基于BP-ANN模型的农村土地生态安全评价及影响因素分析——以重庆市丰都县为例

发布时间：2018-10-17 13:41

【摘要】：针对重庆市丰都县土地生态敏感特性和区域实际情况,将土地生态安全评价指标值作为输入,土地生态安全评价得分作为输出,基于多层反向传播人工神经网络(BP-ANN)构建了具有15个隐含层神经元的3层土地生态安全评价模型,通过Matlab R2012b随机选取2013年317个总样本中的200个样本,将其中的70%,30%分别作为网络的训练样本和检验样本用于训练及检验BP-ANN,将剩余117个样本测试网络,再将2014年317个村的相关指标值代入人工神经网络通过模式识别进行评价;在此基础上分析丰都县农村土地生态安全状况的空间分异现象,并使用因子分析法结合多元线性回归分析法研究影响丰都县农村土地生态安全的因素。结果表明:(1)117个测试样本中相对误差小于5%的样本占98.21%,BP-ANN模型精度良好;(2)2014年丰都县各村土地生态安全评分为40.47~55.73,平均得分49.19,土地生态安全状况处于中等偏下水平,空间上土地生态安全评分全局Morans′I指数为0.34,空间呈自相关性,县域内西北部的农村土地生态状况较好;(3)影响丰都县农村土地生态安全的因素按照影响程度大小依次为:植被生物条件、土壤条件、景观多样性、生态建设与发展协调程度、降水条件、水域条件。BP-ANN模型一定程度上克服了传统评价过程中主观因素导致的误差,精确的评价结果对区域土地生态保护提供指导。
[Abstract]:According to the sensitive characteristics of land ecology in Fengdu County of Chongqing and the actual situation of the region, the evaluation index value of land ecological security is taken as input and the evaluation score of land ecological security is taken as output. Based on multilayer back-propagation artificial neural network (BP-ANN), a three-layer land ecological security evaluation model with 15 hidden layer neurons was constructed. 200 of the 317 total samples in 2013 were randomly selected by Matlab R2012b. 70% of them were used as network training samples and test samples respectively for training and testing BP-ANN,. The remaining 117 samples were used to test the network, and then the relative index values of 317 villages in 2014 were replaced by artificial neural networks for evaluation by pattern recognition. On this basis, the spatial differentiation of the ecological security of rural land in Fengdu County is analyzed, and the factors influencing the ecological security of rural land in Fengdu County are studied by factor analysis combined with multiple linear regression analysis. The results showed that: (1) 98.21% of 117 samples with relative error less than 5% had good accuracy of BP-ANN model, (2) the land ecological security score of Fengdu County in 2014 was 40.47 (55.73), with an average score of 49.19, and the land ecological security situation was at a moderate and lower level. The global Morans'I index of land ecological security score is 0. 34, which is autocorrelation in space. (3) the factors influencing the ecological security of rural land in Fengdu County are: vegetation biological condition, soil condition, landscape diversity, ecological construction and development coordination. To some extent, the BP-ANN model overcomes the errors caused by the subjective factors in the traditional evaluation process, and the accurate evaluation results provide guidance for the regional land ecological protection.
【作者单位】：西南大学地理科学学院;
【基金】：“西南山区生态敏感区典型区域土地生态状况信息提取与精细评估”项目(201315106397)
【分类号】：F301.24

【相似文献】