基于支持向量机的城市土壤重金属污染评价

以太原市城区周边的80个土壤样品为研究对象，测定了土壤中Ni、Cr、As、Cu、Zn、Pb、Cd和Hg8种重金属的含量，运用支持向量机模型进行土壤重金属污染评价，并与 Hakanson 指数法和内梅罗综合污染指数法的评价结果进行对比,探讨支持向量机模型在土壤重金属污染评价中的应用。结果表明，（1）重金属元素的变异系数由高到低顺序分别为：Hg＞Pb＞As＞Cd＞Cu＞Ni＞Zn＞Cr。Hg含量在0.02-0.39 mg·kg^-1之间，变异系数为0.648，最大值为最小值的19.5倍；而土壤中Pb的含量在17.4-86 mg·kg^-1之间，变异系数为0.409；即使变异系数最小的Cr元素的最大值为109 mg·kg^-1，最小值为54.7 mg·kg^-1，变化范围也到达了54.3 mg·kg^-1，可见太原市土壤中各元素含量的变异很大。各元素的含量的平均值除Ni以外均大于太原市的元素背景值，但都在国家土壤质量质量标准的二级标准值之下。（2）内梅罗综合污染指数法、Hakanson 指数法和支持向量机的评价结果中，评价等级为2A、2B、2C的样品数分别为：41、47和45，37、29和33，2、4和2，评价结果相差不大，内梅罗综合污染指数法与支持向量机方法的评价结果相同率为70%，Hakanson 指数法和支持向量机的评价结果相同率为65%。（3）对评价结果有差异的样品进一步分析可知，支持向量机方法的评价结果更为准确。相比较内梅罗综合污染指数法和Hakanson指数法而言，支持向量机降低了人的主观判断对评价结果的影响，在进行综合评价的时候能考虑到各因素的综合影响，使评价结果更接近真实情况，有较为严格的数学基础，泛化能力好，在土壤重金属污染评价中有广泛的应用前景。

Assessment of Urban Soil Heavy Metal Pollution Based on Support Vector Machine

In order to evaluate urban soil heavy metal pollution statement of Taiyuan City, we measured soil nickel（Ni）, chromium（Cr）, arsenic（As）, copper（Cu）, zinc（Zn）, lead（Pb）, chromium（Cd）, and mercury（Hg） contents of 80 soil samples around the research area. Moreover, soil heavy metal pollution was assessed using support vector machines （SVM）. By comparing to Hakanson index and Nemerow pollution index, we discussed the applicability of support vector machines in soil heavy metal pollution evaluation. Results showed that：（1） The variation coefficient of Hg and Pb were 0.648 and 0.409 respectively. Maximum variation coefficients of Hg and Pb were 19.5 and 4.94 times of the minimum. Contents of Hg and Pb were highly heterogeneous in spatial distribution in Taiyuan. Except Ni element, average contents of each element were higher than their corresponding background values in Taiyuan, but are under the secondary standard value. （2） The evaluation results of Nemerow comprehensive pollution index, Hakanson potential ecological risk index, and SVM methods have no great difference. The evaluation results of Nemerow comprehensive pollution index and SVM was 70%identical. The evaluation results of Hakanson potential ecological risk index and SVM was 65%identical. （3） Further analysis of samples with different evaluation results showed that the evaluation results of SVM were more accurate. Comparing to Nemerow comprehensive pollution index and Hakanson potential ecological risk index, SVM could reduce the impact of man＇s subjective judgment to the evaluation result. Meanwhile, SVM could produce a more comprehensive evaluation result involving various factors. Evaluation results were closer to the truth. Besides, SVM had more strict mathematical foundation and good generalization ability. It could be widely applied in the evaluation of soil heavy metal pollution.