Simulation for evolvement of residential spatial patterns in real scene based on multi-agent

Multi-agent modeling provides a new method for urban research. And it is of theoretical and practical significance to adopt multi-agent modeling, a bottom-up approach, to build up the actual residential decision model, and to research the formation and evolvement of the urban residential spatial patterns. However, many multi-agent models usually describe space as homogeneous space which is beyond expression of real geo-space. In order to simulate the urban residential patterns, some evaluation indices of urban residential suitability in the human-environment science are introduced. Integrating multi-agent model with GIS can provide a heterogeneous and dynamic simulated environment. The model based on multi-agent is constructed as follows: simulated inhabitants wander through the universe to find the optimal place of residence according to constraints as their incomes and preference of residential environment. Evaluation indices, such as traffic accessibility, life convenience and landscape beauty, have been chosen and quantified by GIS. The migration decision of agent depends on the current residential pressure. It is suggested that the probability of leaving its current location increases monotonically with an increase in the residential pressure. Current residential pressure of an agent is estimated by the pressure caused by both basic house price and house environmental price. On the Swarm platform, Java language and component of Swarm-GIS are used to realize the model. The proposed model is applied to simulate the formation of residential spatial patterns in Haizhu district of Guangzhou, Guangdong province. The results of simulation reveal that downtown area is still the first choice of housing location for middle and high income residents in Guangzhou, which shows a different residential mode from Western developed countries. The Spearman rank correlation coefficient of simulated house price and actual house price exceeds 0.6, which indicates that the simulation result can reflect the rules in real practice. Although some uncertainty remains in the model, such as the lack of unified criteria for residential environment evaluation, randomness of house choice and the simplification of the model, the simulation can help us to understand and explore causes and dynamics of residential space patterns, and provide an important analytical tool and simulation method to develop and verify urban theories.