关键词: 元胞自动机(CA), 遗传算法(GA), 城市模拟

Abstract: This paper presents a new method to calibrate urban cellular automata (CA) using genetic algorithms(GA).The GA is used to find the optimal parameter values so that CA models can simulate urban expansion in a more realistic way. Traditional multi-criterion evaluation (MCE) and logistic methods have limitations for deriving the transition rules of CA models. The variables should be independent so that the parameter values (coefficients) can be properly estimated by regression analysis. This assumption is not true in most situations. The limitations can be overcome by using GA to estimate these parameter values for these correlated variables. When calibrating urban cellular automata with GA, the parameters of CA models are set to the chromosomes in GA program.The real number encoding way is used to encode chromosomes. The fitness function is defined with mean square error between simulated and actual urban forms.The initial population is set to be 50 randomly. And crossover probability is set to be 0.9, and mutation probability is set to be 0.01. The elitist selection is used to heredity the better individual.If the fitness does not change in the past 50 generations, the genetic procedure will be finished. After properly encoding the chromosomes, the optimal parameter values are automatically found by the evolutionary approach. This method is applied to the simulation of urban expansion in Dongguan, a fast developing city in the Pearl River Delta in South China. The model is able to simulate urban development in 1988-2004 by using the training data from remote sensing data. The analysis indicates that the proposed model can produce better simulation results than MCE-based CA models and logistic calibrated CA models. Moreover, the parameter values can be used to explain the relationships between spatial variables and urban development.

Key words: cellular automata, genetic algorithms, urban expansion, Pearl River Delta