基于卷积长短时记忆网络的短时公交客流量预测  

Short-term Bus Passenger Flow Prediction Based on Convolutional Long-short-term Memory Network

作　　者：陈静[1] 张昭冲 王琳凯 安脉 王伟 Chen Jing;Zhang Zhaochong;Wang Linkai;An Mai;Wang Wei(School of Information Technology and Engineering,Tianjin University of Technology and Education,Tianjin 300222,China;Smart City Development Authority,Sino-Singapore Tianjin Eco-city Management Committee,Tianjin 300467,China)

机构地区：[1]天津职业技术师范大学信息技术工程学院,天津300222 [2]中新天津生态城管委会智慧城市发展局,天津300467

出　　处：《系统仿真学报》2024年第2期476-486,共11页Journal of System Simulation

基　　金：天津市教委科研计划(2021KJ008)。

摘　　要：针对传统的短时客流预测方法没有考虑到时序特征中跨时段客流之间的相似性问题,提出一种改进k-means聚类算法与卷积神经网络和长短时记忆网络相结合的短时客流量预测模型k-CNN-LSTM。通过k-means算法对跨时段时序数据进行聚类,使用间隔统计确定k值,构建交通流矩阵模型,采用CNN-LSTM网络处理具有时空特征的短时客流。该模型能够对具有空间相关性的数据进行较为准确的预测。使用真实数据集对模型进行检验和参数调优,实验结果表明:k-CNN-LSTM模型较其他模型有相对较高的预测精度。To address the problem that the traditional short-time passenger flow prediction method does not consider the temporal characteristics similarity between the inter-temporal passenger flows,a shorttime passenger flow prediction model k-CNN-LSTM is proposed by combining the improved k-means clustering algorithm with the CNN and the LSTM.The k-means is used to cluster the intertemporal timeseries data,the k-value is determined by using the gap-statistic,and a traffic flow matrix model is constructed.A CNN-LSTM network is used to process the short-time passenger flows with spatial and temporal characteristics.The model is tested and parameter tuned by the real dataset.The test results show the model is able to predict the spatially correlated data more accurately.

关 键 词：卷积神经网络 长短时记忆网络 时空数据预测 K-MEANS聚类 客流量预测 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]