基于机器学习的污染土壤修复变量分类与评价

Classification and Evaluation of Contaminated Soil Remediation Variables Based on Machine Learning

为缩短重金属污染土壤修复周期,研究电动+植物强化修复法机理与规律,建立分类评价指标体系,融合随机森林算法和灰色关联法建立RF-GRA模型,进行变量分类和修复效率评价。指标中电压梯度(0.093)的权重值最高,其次是电流大小、电极材料、土壤湿度、富集植物种类和通电时间等。6个样本的模型评价效率高于实验效率,实验环境A1~A3的相对误差分别为3.35%, 4.12%和6.54%,B1~B3分别为11.19%, 9.48%和8.66%;A1(Cd)和A2(Cu)的修复效果优于B1(Cd)和B2(Cu),A3(Zn)低于B3(Zn);RF-GRA的测试结果优于RF-Random。为土壤修复质量评价奠定了基础。

To reduce the remediation cycle of soil contaminated by heavy meta, the mechanism and law of Electrokinetic&Phyto reinforcement remediation contaminated soil was studied. The classification and evaluation index system and RF-GRA model integrating random forest and grey correlation method were established to classify variables and evaluate remediation efficiency. The highest weight value in index is the voltage gradient(0.093), which is followed successively by current value,electrode material, soil humidity, hyperaccumulators type, and power on time. The weight values of the other variables are lower than the average. The evaluating efficiency of the model is higher than that of experiment for 6 samples. The relative errors of experiment environment A1 〜 A3 are 3.35%, 4.12% and 6.54% and B1 〜 B3 are 11.19%, 9.48% and 8.66%respectively;the remediation effect of A1(Cd)and A2(Cu) are better than B1(Cd) and B2(Cu), while A3(Zn) is lower than B3(Zn). RF-GRA model is better than RF-Random, which lays a foundation for the evaluation of soil remediation schemes.