基于BP神经网络的城市公园土壤重金属磁学诊断分析——以上海市莘庄公园为例

Magnetic diagnosis of heavy metals in urban park soil based on BP neural network——A case study of Xinzhuang Park in Shanghai

在上海市莘庄公园采集公园绿地表层土壤,结合磁学方法和化学方法,通过地统计学分析土壤重金属、土壤磁学特征的空间变异特征,对基于非线性BP神经网络和偏最小二乘法建立的土壤重金属磁学诊断模型的稳定性和精确性进行了比较分析。结果表明,土壤中Cu、Mn、Pb和Zn含量平均值分别为27.31、651.91、26.05和96.20 mg/kg,皆超过上海土壤重金属背景值,公园绿地土壤重金属存在富集现象。土壤低频磁化率、饱和等温剩磁和非磁滞剩磁磁化率平均值分别为27.39×10^(−8) m^(3)/kg、3480.67×10^(−6) Am^(2)/kg和182.01×10^(−8) m^(3)/kg,也不同程度超出上海土壤背景值,公园绿地土壤存在磁性增强现象。土壤低频磁化率、饱和等温剩磁与Cu、Pb和Zn呈极显著正相关(p<0.01),与Mn呈显著正相关(p<0.05),通过非线性BP神经网络土壤重金属磁学诊断建模的稳定性和精确性均优于偏最小二乘法。

The surface soil samples were collected from Xinzhuang Park in Shanghai.Combining the magnetic method with the chemical method,the spatial variation characteristics of soil heavy metals and soil magnetic characteristics were analyzed by geostatistics.The stability and accuracy of the magnetic diagnosis model for soil heavy metals based on nonlinear BP neural network and partial least square method were investigated.The results showed that the average contents of Cu,Mn,Pb and Zn in soil were 27.31、651.91、26.05 and 96.20 mg/kg,respectively,which were higher than the background value of heavy metal content in the soil of Shanghai with a certain accumulation.The mean value ofχlf,SIRM andχARM were 27.39×10^(−8) m^(3)/kg,3480.67×10^(−6) Am^(2)/kg,and 182.01×10^(−8) m^(3)/kg,which exceeded the background values of Shanghai soil in certain degrees.Soilχlf and SIRM were significantly positively correlated with Cu,Pb and Zn(p<0.01),and positively correlated with Mn(p<0.05).From the comprehensive effect of the model prediction,the fitting effect of BP neural network model was better than that of partial least square method.