基于GRU算法的轨道交通客流预测模型  

Rail transit passenger flow prediction model based on GRU algorithm

作　　者：钟杰宏 校景中[1] 刘中毅 ZHONG Jie-hong;XIAO Jing-zhong;LIU Zhong-yi(School of Computer Science and Engineering,Southwest Minzu University,Chengdu 610041,China)

机构地区：[1]西南民族大学计算机科学与工程学院,四川成都610041

出　　处：《西南民族大学学报（自然科学版）》2023年第2期206-212,共7页Journal of Southwest Minzu University(Natural Science Edition)

基　　金：西南民族大学中央高校基本科研业务费专项(2021NYYXS122)。

摘　　要：轨道交通的高速发展,使得站点的客流压力加剧,拥挤问题也带来了安全隐患.为简化客流预测模型训练时间,轻量化模型,采用K-means聚类,将客流数据进行分类,归一化数据,简化数据分布.在划分训练集和测试集后,分别利用长短时记忆网络(LSTM)模型和门控循环单元(GRU)模型对数据集进行训练.在不同时间粒度下分析了模型的可行性,对比两种算法的损失函数和运行时间.实验结果表明,在预测结果的准确性相近的情况下,GRU模型比LSTM模型有更短的拟合时间,同时模型本身更加简单,有着更好的适用性.With the rapid development of rail transit,the pressure of passenger flow has intensified at the stations,and the con⁃gestion problems have also brought hidden dangers to safety.In order to simplify the training time of passenger flow prediction model and lightweight model,k⁃means clustering was adopted to classify passenger flow data,normalize data and simplify data distribution.After dividing the training set and test set,LSTM model and GRU model were used to train the data set.The feasi⁃bility of the model was analyzed under different time granularity,and the loss function and running time of the two algorithms were compared.Experimental results showed that GRU model had shorter fitting time than LSTM model when the accuracy of prediction results was similar,and the model was simpler and had better applicability.

关 键 词：客流预测 轨道交通 门控循环单元 K-MEANS聚类 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] U293.13[自动化与计算机技术—控制科学与工程]