InSAR和改进支持向量机的沉陷预测模型分析

Analysis of subsidence prediction model based on InSAR and improved support vector machine

针对传统开采沉陷监测方法的缺陷和现有预测模型精度较低的问题,该文提出了一种基于SBAS-InSAR和差分进化混合灰狼优化算法(DEGWO)优化支持向量机回归(SVR)的预测模型,利用2018年10月—2020年3月的44景Sentinel-1A数据对陕西彬长矿区孟村煤矿进行开采沉陷监测,得到该矿区的年平均沉降速率和时间序列累积沉降值。SBAS-InSAR监测结果表明,该矿区年平均沉降速率最高达到了211 mm/a,最大累积形变量达到335 mm。用矿区GPS数据对InSAR处理结果进行验证,拟合效果较好。并将预测结果与传统SVR预测模型以及灰色GM(1,1)模型的预测结果进行对比,DEGWO-SVR模型的绝对误差、相对误差以及均方根误差,均为三者最小。说明了差分进化混合灰狼优化算法能够起到很好的参数优化效果,该方法优化的SVR预测模型能够在矿区开采沉陷预测中得到应用。

Aiming at the shortcomings of traditional mining subsidence monitoring methods and the low prediction accuracy of existing prediction models,this paper proposes an optimization support vector based on SBAS-InSAR(small baseline subset interferometry,SBAS-InSAR)and differential evolution hybrid gray wolf optimization(DEGWO)algorithm to optimize support vector machine regression(SVR).We used 44 Sentinel-1 A data from October 2018 to March 2020 to monitor the mining subsidence of the Mengcun coal mine in Binchang mining area,Shaanxi,and obtained the average annual subsidence rate and time series cumulative subsidence value of the mining area.SBAS-InSAR monitoring results show that the maximum annual average settlement rate of the mining area reaches 211 mm/a,and the maximum cumulative deformation reaches 355 mm.The InSAR processing results were verified with GPS data from the mining area,and the error was found to be within the tolerance range.Finally,the prediction results are compared with the prediction results of the traditional SVR prediction model and the gray GM(1,1)model.The absolute error,relative error and root mean square error of the DEGWO-SVR model are the smallest of the three,which illustrates that the differential evolution hybrid gray wolf optimization algorithm can play a very good parameter optimization effect,and the SVR prediction model optimized by this method can be applied in mining subsidence prediction.