基于Softmax函数增强卷积神经网络—双向长短期记忆网络框架的交通拥堵预测算法

Traffic Congestion Prediction Algorithm Based on CS-BiLSTM Framework

对交通状态进行预测,需要准确识别和判断交通状态。基于道路自身的自由流速度,将具有不同速度等级的街道统一以旅行时间指数(travel time index, TTI)作为拥堵评价,相较于以车辆速度为基准的传统预测方法更能表现出道路的拥堵状态。提出了一种改进的深度学习预测模型(CS-BiLSTM),该模型基于卷积神经网络(convolutional neural networks, CNN)和双向长短期记忆(bidirectional long short-term memory, BiLSTM),并结合Softmax函数增强CNN提取出的交通空间特征信息。最后以成都市出租车的全球定位系统(global positioning system, GPS)数据进行验证。结果表明,所提出的CS-BiLSTM模型具有更高的准确性,其性能相比CNN-BiLSTM网络预测框架提升了13%。

To predict the traffic state,it is necessary to accurately identify and judge the traffic state.Based on the free flow speed of the road itself,the travel time index(TTI)was used as the congestion evaluation of the streets with different speed levels,which can better show the congestion state of the road than the traditional prediction method based on the vehicle speed.An improved deep learn-ing prediction model(CS-BiLSTM)was proposed,which was based on convolutional neural networks(CNN)and bidirectional long short term memory(BiLSTM),and combining Softmax function to enhance the traffic spatial feature information extracted by CNN.Fi-nally,the global positioning system(GPS)data of taxis in Chengdu were verified.The results show that the proposed CS-BiLSTM mod-el has higher accuracy,and its performance is 13%higher than that of CNN-BiLSTM network prediction framework.