BP神经网络在土壤重金属污染分析中的应用

Application of BP neural network on the analysis of heavy metal contamination in soil

以湖南株洲市区中西部为研究区域,获取该区域35个土壤样本和多光谱数据,基于多元线性回归(MLR)、偏最小二乘回归(PLS)、BP神经网络回归模型(BP),分别建立土壤重金属(Cr、Cu、Ni)含量的反演模型,并对模型预测效果进行检验。建模与预测综合效果:BP模型>PLS模型>MLR模型,BP神经网络回归模型的效果远远好于其他2组,尤其适合分析具有非明确关系的2组数据。其中,Cr元素回归模型为最佳拟合模型,建模和预测R^2分别为0.917 4、0.811 0,建模均方根误差和预测均方根误差分别为8.269 3、16.870 7,说明基于多光谱数据反演土壤重金属含量有一定的可行性。

Targeted at the region in the midwest of urban Zhuzhou area,35 soil samples and multi-spectral data are obtained in this research. Then,based on these data,three approaches are adopted to construct the model of heavy metal（ Cr,Cu and Ni） contamination in soil： multiple linear regression（ MLR）,partial least squares regression（ PLS） and BP neural network（ BP）. Moreover,some indexes（ R2,RMSEC and RMSEP） are defined in this paper to evaluate the effectiveness of these approaches. The results show that,BP comes first followed by PLS and MLR in terms of the model＇s effectiveness. Besides,indexes of BP are much better than those of the other two approaches and this approach is especially suitable for the analysis of data with uncertainty relation. Among the three elements analyzed in the paper,the regression model of Cr is best in terms of imitative effect. For Cr,R-2 of modelling and prediction are 0. 917 4 and 0. 811 0 respectively,with RMSEC and RMSEP being 8. 269 3 and 16. 870 7 respectively. Meanwhile,to some extent,it is reasonable to construct quantitative inversion of heavy metal contamination in soil using multi-spectral data.