基于神经网络的土壤重金属预测及生态风险评价

发布时间：2018-02-26 00:27

本文关键词： RBF神经网络重金属预测生态风险　出处：《长江流域资源与环境》2017年04期 　论文类型：期刊论文

【摘要】：采用单隐层RBF神经网络模型预测土壤重金属Cr、As、Ni、Pb、Zn 5种元素的含量,实测35组数据做为训练数据,另用6组做验证数据,该模型是以利用采样的10组数据预测其后的连续5组数据,输入层的神经元个数是10,输出层是5,隐含层的传递函数为径向基函数radbas,输出层的传递函数为线性函数Purelin,其结果表明:采用RBF神经网络模型预测有较高的精度。通过多元统计分析采样样品与预测样品,研究区域As、Ni、Zn的均值超过了上海市土壤环境背景值,As元素达到高度变异,Pb、Zn、Ni 3种元素达到中度变异。通过因子分析,前2个因子基本包含了全部元素变量的主要信息,第1因子中载荷最高是元素Ni(0.946),第2因子中则为元素As(0.930)。通过潜在生态风险指数评价,研究区域整体呈轻度生态风险水平。采用RBF神经网络模型可以降低采样分析成本,更好的评价区域土壤重金属的生态风险。
[Abstract]:The single hidden layer RBF neural network model was used to predict the content of 5 elements of soil heavy metal CrAsN NiNiPbN Zn, 35 groups of measured data were used as training data, and 6 groups of data were used as validation data. The model used 10 groups of sampled data to predict 5 successive groups of data. The number of neurons in the input layer is 10, the output layer is 5, the transfer function of the hidden layer is Radbas-based radial function, and the transfer function of the output layer is the linear function Purelin. The results show that the prediction accuracy is high by using the RBF neural network model. Multivariate statistical analysis of sample and prediction sample, In the study area, the average value of As-NiN Zn exceeded the background value of soil environment in Shanghai, and the elements of as reached a high variability and reached a moderate variation. By factor analysis, the first two factors basically contained the main information of all element variables. In the first factor, the highest load was element Nijiu 0.946, and in the second factor, element As0.930. Through the evaluation of potential ecological risk index, the whole study area presented a mild ecological risk level. The cost of sampling and analysis could be reduced by using RBF neural network model. Better evaluate the ecological risk of heavy metals in regional soil.
【作者单位】：昆明理工大学现代农业工程学院;昆明理工大学建筑与城市规划学院;
【基金】：国家自然科学基金项目(20264002) 云南省科技厅基金项目(2011FB032)~~
【分类号】：X53;X825

【相似文献】