一种城市出行需求预测的时空方法  

A Spatio-Temporal Approach for Urban Travel Demand Forecasting

作　　者：楚本嘉 李建波 刘阳[2] 马照斌[1] 夏丰千 

机构地区：[1]青岛大学计算机科学技术学院,山东 青岛 [2]青岛大学泛在网络与城市计算研究所,山东 青岛

出　　处：《计算机科学与应用》2023年第3期518-527,共10页Computer Science and Application

摘　　要：预测城市出行需求对于交通管理、保障公共出行安全和建设智慧城市具有重要意义。然而,由于受区域间交通状况、天气、节假日等诸多复杂因素的影响,城市出行需求数据往往存在高频噪声和复杂的波动模式。本文提出了一种基于深度学习的城市出行需求预测模型(Spatio-Temporal Urban Travel Demand Forecasting Model, STUTDFM)。该模型的架构由外部因素影响组件、时空特征提取组件和数据融合组件组成。外部因素影响组件可以处理城市出行需求影响因素的数据,从而拟合一些局部极值,时空特征提取组件可以捕获城市出行需求数据的空间依赖性和时间依赖性,数据融合组件可以将外部因素影响组件和时空特征提取组件调整到整体预测模型中。对四个真实数据集的实验表明,所提出的城市出行需求预测模型方法优于八种众所周知的方法。Predicting urban travel demand is of great significance for traffic management, ensuring public travel safety and building smart cities. However, due to the influence of many complex factors such as inter-regional traffic conditions, weather and holidays, urban travel demand data often has high frequency noise and complex fluctuation patterns. This paper proposes a deep learning-based urban travel demand prediction model—Spatio-Temporal Urban Travel Demand Forecasting Model (STUTDFM). The architecture of the model consists of an external factor influence component, a spatio-temporal feature extraction component and a data fusion component. EIFC can process the data of the factors affecting urban travel demand so as to fit some local extremes, SPFEC can capture the spatial dependence and temporal dependence of urban travel demand data, and DFC can adjust EIFC and STFEC to the overall prediction model. Experiments on four real datasets show that the proposed STUTDF method outperforms eight well-known methods.

关 键 词：深度学习 城市出行需求 预测 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]