Prediction Of Flow Evolution Based On Built Environment

For urban planners,forecasting the 24-hour population fluctuation after the completion of the planning program is of great help to guide regional planning.At present,the main research direction in the field of urban planning and computing is focused on short-term traffic forecasting for traffic management and urban safety,which can not effectively help urban planners to design.This paper presents a complete process for predicting the 24-hour flow fluctuation in the urban area based on the built environment data.The results are realized and analyzed in Nanjing's built environment data and cell phone signaling data in 2015.The important factors affecting the 24-hour population fluctuation in the built environment are analyzed.The difficulties and solutions are as follows:Firstly,the fluctuation of the number of people in the region is related to the built environment in a large range,even with the relative orientation and distance.Therefore,this paper visualizes the built environment characteristics of the region,retains the relative location information and the distance information.In addition,the Pearson correlation coefficient sum was used to screen the data for the less correlated variables.Secondly,due to the low quality of LBS data,the discontinuity of space,and the low density of data per user,especially mobile signaling data,it can not directly support the statistics of regional users.For this reason,this paper first carries out path planning to fill the information gap through the Contraction Hierarchies algorithm at the micro level,and then restores the data to the macro-statistical level,and obtains the 24-hour expected number of users.Thirdly,considering the correlation and continuity of the 24-hour fluctuation in the number of people in the region,this paper firstly performs PCA dimensionality reduction on 24-hour flow evolution,obtains the principal component vector,and then uses the CNN network model based on ResNet residual unit to predict the regional flow of people in each main.The weights on the component vectors,and the evolution of the 24-hour number of people,reduce the prediction error.After the establishment of the model,we compare the predicted results with the results of other common algorithms,analyze the error of the forecast results in some regions,and find the built environment features with significant influence on the total number of changes in the 24-hour expected number and the number of people staying up late.