基于区块多源时空数据的道路网密度预测模型  

Road network density prediction model based on block multi-source spatio-temporal data

作　　者：黄士琛 邵春福[1] 王晟由 HUANG Shichen;SHAO Chunfu;WANG Shengyou(School of Traffic and Transportation,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]北京交通大学交通运输学院,北京100044

出　　处：《北京交通大学学报》2021年第5期93-100,共8页JOURNAL OF BEIJING JIAOTONG UNIVERSITY

基　　金：国家自然科学基金(52072025)。

摘　　要：城市化进程与人口增速的不协调导致了城市蔓延现象,为科学合理地统筹引导道路建设,缓解城市蔓延引发的交通拥堵和环境污染,需要对建成区的道路网指标进行精细化管理.以道路网密度为例,考虑道路建设进度和政策实施惯性,构建基于深度学习的区块化时空数据预测模型BiConvlstm2DNet,与之对应地构造基于区块的多源数据融合流程,并以某市为例进行实验.将该市时序的土地覆盖、人口和道路网拓扑结构融合为时空多源数据集,而后在数据集上将本模型同其他经典预测模型进行参数标定和对比.研究结果表明,BiConvlstm2DNet在该数据集上得到的准确率为91.5%,较支持向量回归和随机森林回归等模型的准确率平均提升了8.0%,是一种分区块预测建成区道路网指标的可靠模型.The disharmony between urbanization and population growth leads to urban sprawl. To guide the road construction in a scientifical and reasonable way as well as ease the traffic congestion and environmental pollution caused by urban sprawl, we need to finely manage the road network indexes in the built-up area. In this paper, the road network density is used as an indicator;considering the influence of the road construction process and policy implementation time, this paper proposes a block-based spatio-temporal data prediction model, Biconvlstm2 DNet, on the basis of deep learning. Correspondingly, a block-based multi-source data fusion process is constructed and experimented with a certain city as an example. The temporal land cover, population and road network topology of the city are fused into a spatio-temporal multi-source dataset, and then the parameters of this model are calibrated and compared with other classical prediction models in the dataset. The results show that BiConvlstm2 DNet obtains a prediction precision of 91.5% on the above dataset, which is 8.0% higher than the average accuracy of models such as support vector regression and random forest regression, so Biconvlstm2 DNet is a reliable block-based model for predicting the road network indicators in built-up areas.

关 键 词：智能交通 道路网密度预测 深度学习 用地覆盖 多源数据融合 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]