一种基于机器学习的城市发展模拟元胞模型  

Modelling land use changes with machine learning-based cellular automata in a GIS environment

作　　者：冯永玖[1] 刘妙龙[2] 

机构地区：[1]上海海洋大学海洋科学学院,上海201306 [2]同济大学测量与国土信息工程系,上海200092

出　　处：《测绘科学》2011年第3期216-218,共3页Science of Surveying and Mapping

基　　金：国家自然科学基金资助项目(40771174);上海高校选拔培养优秀青年教师科研专项基金资助项目(ssc09018)

摘　　要：利用元胞自动机(Cellular Automata,CA)模拟土地利用变化,已经成为认识和理解其复杂动态演化过程的有效手段。传统的元胞自动机基于线性转换规则,较难表达土地利用变化的非线性边界问题。本文研究利用最小二乘支持向量机方法(LS-SVM),将原空间下的非线性可分问题,通过高斯径向基核函数映射到高维特征空间,简化其求解过程,从而建立了一种非线性的土地利用元胞自动机模型LS-SVM-CA。利用该模型对上海市嘉定区1989-2006年的土地利用变化进行模拟的试验表明,其模拟结果与该区域土地利用实际格局非常符合,且其总体精度和Kappa系数比基于标准SVM的元胞自动机模型更高。Modelling land use changes by using cellular automata becomes an effective way to understand its complex dynamic evoluation recently.Linear methods-based conventional CA models have difficulties in dicribing the nonlinear dynamics of land use patterns.In the paper,with least square support vector machine mentods（LS-SVM）,nonlinear problems in original space were projected into high-dimensional feature space by using gaussian radial basis kernel function to create a new CA model——LS-SVM-CA.The proposed LS-SVM-CA model was successfully applied in modelling land use changes of Jiading district,Shanghai from 1989 to 2006 which demonstrated that the simulated results matched well with the actual land use structure,and the overall accuracy and Kappa coefficients were higher than that of the CA model based on standard SVM.

关 键 词：元胞自动机 最小二乘支持向量机 土地利用变化模拟 遥感 GIS 

分 类 号：P208[天文地球—地图制图学与地理信息工程] P237[天文地球—测绘科学与技术]