杨俊宴 史宜 孙瑞琪 王桥 顾杰

Yang Junyan, Shi Yi, Sun Ruiqi, Wang Qiao, Gu Jie

关键词：人群时空分布；建成环境；卷积神经网络；时空预测模型；人地关系

Keywords：Spatiotemporal Distribution of Population; Built Environment; Convolutional Neural Network; Spatiotemporal Prediction Model; Man-Land Relationship

摘要：

人群时空分布模型的建构方法是城市规划学科的长期研究议题，通信技术的发展为通过建构人群时空分布模型研究城市复杂系统提供了重要的数据来源和技术支撑。本文以500 m×500 m 的矩形栅格为空间分析单元，利用手机信令数据建构24小时人群密度分布特征数据集；集成32个建成环境因子形成指标库，建构建成环境特征数据集。基于卷积神经网络建构南京市人群时空分布预测模型，以城市建成环境特征为自变量、人群多时段密度分布情况为因变量，学习两者间的非线性关系，实现了对南京市人群分布情况的模拟。基于平均百分比误差值的空间分布，该模型呈现较高的拟合度。结果发现，模拟结果误差较小的空间分析单元位于南京市中心区域和远郊未集中建设区域，误差较大的空间分析单元位于城市拓展区和正在建设中的地区等人群活动不稳定的区域。

Abstract：

The construction method of population spatiotemporal distribution model is a longterm research topic of urban planning. The development of information and communication technology provides important data source and technical support for the construction of population spatiotemporal distribution model to study urban complex systems. In this paper, 500m × 500m rectangular grid is used as the research unit to construct the 24h population density distribution characteristic data set by using the mobile phone signaling data. It integrates 32 built-up environmental factors to form an index database, and constructs the built-up environmental characteristic data set. Based on the convolution neural network, the temporal and spatial distribution prediction model of population in Nanjing is constructed. Taking the characteristics of urban built environment as the independent variable and the multi period density distribution of population as the dependent variable, the nonlinear relationship between them is learned, and the simulation of population distribution in Nanjing is realized. Based on the spatial distribution of MAPE value, the model shows a high degree of fitting. Among them, the spatial units with small error in the simulation results are located in the central area of Nanjing and the non-centralized construction area in the outer suburbs, and the spatial units with large error are located in the areas with unstable crowd activities such as urban expansion area and construction area.

杨俊宴，博士，东南大学建筑学院，教授

史宜（通信作者），博士，东南大学建筑学院，副教授。shiyi@seu.edu.cn

孙瑞琪，硕士，江苏省规划设计集团有限公司，城乡规划师

王桥，博士，东南大学信息科学与工程学院，教授

顾杰，东南大学建筑学院，硕士研究生

• [1] 徐东云, 张雷. 城市交通拥堵与城市化进程的关系初探[J]. 综合运输, 2007(10): 12-16.

  [2] 杜江, 刘渝. 城市化与环境污染：中国省际面板数据的实证研究[J]. 长江流域资源与环境, 2008, 17(6): 825-830.

  [3] 陈海燕, 贾倍思. 紧凑还是分散？——对中国城市在加速城市化进程中发展方向的思考[J]. 城市规划, 2006, 30(5): 61-69.

  [4] 张学勇. 我国大城市地区新城成长与主城共生策略研究[D]. 哈尔滨: 哈尔滨工业大学, 2011.

  [5] 席广亮, 甄峰. 智慧城市建设推动新型城镇化发展策略思考[J]. 上海城市规划, 2014(5): 26-29.

  [6] CLARK C. Urban population densities[J]. Journal of Royal Statistics Society: series a, 1951, 114: 490-494.

  [7] 毛汉英, 方创琳. 我国新一轮国土规划编制的基本构想[J]. 地理研究, 2002, 21(3): 267-275.

  [8] 兰宗敏, 冯健. 城中村流动人口日常活动时空间结构——基于北京若干典型城中村的调查[J]. 地理科学, 2012, 32(4): 409-417.

  [9] 任常兴, 吴宗之, 刘茂. 城市公共场所人群拥挤踩踏事故分析[J]. 中国安全科学学报, 2005, 15(12): 102-106, 137.

  [10] SANTA GUZMáN L F. A statistical approach for studying urban human dynamics[D]. Lisbon: Universidade NOVA de Lisboa, 2018.

  [11] HOTEIT S, SECCI S, SOBOLEVSKY S, et al. Estimating real human trajectories through mobile phone data[C] // 2013 IEEE 14th International Conference on Mobile Data Management, 2013: 148-153.

  [12] BATTY M. The new science of cities[M]. Cambridge, MA: MIT Press, 2013.

  [13] LUCA M, BARLACCHI G, LEPRI B, et al. Deep learning for human mobility: a survey on data and models[J]. arXiv: 2012.02825.

  [14] LUCA M, BARLACCHI G, LEPRI B, et al. A survey on deep learning for human mobility[J]. ACM Computing Surveys (CSUR), 2021, 55(1): 1-44.

  [15] QIAN W. Discovering human mobility from mobile data: probabilistic models and learning algorithms[D]. Besancon: Université Bourgogne Franche-Comté, 2020.

  [16] 吴志强. 人工智能推演未来城市规划[J]. 经济导刊, 2020(1): 58-62.

  [17] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[J]. Communications of the ACM, 2012, 60: 84-90.

  
  

• 收起