【作者】 王宇；

【导师】 徐建刚；

【作者基本信息】 南京大学 ， 城乡规划学， 2019， 硕士

【副题名】以长汀县城为例

【摘要】 改革开放40多年来,随着社会经济的快速发展和城镇化的快速推进,我国城市用地发生了显著变化,研究城市用地演变规律已成为规划学科的前沿热点。然而,现有的用地演变模拟研究大多集中在较大区域空间层面,用地类型划分简单,难以精确描述城市内部各种用地类型的空间分布特征、相互作用模式和驱动因子间的耦合关系。同时,由于城市用地演变存在明显的不确定性、非线性、多样性等复杂特征,传统建模方法显得捉襟见肘。近年来,以人工神经网络、决策树为代表的机器学习算法逐渐被应用到用地模拟和预测的研究中,为探究城市内部各种用地与驱动因子之间的复杂空间耦合规律提供了新的方法思路。时至今日,国内相关研究大多聚焦于大城市,而对山地小城市的关注较少。与大城市相比,山地小城市在规模尺度、地理特征、产业结构和发展进程等方面存在很大差异,因而其用地的演变规律也可能与大城市有所差异。为了避免重走大城市无序蔓延式的城市发展道路,本文以福建省长汀县城为例,借助机器学习耦合元胞自动机模型,构建一种适用我国山地小城市的用地演变模拟分析方法,来探究山地小城市新型城镇化发展的时空演变规律。全文分为三个部分。第一、二章主要阐述研究背景,界定复杂性科学、元胞自动机、机器学习等基本概念。通过回顾国内外关于元胞自动机及其耦合算法的相关研究进展,提出研究内容和方法,阐述研究意义和研究框架。并介绍了研究区概况、数据来源及其空间化方法。第二部分第三到六章为研究主体部分。第三章是城市用地演变的系统分析,通过建立城市系统动力学模型来定量阐述用地演变机制,具体利用Vensim PLE软件,预测不同发展情境下各种用地规模的目标总量,为后续城市用地演变的预测提供目标总量约束参数。第四章是基于机器学习的城市用地适宜性的训练,利用FLUS软件中的人工神经网络模块计算社会经济、空间规划、生态环境、道路交通四大类因子驱动下的用地适宜性概率,引入ROC曲线分析和空间相关分析方法,探讨了城市用地布局和各类驱动因子的空间耦合规律,并筛选出18个强相关驱动因子。第五章是基于机器学习的城市用地转换规则的训练,根据第四章筛选出的强相关驱动因子,利用GeoSOS软件中的决策树模块探究各类驱动因子对用地演变的影响,获得各种用地的转换规则。第六章是基于元胞自动机的城市用地演变的模拟和预测,基于前两章得到的用地适宜性和转换规则,开展用地演变模拟和精度检验,并根据第三章获得的各种用地规模的目标总量预测结果,预测理想发展情境下城市用地空间演变模式。第三部分为总结与展望,主要研究结论如下:本文通过构建一种基于机器学习耦合元胞自动机的城市用地演变模拟和预测方法及实现技术,发展了山地小城市用地演变是由宏观用地规模目标总量整体控制和微观用地适宜性与转换规则共同决定的城市用地演变过程化分析的新途径。通过长汀县城的实证模拟,初步提炼出山地小城市用地规模扩展和空间功能演变的如下基本规律:(1)城市用地规模演变是土地利用、人口、经济、交通和环境五个子系统相互作用的结果。人口决定了各种用地的规模,固定资产投资、居民消费支出和货运量等经济指标是直接推动各种用地规模演变的经济驱动力,但是城市用地规模的扩张也会受到环境因素的制约,从而导致盲目扩张式的发展模式难以为继。(2)城市用地空间布局主要受社会经济、空间规划、生态环境、道路交通等因子驱动下的用地适宜性概率的影响。长汀县城实证表明:常住人口、就业密度与居住用地布局存在高度相关性;空间规划与公服用地、商业用地、道路用地和市政用地之间存在高度相关性;大部分城市用地更倾向于分布在高程较低、地势平坦的地区,二类工业用地与自然保护区之间存在互斥性,教育科研用地和工业用地往往远离洪水淹没区布局;街巷和主干路是集聚城市公服和商业用地和的重要媒介,生产类用地主要分布在远离城市的公路周边地区,在保证交通通达性的同时降低邻避效应。(3)城市用地空间演变是由用地转换规则直接决定的。长汀县城实证表明:发现人口密度的提高加速农林用地向建设用地的转换,人口密度的降低促进居住用地向公服用地或商业用地发生功能置换;高房价催生出更多的公服用地需求,而生产类用地、居住用地和农林用地之间的相互转换更容易发生在房价水平较低的地区;公服用地、商业用地、道路用地和市政用地更倾向于符合用地规划进行空间转换,而用地规划对于农林用地向二类工业用地转换方面约束较弱;农林用地转换为建设用地常常发生在海拔较低的地区;生态类用地和生产类用地更倾向于在距街巷较近的地区转换为公服和商业用地;公服、商业、农林和一类工业等用地在用地演变过程中表现出明显的集聚效应。本文正文共约60 000字,图表96幅。更多还原

【Abstract】 Over the past 40 years of reform and opening up,with the rapid development of social economy and urbanization,significant changes have taken place in urban land use in China.The study on the evolution law of urban land use has become the frontier hot spot in the planning discipline.However,most of the existing land use evolution simulation studies focus on the large regional spatial level.The classification of land use types is simple,and it is difficult to accurately describe the spatial distribution characteristics,interaction patterns and the coupling relationship between driving factors of various land use types in cities.At the same time,traditional modeling methods appear to be stretched due to the obvious complex characteristics of urban land use evolution,such as uncertainty,non-linearity and diversity.In recent years,machine learning algorithms,such as artificial neural networks and decision trees,have been gradually applied to the study of land use simulation and prediction,which provides a new method for exploring the complex spatial coupling law between land use and driving factors in cities.Up to now,most of the relevant domestic research has focused on big cities,while less attention has been paid to small mountain cities.Compared with big cities,small mountain cities have great differences in scale,geographical features,industrial structure and development process,so the evolution law of land use may be different from that of big cities.In order to avoid re-taking the urban development path of the disorderly spread of big cities,this paper takes Changting County of Fujian Province as an example,constructs a simulation analysis method of land use evolution suitable for small mountain cities in China by means of machine learning coupled cellular automata model,and explores the temporal and spatial evolution law of the new urbanization development in small mountain cities.The full text is divided into three parts.Chapters 1 and 2 mainly elaborate the research background and define the basic concepts of complexity science,cellular automata,machine learning and so on.By reviewing the research progress of cellular automata and its coupling algorithms at home and abroad,the research contents and methods are proposed,and the research significance and framework are expounded.The overview,data sources and spatialization methods of the study area are also introduced.Chapter 3 to 6 of the second part are the main parts of the study.Chapter 3 is the systematic analysis of urban land use evolution.The mechanism of land use evolution is described quantitatively by establishing a dynamic model of urban system.The Vensim PLE software is used to predict target quantity of various land use under different development scenarios,which provides target total constraints parameters for the subsequent prediction of urban land use evolution.Chapter 4 is the training of urban land suitability based on machine learning.The artificial neural network module in FLUS software is used to calculate the probability of land use suitability driven by four major factors:social economy,spatial planning,ecological environment and road traffic.The ROC curve analysis and spatial correlation analysis are introduced to discuss the spatial coupling law of urban land distribution and various driving factors.18 strong correlation drivers were screened out.Chapter 5 is the training of urban land use conversion rules based on machine learning.According to the strong correlation driving factors selected in Chapter 4,the decision tree module of GeoSOS software is used to explore the influence of various driving factors on land use evolution,and obtain various land conversion rules.Chapter 6 is the simulation and prediction of urban land use evolution based on cellular automata.Based on the land suitability and conversion rules obtained in the previous two chapters,the simulation and accuracy test of land use evolution are carried out.According to the target forecasting results of various land use obtained in Chapter 3.the spatial evolution model of urban land under ideal development scenarios is predicted.The third part is the summary and prospect.The main conclusions are as follows:In this paper,a new method of urban land use evolution simulation and prediction based on machine learning coupled cellular automata and its implementation technology is constructed.A new procedural analysis way of urban land use evolution in small mountain cities,which the land use evolution is determined by the overall control of the macro-scale target and the micro-scale suitability and conversion rules,is developed.Through the empirical simulation of Changting County,the following basic laws of land use expansion and spatial function evolution of small mountain cities are preliminarily extracted:(1)The quantitive evolution of urban land use is the result of the interaction of five subsystems:land use.population,economy,transportation and environment.Population determines the scale of various land use.Economic indicators such as fixed assets investment,consumer expenditure and freight volume are the economic driving forces that directly promote the quantitive evolution of land use.However,the expansion of urban land use will also be constrained by environmental factors,which leads to the unsustainable development of blind expansion.(2)The spatial distribution of urban land use is mainly influenced by the probability of land suitability driven by social economy,spatial planning,ecological environment and road traffic.The empirical study of Changting County shows that:it is found that there is a high correlation between permanent population,employment density and the distribution of residential land;there is a high correlation between spatial planning and public service land,commercial land,road land and municipal land;most of the urban land tends to be distributed in areas with lower elevation and flat terrain,there is mutual exclusion between the second-class industrial land and the nature reserves,educational land and industrial land are often far away from the distribution of flood-inundated areas;alleys and main roads are important media for gathering urban public service and commercial land,and production land is mainly distributed in the surrounding areas of highways far away from city,which ensures traffic accessibility and reduces the neighborhood avoidance effect.(3)The spatial evolution of urban land use is directly determined by land use conversion rules.The empirical study of Changting County shows that:it is found that the increase of population density accelerates the conversion of agricultural and forestry land to construction land,and the decrease of population density promotes the functional replacement of residential land to public service or commercial land;high housing prices generates more demand for public service land,while the conversion between production land,residential land and agricultural and forestry land is more likely to occur in areas with lower housing prices;public service land,commercial land,road land and municipal land are more inclined to conform to land use planning for spatial conversion,while land use planning has weak constraints on the conversion of agricultural and forestry land to second-class industrial land;conversion of agricultural and forestry land to construction land often occurs in areas with lower elevation;ecological land and production land are more likely to be converted to public service and commercial land in areas closer to alleys;public service and commercial land,agricultural and forestry land and first-class industrial land show obvious agglomeration effect in the process of land use evolution.The whole thesis contains about 60 000 words,96 pictures and charts.更多还原