基于聚类算法和CNN+LSTM短时交通流量预测  

Clustering Algorithm-based Short-term Traffic flow Prediction with CNN and LSTM

作　　者：尹迁齐 高永强 齐龙 周士谦 YIN Qian-qi;GAO Yong-qiang;QI Long;ZHOU Shi-qian(School of Automotive Engineering,Shandong Jiaotong University,Jinan 250357,China)

机构地区：[1]山东交通学院汽车工程学院,山东济南250357

出　　处：《山东工业技术》2024年第6期57-65,共9页Journal of Shandong Industrial Technology

摘　　要：为了更精确地预测交通量,减轻交通阻塞的问题,采用聚类算法,卷积神经网络(CNN)与长短记忆网络(LSTM)相结合,建立一个交通流预测模型。该模型在兼顾时间序列特征地同时,兼顾了交通流量方式与空间特性之间地关系。首先,构造数据张量X,利用CNN挖掘相邻路口的数据X,并整合空间关联信息,将经过CNN处理后的数据输入LSTM模型,用于预测下一时间序列的数值。其次,将LSTM输出经过K-means聚类算法对交通模式进行划分,并计算聚类中心和权重,最终与LSTM的输出做加权计算,得到预测结果。通过多组实验证明本文所提出的模型预测精度优于其他模型,验证了在不同的流量模式下,将时空特征纳入其中,对短时交通流预测的有效性。To boost traffic volume prediction accuracy and mitigate congestion,a traffic flow prediction model was designed by integrating clustering algorithms with convolutional neural networks(CNN)and long short-term memory networks(LSTM).This model accounts for both the temporal aspects and the connections between traffic patterns and spatial features.Initially,a data tensor X is constructed,and CNN is used to analyze the data X from adjacent intersections,incorporating spatial correlation information.The CNN-processed data is then fed into the LSTM model to predict the next time series values.Next,the LSTM output is categorized into traffic patterns using the K-means clustering algorithm,and clustering centers and weights are calculated.Finally,the LSTM output is weighted to obtain the prediction results.A series of experiments have revealed that the proposed model outperforms other models in terms of prediction accuracy,showcasing the benefits of incorporating spatiotemporal features in short-term traffic flow prediction across different flow patterns.

关 键 词：短时交通流量预测 K-MEANS聚类 卷积神经网络 长短期记忆网络 

分 类 号：U4-9[交通运输工程—道路与铁道工程]