摘要
针对高速公路短时交通流预测问题中待测站点上下游的交通流量时空信息利用不充分,且上下游观测点选择不合理的问题,提出了基于观测点遴选并充分挖掘时空信息的短时交通流预测方法。首先使用KNN算法对待测站点的上下游节点进行遴选,将与待测站点欧氏距离较小的上下游节点历史数据组织成包含时空信息的二维矩阵;然后使用卷积神经网络提取空间特征,将所得的特征向量送入双向LSTM模型进行时间信息的提取并完成最终预测。结果表明,经过观测点遴选后的KNN-CNN-BiLSTM预测模型准确率较遴选前提升了19.3%,实现了交通流时空信息的充分挖掘,是一种有效精准的短时交通流预测模型。
In view of insufficient utilization of spatio-temporal information about upstream and downstream short-term traffic flow of stations under test on highways as well as unreasonable selection of upstream and downstream observation points, a short-term traffic flow prediction method was proposed in this paper on the basis of observation point selection and full mining of spatio-temporal information. Firstly, upstream and downstream nodes for the station under test were selected through the KNN algorithm, and the historical data of the upstream and downstream nodes with a smaller Euclidean distance from the station under test were organized into a two-dimensional matrix containing spatio-temporal information, Then, spatial features were extracted by using the convolutional neural network, and the resulting feature vectors were sent to the bidirectional LSTM model for extraction of time information and completion of final prediction. The results showed that the accuracy of the KNN-CNN-BiLSTM prediction model after the selection of observation points was raised by 19.3%, as compared with that before such selection, thus realizing full mining of spatio-temporal information about traffic flow. What was presented in this paper was an effective and accurate short-term traffic flow prediction model.
作者
徐先峰
宋亚囡
黄刘洋
夏振
潘卓毅
Xu Xianfeng;Song Yanan;Huang Liuyang;Xia Zhen;Pan Zhuoyi(College of Electronics and Control Engineering,Chang’an University,Xi’an Shaanxi 710064,China)
出处
《电气自动化》
2021年第5期95-96,114,共3页
Electrical Automation
关键词
交通流预测
卷积神经网络
时空信息
traffic flow prediction
convolutional neural network
spatio-temporal information
