Study On The Urban Spatial Growth Model Based On Multi-agent System

The urbanization growth will continue since China is in the context of promoting the construction of new urbanization. With the influx of immigrants into the city, the urban construction land demand will continue to increase, which will lead to a waste of land resources, a loss of agricultural land in the suburbs and other contradictions between people and land. Land use change brought by urban spatial growth is an important research content of LUCC.On the micro level, urban spatial growth is characterized by the accumulation and concentration of the practice of human being in the city’s geographical environment, which has the characteristics of complex adaptive systems. To better understand the process of the micro-power under macroscopic phenomena and the complexity of urban growth, the study chooses a "bottom-up" modeling method and constructs an urban spatial growth model based on multi-agent system.The main contents of the study include:(1) The microscopic agents involved in the urban spatial growth was classified into four types:residents, industry, developers and government. Construction suitability of land use units can reflect the geographical preferences of different types of agents. Learning function was designed to reflect the microscopic interactions between agents and government spatial regulation zones was considered to reflect the macro guidance role of government.(2) Learning from previous studies, the driving factors that affect the growth of urban space was selected. For residents (high, medium and low-income residents) and industrial enterprises, construction suitability of land use units was assessed by AHP which is a combination of qualitative method and quantitative method. For developers, construction suitability of land use units was assessed by logistic regression. Global logistic regression model and geographically weighted logistic regression (GWLR) model was compared according to the model’s fitting effect of the sampling points. The better way was selected as the assessment result.(3) At first, the virtual experimental data was used to illustrate the features of the model, then the model is applied to simulate urban spatial growth in the Research Area. Different scenarios was compared to test sensitivity of the model’s parameters. The simulation results show the impact of the learning function and government factor.The results show that:(1) Multi-agent model constructed in this paper can be used to simulate the urban spatial growth. On the basis of the classification of agents, urban spatial growth simulation was achieved under the interactions of agents by the iterations of four stages, which are alternative land use units selection stage, location selection stage, location decision stage and optimization stage. In the model verification, simulation precision of three experiments are more than 70% and the value of Lee-Sallee shape index and Kappa coefficient are acceptable, which shows the effective of urban spatial growth model based on multi-agent.(2) GWLR model can reflect the non-stationary characteristics of geographic data, and has better fitting effect of sampling points compared with the global logistic regression models, which means GWLR model is more suitable for construction land suitability assessment.(3) Government spatial regulation zones of construction land have played a guiding role in the urban spatial growth. Government spatial regulation zones of construction land are classified as 4 types:Constructive Expansion Permitted Zone, Constructive Expansion Conditionally-permitted Zone, Constructive Expansion Restricted Zone and Constructive Expansion Prohibited Zone. In the model verification, the model including the government factor achieves a higher simulation accuracy, which means microscopic individual activities are bound by "top-down" planning. Therefore, the government should design a reasonable government spatial regulation zones and urban development boundary in order to correctly guide the urban spatial growth.(4) Learning function enables the dynamic evaluation of construction suitability of land-use units. Learning function considers the interactions between the agents so that the suitability of land use units around the new construction units can have a continuous improvement. As a result, the land-use simulation results tend to gather and compact. In the model verification, the model including the learning function achieves a higher simulation accuracy, which means learning function can reflect the reality of construction land characterized by clustering.