Study Of Simulation And Prediction Of Population Spatial-temporal Dynamics

The population problem is the contradiction between human beings,nature,and society and economy caused by rapid changes in population size and spatial distribution.Chinese census data has the problem that the spatial and temporal resolution is not high and it cannot reflect the true spatial and temporal distribution of the population in the administrative unit.Most of the existing studies divide the time and space,and carry out the population spatialization,and the temporal prediction of the administrative area separately;the model accuracy of a small part of the combination of space and time is limited.How to achieve high-resolution population spatialization based on demographic data and construct a population spatial-temporal dynamic model to simulate population spatial-temporal changes are of great significance to scientifically predict future population distribution and alleviate the contradiction between population and nature,society and economy.Taking Chongqing as the research area,this article discusses three aspects of population spatialization,spatial-temporal coupling model construction and population pattern rationality.Firstly,based on the 2010 county statistical population,a500m-resolution population spatialization model is constructed through two steps of partitioned multivariate statistical regression and super-resolution convolutional neural network to complete global and local feature learning.Then,three spatial-temporal coupling models of population were constructed by using partitioned multivariate statistical regression-super-resolution convolutional neural network spatialization and ensemble learning methods of to realize future population spatial-temporal prediction.Finally,the Moran'I measuresed the imbalanced population distribution in Chongqing.To measure whether this imbalance is reasonable,an analysis of the population-land-industry coordination coupling degree in Chongqing is made,and constructive suggestions are made to optimize the population pattern.The main research results are as follows:1.The partitioned multivariate statistical regression-super-resolution convolutional neural network spatial model proposed in this thesis has a minimum root mean square error of 1.51,and its spatial residuals are generally better than other schemes.Studies have shown that the combination of global and local features makes up for the problem of incomplete learning ability of a single model to a certain extent.2.Spatial-temporal prediction model coupled with spatial multivariate statistical regression-super-resolution convolutional neural network-XGBoost,compared with the previous population forecast based on random forest administrative unit level,spatial resolution has been improved.The root-mean-square error and relative error of the XGBoost time series prediction model are 1.68 and 2.88%,which are smaller than 1.79 and 6.80% of GBDT,and 2.18 and 7.14% of random forest.Studies have shown that the spatio-temporal coupled prediction model can improve the spatial resolution of population distribution while ensuring the accuracy of time series prediction.3.Due to the imbalance of population distribution in Chongqing,this thesis conducted a population-land-industry coordination coupling analysis.Studies have shown that in major urban areas with high population imbalances,the coordination coupling is higher than 0.7.Studies have shown that uneven population distribution does not mean that it is unreasonable.In the future,the optimization of Chongqing' s population pattern should focus on land use efficiency,economic development,and population evacuation and guidance.