CAJ | 학술논문

元胞自动机（Cellular Automata，CA）是进行城市空间演变模拟的重要建模工具。经典城市扩张CA模拟规则提取，主要利用一段历史变化样本对城市化（1值）和未城市化（0值）进行双向拟合，存在0值过度拟合现象，即历史观测不变化的元胞样本并不代表其没有转变的潜在可能性。为此，本文将城市空间增长潜力引入CA模型，重新构建CA规则学习样本和参数拟合目标，并利用粒子群优化算法进行参数挖掘，弥补传统CA规则提取的局限性。研究以广州市为案例区，基于主体功能区规划思想构建空间开发潜力，对改进的城市扩张CA模拟模型进行实例应用。结果表明，本文改进的CA模型不论在整体格局还是细节呈现上，均比传统CA模型表现出更高的可信度，模型整体评估精度高于70%，结果可为中长期城市规划提供更好的参考。
원포자동궤（Cellular Automata，CA）시진행성시공간연변모의적중요건모공구。경전성시확장CA모의규칙제취，주요이용일단역사변화양본대성시화（1치）화미성시화（0치）진행쌍향의합，존재0치과도의합현상，즉역사관측불변화적원포양본병불대표기몰유전변적잠재가능성。위차，본문장성시공간증장잠력인입CA모형，중신구건CA규칙학습양본화삼수의합목표，병이용입자군우화산법진행삼수알굴，미보전통CA규칙제취적국한성。연구이엄주시위안례구，기우주체공능구규화사상구건공간개발잠력，대개진적성시확장CA모의모형진행실례응용。결과표명，본문개진적CA모형불론재정체격국환시세절정현상，균비전통CA모형표현출경고적가신도，모형정체평고정도고우70%，결과가위중장기성시규화제공경호적삼고。
Geography simulation model such as Cellular Automata (CA) is one of the most important tools for simulating and early warning the urban growth. The CA model can simulate urban sprawling accurately only when suitable conversion rules for every cell are achieved. Hence, the core of CA is to derive the conversion rules, and many researchers have been interested in discovering the rules. However, the conversion rules of tradi-tional CA are mainly derived from historic samples, in which both changed samples and unchanged ones are con-sidered for function fitting to retrieve parameters simultaneously. In this approach, it is assumed that if the urban sprawling occurred, samples were labeled as 1; otherwise, samples were accordingly labeled as 0. However, it will result in over fitting for the unchanged samples, because those samples with labels of 0 may have the potenti-ality to transform in future, especially for those located at the rural-urban fringe. Therefore, we proposed a gradi-ent CA for simulating urban sprawling. In this model, whether or not urban growth would occur was determined by the developing probability instead of its developed or undeveloped status. Accordingly, the unchanged sam-ples were set to the values ranging from 0 to 1. And in this research, the developing potentiality was estimated ac-cording to present planning maps. Compared with traditional CA, the gradient CA could avoid the over fitting problem for the unchanged samples to a certain degree. Moreover, the fitting objective was distinguished from traditional CA for its ability in retrieving conversion rules. In addition, particle swarm optimization algorithm was used to obtain the parameters of spatial indices. Finally, Guangzhou City, which locates in the Pearl River Delta of China, was chosen as the study area for model implementation and validation. In this case study, the spa-tial developing potentiality was allocated referring to the major function zone (MFZ) planning, because MFZ is currently one of the most significant planning policies for Chinese government to control the chaotic urbaniza-tion. In order to evaluate the model’s efficiency, a comparison analysis was carried out between the gradient CA and traditional CA. Global and local patterns of the simulation results were analyzed respectively in details. Re-sults demonstrate that the model modified in this paper can perform efficiently and the overall accuracy of the model is greater than 70%, which can provide better and reasonable spatial scenarios for medium-and long-term urban planning.