Spatial Distribution Expression And Feature Mining Of Port Freight Volume Based On Random Forest

The rapid growth and high concentration of port throughput have brought challenges to the planning of the port's hinterland transportation system and the relationship between the port and the city.Carrying out the research on the spatial distribution of freight volume of the port is of great significance to the spatial distribution of freight volume at the micro scale,the exploration of the new type of socio-economic data spatialization,and the planning of the collection and distribution system.At present,there are research methods for the spatial distribution pattern of freight volume,where indicators such as unbalanced coefficients are used to analyze the relationship and scale of the spatial transfer of freight volume at the macro scale.The lack of research on the relationship between freight volume statistics and explanatory variables makes it difficult to predict freight volume values and express spatial distribution on the micro-grid scale.This paper takes DY Port and its hinterland as the research object,and selects seven kinds of data,which are of much information and contain different interpretation dimensions,such as port attractiveness indicators,night lighting data,land use,population density,POI data,road traffic and topography.And the features closely related to the port hinterland freight volume are extracted.They are trained in a random forest regression model and grid data of spatial distribution of freight volume is generated after evaluation and interpretation.The specific research results are divided into the following three aspects:(1)Extract port impact data,POI data,land use data,night light remote sensing data,population density data,road traffic data,and topography data.After feature engineering and feature screening,establish a multi-dimensional feature database of random forest regression models.(2)In this paper,a random forest regression model and multi-source spatial data are used to simulate the spatial distribution of freight volume in the hinterland of DY Port.The effect of the simulation is tested using the coefficient of determination(R~2),root mean square error(RMSE)and mean absolute error(MAE).And the R~2,RMSE,and MAE of the model are 0.7677,2.5606,and 1.3210,respectively.The test results of the model indicate that the model is of high accuracy and strong generalization ability.(3)By estimating the decision path,the contribution of the features that have an impact on the regression prediction of the model is analyzed,and the relationship between the random forest model and the feature contribution is discussed.The study finds that the port gravity index has an important influence on the model estimation results,and its contribution value increases with the increase of the eigenvalues;some factors that reflect the economic development level of the region also have a certain impact on the model estimation results,such as night light intensity,commercial Service POI kernel density,etc.The research method in this paper can intuitively show the spatial distribution of the freight volume of the port,and can provide support for the study of the relationship between the city and the port and the planning of the port's collection&distribution system.The research result can provide a reference for the research on the distribution of freight volume in other ports and their hinterland.It can also be used to further develop the model and simulate the spatial distribution of freight volume in the port hinterland under different hinterland development conditions to provide certain guidance for port construction and planning.