以成都市为研究区域,利用城镇建设用地和社会经济统计数据,从因素和时间二维角度出发,分别选用RBF神经网络模型和GM(1,1)模型,在模型比较优选的基础上对成都市城镇建设用地需求进行了预测。结果表明:RBF神经网络模型与GM(1,1)模型相比,前者的平均绝对误差和误差均方根较低,且线性拟合效果更佳,是一种精度较高的预测方法;据预测,近期2015年和远期2020年研究区城镇建设用地需求量将分别达到145 986.10hm2和182 321.26hm2,研究结果既能为具有类似数据"突变"特征的城市开展城镇建设用地需求预测提供方法借鉴,又能为研究区土地资源的可持续利用和土地利用总体规划的定期修编提供决策依据。
With rapid urbanization and industrialization China is facing great urban construction land pressures. The expansion of construction land has also become one of the greatest features of Chinese land-use change. Chengdu is a city under rapid development and transformation in southwest China, however, the area of construction land shows a particular mutation feature. Here we focus on the prediction of urban construction land demand based on comparison and selection between a RBF neural network model and a GM (1, 1) model from two dimensions (factor and time). In order to reach a precise prediction, data about constriction land and social economy is applied. We found that the RBF neural network model has a better performance than the GM (1,1) model: the former has a lower absolute error and root mean square error, and performs better in linear fitting than the latter. By 2015, Chengdu' s construction land demand will reach 145 986.10hm^2 and by 2020 will reach 182 321.26hm^2. With the construction of an urban-rural infrastructure system and Tianfu district, it is estimated that construction land demand for both urban and rural areas in Chengdu will increase rapidly. The findings will provide a reference for urban construction land demand forecasting for city planners faced with similar 'mutation-featured' data. These findings will also be vital to the periodic revision of land-use planning in Chengdu.