基于RBF-BP神经网络的人居环境自然适宜性评价

Evaluation of natural suitability for human settlement based on RBF-BP neural network

为了有效解决人居环境自然适宜性评价中,评价指标权重难以科学确定、评价结果精度低等问题。以东莞市为例,构建了地形、气候、水文和植被等4要素11个指标的人居环境自然适宜性评价体系;综合利用BP算法良好的泛化性能和RBF算法学习收敛速度快,以及克服局部极小值问题等优点,建立RBF-BP神经网络模型,用以求出评价单元各因子间的相关性,将训练结果进一步分析和处理求出评价指标的权重;最后将权重应用于人居环境指数(HEI)评价模型,完成了东莞市人居环境适宜性评价,并将适宜性划分了六个等级。结果表明,评价结果与其实际人居状况大致吻合;同时也说明了此模型收率速度高、学习速度快、拟合精度较高和泛化能力更强,能较准确、科学地评价人居环境自然适宜性。

In order to solve the problems of the difficulty of scientifically determining the weight of the evaluation index and the low accuracy of the evaluation result in the evaluation of natural suitability for human settlement.Taking Dongguan as an example,the evaluation system of natural suitability for human settlement environment was constructed with 11 indexes of 4 elements which including topography,climate,hydrology and vegetation.Comprehensive use of the good generalization performance of the BP algorithm and the advantages of RBF algorithm which has fast convergence and overcoming local minimum problem,establish the RBF-BP neural network model to calculate the correlation between the factors of the evaluation unit,and further analyze and process the training results to obtain the weight of the evaluation index;Finally,the weight was applied to the human environment index(HEI)evaluation model to evaluate the s natural suitability for human settlement of Dongguan,and the suitability of Dongguan was divided into six levels.The results show that the evaluation results are roughly consistent with the actual living conditions,at the same time,the results show that this model has many advantages,such as high yield rate,fast learning speed,high fitting accuracy and strong generalization ability,which can accurately and scientifically evaluate the natural suitability for human settlements.