Abstract: The remediation process of heavy metal contaminated soil in tailings ponds is complicated and slow. To reduce the remediation cycle and accelerate the development of intelligence and refinement of soil remediation technology, the mechanism and law of the phyto-electrokinetic method were used, and the evaluation index system of soil remediation variables was established flexibly. A new GRA-RF model for soil simulated remediation was established by combining a random forest classifier and GRA simulation algorithm, and then the simulated remediation and experimental repair were compared. The results showed that:1) the highest grade value in remediation variables was the voltage gradient (0.092), followed successively by current value (0.078), soil humidity (0.074), power-on time (0.069), dynamic pH level of the soil (0.066), electrode space (0.063), electrode arrangement way (0.06), additive type (0.057), and electrode material (0.053). The weight values of the other variables were lower than the average. The size of the remediation variables would be changed following the importance order of the variables until the optimal repair effect was obtained; 2) the remediation efficiency of the simulation was higher than that of the experiment for 8 samples. The relative errors of remediation environment, for A1 to A4 were 2.22%, 4.72%, 8.75%, 3.89% and from B1 to B4 were 9.91%, 8.28%, 6.74% and 5.63% respectively. The remediation effect of Cd, Cu and Pb in environment A were better than Cd, Cu and Pb, while Zn had an opposite remediation effect; 3) by comparing the error rate of the algorithm, it was found that GRA-RF model was better than Random-RF. This method more precisely simulated the process of enhanced remediation of contaminated soil. The efficiency of soil remediation was improved by optimizing the remediation variables, it laid a foundation for making and improving soil remediation schemes.