的问题,提出一种引入注意力机制的卷积神经网络和双向门控循环单元的组合预测模型(ACBiGRU).该组合模型利用引入注意力机制的卷积神经网络挖掘邻近道路交通流量的空间相关性,将注意力机制嵌入到卷积神经网络中,对卷积层的结果进行不同权重的关注,有效提取交通流的空间特征;然后通过BiGRU模型提取交通流的时间序列特征;最终将提取到的时间和空间特征融合,完成短时交通流预测.实验结果表明:在真实数据集上的ACBiGRU模型预测优于其他模型,预测结果的RMSE(均方根误差)比传统时间序列模型平均降低了8%,验证了结合时空特性的短时交通流预测的有效性和优越性.

Aiming at the traditional short-term traffic flow prediction methods only focused on the temporal characteristics of traffic flow without considering the spatial characteristics,a combined prediction model of convolutional neural network and bidirectional gated circulation unit(ACBiGRU)model with attention mechanism was proposed.The combined model used attention mechanism of convolution neural network to mine spatial correlation of adjacent road traffic flow,and embeded the attention mechanism in the convolution neural network.The results of the convolution layer with different weight were effective to extract the spatial characteristics of traffic flow,and then the BiGRU model was used to extract the traffic flow time series characteristics.Finally,the extracted temporal and spatial features were fused to complete the short-term traffic flow prediction.Experimental results show that the ACBiGRU model is better than other models in the real dataset,and the RMSE(root mean square error)of the prediction results is reduced by 8%on average compared with that of the traditional time series model,verifying the effectiveness and superiority of the short-term traffic flow prediction combining with the spatio-temporal characteristics.