Prediction of Urban Underground Space Demand by Machine Learning
Accurate prediction of urban underground space demand is an important work for urban underground space planning.In view of the shortcomings of the current research,such as less consideration,qualitative focus,strong subjectivity and low prediction accuracy,based on 43 groups of urban underground space related data in the litera-ture,a multi factor urban underground space demand prediction model was established based on 9 kinds of machine learning algorithm for the first time in this paper.In the process of establishing the model,the characteristic data was normalization processed to eliminate the effect of feature dimension on the model performance.The feature was extracted to select the optimal feature combination.The grid search cross validation technology was used to optimize the model parameters.Finally,the root mean square error and the determination coefficient were used to evaluate the model performance.The calculation results show that the three most 3 important influencing factors of urban un-derground space demand are resident population density,regional average car ownership and regional average GDP,which the mean value of characteristic importance in different algorithm models is 0.342,0.187 and 0.172 respectively;The feature combination of F-l(that is all eight features used)is the optimal feature combination.At this time,the XGB algorithm model has the highest performance with a determination coefficient of 0.970 and a root mean square error of 460.2;Finally,the BAG algorithm model predicts the development intensity of underground facilities in Beijing in 2020 with the prediction error of 9.23％,which further reflects the high accuracy of the model.
国家科技期刊平台
NSTL
万方数据
