摘要
住宅房地产价格与交通、环境等各类影响因子之间存在着非线性复杂关系,住宅价格的空间自相关性对住宅价格建模也有重要的影响。考虑到住宅价格的空间自相关性,构建了3种空间型BP神经网络模型,并利用遗传算法(GA)进行模型训练。第一种空间型模型的输入层神经元为样本坐标,第二种空间型模型的输入层神经元为空间滞后向量,第三种空间型模型的输入层神经元既包括样本坐标也包括空间滞后向量。以武汉市为例进行实证分析,选取了2010年291个住宅价格样本。实验结果表明,空间型BP神经网络模型的拟合精度优于普通BP神经网络模型及空间滞后模型,其中第三种空间型BP神经网络模型效果最优,输出结果与实际价格相关性达86.69%,均方根误差明显小于其他模型。
This paper built three spatial BP neural network models by combining the BP neural network and spatial autocorrelation. The evaluation of residential real estate price of Wuhan City was used as a case study. Sample point coordinates reflected the overall spatial distribution of the urban residential real estate price, and spatial autocorrelation variables displayed the residential real estate price's local spatial features. Besides, genetic algorithm(GA) was employed to optimize the initial weights and thresholds of the models. 291 residential real estate sample points of Wuhan City in 2010 were selected. The results show that the residential real estate price of Wuhan City exhibits obvious spatial autocorrelation for the value of the Moran's I, is 0.360. The spatial lag model is inferior to BP neural network model in the accuracy of assessing housing prices, and the spatial BP neural network model is superior to the ordinary BP neural network model. The BP neural network model with the combination of spatial autocorrelation variable and sample point coordinates has an optimal accuracy. The correlation between the output and the actual price is 86.69%, and root-mean-square error is minimum. Therefore it can be used in the evaluation of urban residential real estate price.
出处
《地理空间信息》
2017年第2期86-90,共5页
Geospatial Information
基金
国家自然科学基金资助项目(41171312)
