文献类型：期刊文献

英文题名：Fine-Scaled Spatiotemporal Prediction of Bi-Hourly Urban Population Dynamics by Integrating Ga-Lstm Model and Neighborhood Annulus Features

作者：Zhang, Chenming[1]; He, Bei[2]; Fan, Qindong[1]; Zhang, Qian[3]; Ping, Xiaoying[4]; Li, Chunlin[5]; Wang, Qingzheng[6]

第一作者：Zhang, Chenming

机构：[1] College of Architecture, North China University of Water Resources and Electric Power, Henan, Zhengzhou, 450046, China; [2] School of Construction Management and Real Estate, Henan University of Economics and Law, Henan, Zhengzhou, 450046, China; [3] Collage of Tourism and Planning, Pingdingshan University, Henan, Pingdingshan, 467000, China; [4] School of Public Administration, North China University of Water Resources and Electric Power, Henan, Zhengzhou, 450046, China; [5] CAS Key Laboratory of Forest Ecology and Management, Institute of Applied Ecology, Chinese Academy of Sciences, Liaoning, Shenyang, 110016, China; [6] School of Information Engineering, North China University of Water Resources and Electric Power, Henan, Zhengzhou, 450046, China

第一机构：College of Architecture, North China University of Water Resources and Electric Power, Henan, Zhengzhou, 450046, China

年份：2023

外文期刊名：SSRN

收录：EI(收录号：20230416086)

语种：英文

外文关键词：Long short-term memory - Numerical methods - Population distribution - Population dynamics - Population statistics - Risk management

摘要：The prediction of population density with fine spatiotemporal granularity is fundamental to a variety of applications, from spatial planning to emergency management. Existing research has not yet achieved fine-grained mapping in both time and space dimensions simultaneously, nor has it utilized interactions between adjacent prediction units to increase accuracy. In this study, an integrated method incorporating GA-LSTM and neighborhood annulus features is proposed to predict and represent the bi-hourly urban population density in 250m×250m grids. Initially, a structure of multi-layer annuluses was devised to connect self and neighborhood attributes to a specific grid granular. Next, the model is trained and tested using data from the year 2020, and the prediction is made using data from the year 2021. Finally, the model’s efficacy is estimated. (1) The hybrid method can effectively enhance the performance of prediction, and both the numerical value and spatial pattern of the population map can be accurately predicted. (2) As the radius of the annulus increases, the effect of neighborhood characteristics on population dynamics ceases to be significant when the radius exceeds the threshold. (3) The contribution of various statistical indicators derived from the neighborhood annulus varies, indicating the complexity of population agglomeration mechanisms. ? 2023, The Authors. All rights reserved.