Research On Population Spatialization And Simulation Forecast Based On AWA-ESPCN

Population spatialization and population time-space simulation prediction are two important parts of the population research process,which are of great significance to the study of population distribution laws,population future planning,and economic policy formulation.In this thesis,in response to the series of problems in the process of traditional machine learning population spatialization modeling,such as incomplete local feature extraction and low simulation accuracy,taking Chongqing as an example,the introduction of Efficient Sub-Pixel Convolutional Neural Network(ESPCN)And LightGBM and other methods,starting from the three aspects of driving factor selection,model construction,and simulation prediction,the population spatialization model is constructed and the population simulation prediction for a certain period of time in the future is realized.The main research contents and conclusions are as follows:1.This thesis selects 9 driving factor data that can reflect the objective law of population distribution in the three aspects of natural factors,social factors and distance factors for model training.2.In order to solve the problem of incomplete local feature extraction,on the basis of Area Weighted Average(AWA),ESPCN was introduced to complete the spatialization of the population resolution of 500 meters in 2010.The results show that the mean square error of ESPCN is 2.1327.It is significantly lower than 4.1733 of Super-Resolution Convolutional Neural Network(SRCNN)and 3.5549 of Fast Super-Resolution Convolutional Neural Network(FSRCNN);its residual distribution effect is significantly better than the other 2 schemes.Studies have shown that ESPCN can effectively extract local features,thereby making the spatialization accuracy higher.3.The decision tree integrated learning model has mainly experienced the development of the core algorithms of random forest,Ada Boost,XGBoost,and LightGBM,and LightGBM is superior to XGBoost’s layered growth strategy due to the split strategy of the maximum revenue node.Based on the use of AWA-ESPCN to construct a spatialized data set of population and driving factors,this thesis uses LightGBM to learn the temporal characteristics of population spatial distribution to complete the population forecast of Chongqing in 2025,and compare it with random forest,Ada Boost,and XGBoost.The results show that the R2 and the predicted root mean square error of LightGBM are 0.9818 and 0.0479,respectively.R2 is higher than the random forest’s 0.9563,Ada Boost’s 0.9358 and XGBoost’s 0.9678,and the predicted root mean square error is less than the random forest’s 1.2022,Ada Boost’s 3.3767 and XGBoost’s 0.0579.Research shows that LightGBM integrates the advantages of splitting strategy according to the maximum revenue node,cache optimization and category feature processing,so that it can better learn the population timing characteristics,and the constructed coupling model can be effectively used for population simulation prediction.The innovation of this thesis is embodied in the introduction of ESPCN in the image field to improve the accuracy of population spatialization and the introduction of LightGBM in the field of decision tree integrated learning to improve the accuracy of population simulation prediction.