| 摘要: |
| 乡村振兴战略提出以来，乡村聚落空
间重构的特征及其内在机理剖析一直是讨论的
重点。传统定性分析已不足以在当今数据迸发
与技术进步的情况下客观、准确量化描述乡村
聚落空间重构的动力机制。基于多源多维数据的
转译与整合，文章运用机器学习中的监督学习算
法（GBDT回归模型），构建一种量化剖析乡村
聚落的空间重构规律及其动力机制的方法，从
而建立“空间—动力”耦合机制的客观分析过
程，获得更科学合理的乡村发展动力识别结果。
并以江苏省溧阳市为实证对象，定量分析其聚
落收缩与扩张的空间特征，并运用机器学习方法解析溧阳市乡村聚落空间重构的动力作用机制，发现城区距离、景区距离、社会活力对溧阳的
聚落空间扩张和收缩影响程度较大。 |
| 关键词: 乡村聚落 “空间—动力”耦合机制 机器学习 溧阳市 |
| DOI：10.13791/j.cnki.hsfwest.20220401 |
| 分类号: |
| 基金项目:“十三五”国家重点研发计划项目（2018YFD1100300） |
|
| The Method of “Space-Dynamic” Coupling Mechanism of Rural Settlements Based onMachine Learning: Taking Liyang City, Jiangsu Province as an Example |
|
FU Peng,XIAO Jing,ZHAO Zhiqi,XIE Xin
|
| Abstract: |
| Since the “rural revitalization strategy” was put forward, the characteristics of
rural settlement space reconstruction and the analysis of its internal mechanism have been
the focus of discussion. Traditional qualitative analysis is no longer enough to objectively and
quantitatively describe the dynamic mechanism of spatial reconstruction of rural settlements in
the context of today’s data burst and technological progress.
After more than ten years of research and development, the dynamic mechanism of rural
settlement reconstruction has gradually shifted from a single drive to a multi-interaction
mechanism. The research method has gradually shifted from qualitative analysis to quantitative
research methods such as spatial description, index system, and statistical model, in order to
more accurately and intuitively describe the characteristics of spatial reconstruction and its
mechanism under the action of different dynamic factors. Based on this, machine learning has
shown great advantages over traditional methods in evaluating, verifying, and predicting land
use changes. Scientific explanations and solutions to urban and rural complexities. The machine
learning method for analyzing the “spatial-dynamic” mechanism of rural settlements is to
identify, extract and assign the “spatial” features of the reconstruction of the settlement, and
at the same time correlate with its “dynamic” factors to form a complete data set. The learning
algorithm calculates the coupling relationship between different “spaces” and “dynamics”, and
extracts the effect strength of different influencing factors. The specific method is based on the
construction of a comprehensive cross-temporal and spatial scale database for rural settlement
reconstruction.
The method is divided into four steps. The first step is to translate and express the data
through relevant dimensionality reduction and quantification methods, and establish multi-
dimensional correlation basic data. Overlay analysis of the rural land vector data before andafter settlement reconstruction in the region to identify the expansion or contraction of construction land patches, and use ArcGIS to convert the
patches into spatial points with area attributes. The spatial markers were subjected to kernel density analysis by using the Kernel Density analysis
tool of ArcGIS. The calculated sum density values are used as dependent variables to characterize the intensity of the village spatial reconstruction.
The second step, the selection and translation of multi-dimensional driving data, is to select a total of 14 influencing factors including four
dimensions. Prepare relevant land use, POI, economic and social data, translate and obtain indicators such as settlement location distance, spatial
resources, physical geographic metrics, and social and economic conditions, integrate them into the three-dimensional attributes of the settlement
space, and prepare a data set of reconstructed influencing factors. The third step, the choice of machine learning method for dynamic mechanism
analysis is to study the influence factors and dynamic mechanism of GBDT model based on gradient boosting algorithm on space. The fourth step,
feature importance calculation and action mechanism explanation for the two types of data sets of settlement expansion and settlement contraction,
parameter settings and model operations are performed. The result of feature importance is directly related to the strength of the dynamic factors
in the spatial reconstruction of settlements, which reflects the significant degree of the driving force of the rural settlement reconstruction, and
represents the strength of the role of different dynamic factors on the spatial reconstruction.
During the period from 2010 to 2018, the rural settlements in Liyang City generally showed a trend of settlement shrinkage higher than
settlement expansion. The marked 1 060 settlement expansion feature point datasets and 768 settlement shrinkage feature point datasets were
used to establish a gradient boosting tree (GBDT) regression model using the training set data, and the established regression model was applied
to the training and testing data. Get model evaluation results. The study found that the spatial vitality and location advantage of the area have a
greater impact on the expansion of settlement space; urbanization development and economic income have the strongest impact on the shrinkage of
settlement space; tourism development has a prominent effect on the reconstruction of rural settlements in Liyang City.
Due to the predictability of the machine learning regression model, this method can be applied to the layout planning of urban and rural spaces
to improve the efficiency and scientificity of planning. However, the machine learning method to analyze the dynamic mechanism of settlement space
reconstruction also has certain limitations. Machine learning relies on data-driven, and the selection and absence of data have a large difference in
the results of machine learning. |
| Key words: Rural Settlement Space-Dynamic Coupling Mechanism Machine Learning Liyang City |
