基于神经网络的元胞自动机与土地利用演化模拟——以广州市白云区为例

城市发展过程中存在多种土地利用类型的相互转换,掌握其演化规律有助于制定出合理的土地利用规划。传统元胞自动机(CA)在模拟城市扩张过程时,多种土地利用类型间的转换十分复杂,往往难以获得转换规则。本文利用神经网络构建了多类型演化的CA模型;从城市演化的历史数据中进行学习,挖掘出控制土地利用方式转变的空间要素权重,利用广州市白云区2005—2007年间的土地利用历史演化数据训练神经网络后,对2009年研究区的土地利用结构进行了模拟。对比同期的真实土地利用格局,模拟结果的平均精度达到77.65%。

The Simulation of Cell Automaton and Land Use Evolution Based on Neural Network:Taking Baiyun District of Guangzhou as a Case Study

The process of urban growth involves various types of land use changes,the understanding of whose law of evolution helps to make a rational plan of land utilization.The conversion of land utilization pattern of conventional cell automaton in the simulation of urban development is usually difficult and the law of conversion thereof is hard to acquire.This paper sets up a model of conversion of various types of land utilization based on neural network,discovers the weighting of spatial elements which controls the evolution of land utilization pattern by analyzing historical data of urban development,and simulates the land utilization pattern of the studied area of 2009 by training the neural network using the historical land utilization evolution data from 2005 to 2007.The average accuracy of the simulation reaches 77.5% compared with the actual land utilization of the same period.