摘要
现有的共享单车预测模型大多将共享单车视为封闭交通系统,忽略了不同交通系统之间的交互影响,因此设计了一种使用非负矩阵分解算法改进的图卷积神经网络。首先,利用非负矩阵分解算法将其他交通系统的需求数据分解为不同的出行模式;其次,确定不同出行模式的含义;最后,将分解后的需求信息作为辅助信息与共享单车需求数据一同输入图卷积神经网络中进行预测。实验结果表明:与不考虑其他交通方式影响的模型相比,使用非负矩阵分解算法改进的图卷积神经网络的平均绝对误差下降了10.84%,并且非负矩阵分解方法能较好地解释辅助交通系统是如何提升单车需求预测效果的。
Most existing shared bicycle prediction models consider shared bicycles as closed transportation systems and ignore the interaction between different transportation systems.Therefore,this paper proposes to design a graph convolutional neural network improved by non-negative matrix decomposition algorithm.Firstly,the demand data of other transportation systems are decomposed into different travel modes using the non-negative matrix decomposition algorithm.Secondly,the meanings of different travel modes are determined.Finally,the decomposed demand information is used as auxiliary information and input into the graph convolutional neural network along with the shared bicycle demand data for prediction.Experimental results show that compared with models that do not consider the influence of other transportation modes,the average absolute error of the graph convolutional neural network improved by the non-negative matrix decomposition algorithm has decreased by 10.84%,and the non-negative matrix decomposition algorithm can better explain how auxiliary transportation systems improve the effectiveness of bicycle demand forecasting.
作者
胡智维
张建同
HU Zhiwei;ZHANG Jiantong(Department of Management Science and Engineering,Tongji University,Shanghai 200092,China)
出处
《软件工程》
2024年第2期10-15,共6页
Software Engineering
基金
国家自然科学基金面上项目(71971156)。
关键词
非负矩阵分解
图卷积神经网络
共享单车需求
可解释性
non-negative matrix decomposition
graph convolutional neural network
shared bicycle demand
interpretability