基于CNN-BiLSTM的高速公路交通流量时空特性预测

Prediction of temporal and spatial characteristics of freeway traffic flow based on CNN-BiLSTM

目前,高速公路交通管控部门对准确交通数据的掌握还存在局限性,预测值也不够精确,为给出行者提供更好的交通引导,必须采用新方法预估误差较小的交通流量数据。提出一种同时考虑时间与空间因素的卷积-双向长短期记忆(CNN-BiLSTM)模型,将具有空间局部特征提取能力的卷积神经网络(CNN)和具有能同时考虑前后方向长时间信息的双向长短期记忆(BiLSTM)相结合,将其用于预测更能体现随时空变化不断波动的交通流量。以一些简单的基准方法作为对比模型,采用均方误差(MSE)等5项评估指标,在美国加州高速公路数据集上进行训练和测试,结果表明:由CNN-BiLSTM得出的预测结果更符合实际交通流量的变化趋势,在高峰期和波动较大时间段均能较好地拟合真实值,灵敏度较高。

At present,highway traffic control departments still have limitations in their grasp of accurate traffic data,and the predicted values are not accurate enough.There are still limitations in the traffic control department's grasp of accurate traffic data.In order to provide pedestrians with better traffic guidance,new methods must be adopted to estimate the traffic flow data with less error.This paper proposes convolutional neural network-bidirectional long short-term memory(CNN-BiLSTM)model that considers both temporal and spatial factors.It combines convolutional neural network(CNN)with spatial local feature extraction capabilities and bidirectional long short-term memory(BiLSTM)that can simultaneously consider long time information in the front and back directions.Using it to predict traffic flow that better reflects the ever-changing changes in both time and space.This paper uses some simple benchmark methods as comparison models,applying five evaluation indicators such as mean squared error(MSE)and so on to train and test on the California highway dataset.The result shows that the predictions obtained by CNN-BiLSTM are more in line with the changing trend of actual traffic flow,the peak periods and time periods with large fluctuations can better fit the true traffic volume,and the sensitivity is higher.