Simulation and Empirical Analysis of Regional Land-use Change Based on Multi-Agent System (MAS) Model
区域土地利用变化情景模拟设计与实证分析——基于多智能体系统(MAS)模型

As a typical model of artificial intelligence, MAS has become a hot research area within the field of land science. It is good at explaining complex process and has certain tolerance for uncertainty, fuzziness and inaccuracy. Besides, it can conduct dynamic simulation and phenomenon analysis on land-use change. This article expanded the application of MAS in land-use change to medium scale. With the city of Yixing, which is undergoing the dilemma of development and environment, as the research region, it mainly considered four agent types related to land-use change, namely the government, farmers, urban residents and owners of industrial undertakings. In the design of the scene of land-use change, this article focused on the following four major land-use patterns: the increase of the construction land; the conservation of arable land; land consolidation in rural regions and the intensive use of industrial land. After the preparation of data processing by the approach of geographic and spatial analysis in the platform of ArcGIS 9.3 and statistical analysis of socioeconomic data, this article pointed out the influence factors and driving forces of the land-use change in this city. In addition to canonical correlation analysis, logistic regression analysis and other analysis methods were used to analyze and design the behavior rule of the agent and comprehensively weigh the conversion probability of the land. Moreover, the interaction among agents and the interaction between agent and environment were also used to simulate human's action of decision-making on land-use change. The simulation was carried out in the platform of Swarm 2.2 and the results are as follows: 1) The reduction of arable land, especially the transformation from arable land to urban land at the edge of former urban centers, is dramatic and lasting. And this change would cause some social and environmental problems; 2) The concentration trend of rural residential land is obvious due to policies about rural land planning which are strongly emphasized in the design of the model; 3) The change of mining land is random. The overall accuracy of the model analog is 73.35%, which proves the feasibility of the study which applies MAS model into land-use change simulation. However, the simulation and prediction of rural residential land change are still very difficult and remain to be discussed and improved.