结合SBAS-IPTA与随机森林算法的开采沉陷监测与预计

Monitoring and prediction of mining subsidence combining SBAS-IPTA and random forest algorithm

针对矿区植被多、易造成失相干的问题,提出了一种将小基线集和干涉点目标分析(small baseline subsat and interferometric point target analyisis,SBAS-IPTA)方法与随机森林(random forest,RF)算法相结合的开采沉陷监测与预计方法。首先提取陕西省桑树坪矿区的沉陷速率及时序累积沉陷值,然后定量分析沉陷规律,最后将沉陷值作为RF的训练样本建立预计函数,并与灰色模型(gray model,GM)算法预计结果进行对比分析。结果表明:通过对该矿区在2007—2011年期间沉陷面积及沉陷值的分析,SBAS-IPTA方法可监测植被多的矿区的地表沉陷;结合GM算法预计的3个测试点的绝对误差和相对误差分别为35~423 mm和7%~88%,而结合RF算法预计的绝对误差和相对误差分别为1~29 mm和0~16%,RF算法预计结果明显优于GM算法预计结果,可将RF算法应用于开采沉陷预计。

Aiming at the problem of a lot of vegetation in the mining area,which is easy to cause loss of coherence,a method of mining subsidence monitoring and prediction was proposed that combined the small baseline subsat and interferometric point target analysis(SBAS-IPTA)method with the random forest(RF)algorithm.This method first extracts the subsidence rate and time-series cumulative subsidence value of the Sangshuping mining area in Shaanxi province,and then quantitatively analyzes the subsidence law,and finally uses the subsidence value as the training sample of RF to establish a prediction function,and compares the prediction results with the gray model(GM)algorithm.The results show that the SBAS-IPTA method can monitor the surface subsidence of the mining area with a lot of vegetation through the analysis of the subsidence area and subsidence value of the mining area from 2007 to 2011.The absolute error and relative error of the three test points predicted by the GM algorithm are 35-423 mm,7%-88%,and the absolute error and relative error predicted by the RF algorithm are 1-29 mm,0-16%respectively,Which indicates that the predicted result of the RF algorithm is significantly better than the predicted result of GM algorithm,and the RF algorithm can be applied to mining subsidence prediction.