结合时序InSAR与优化SVR模型的开采沉陷预计研究

Mining subsidence prediction study combining time-series InSAR and optimised SVR models

针对矿区开采沉陷传统监测手段难以获取长时间序列监测结果以及沉降预测模型精度较低的问题,本文采用SBAS-InSAR技术对矿区进行开采沉陷时序监测,并提出了并提出了一种鲸鱼优化支持向量回归(WOA-SVR)结合时序InSAR的开采沉陷预计模型。选取2021年10月—2023年3月共45景Sentinel-1A影像对贵州省部分矿区进行开采沉陷监测,得到该时段内研究区的年平均沉降速率与累积沉降结果。监测结果表明,矿区内年平均沉降速率最高达到373mm/a,最大累积沉降量达到600mm。将监测结果作为训练集与测试集输入模型进行预测,并与传统SVR模型和GM(1,1)模型预测结果进行对比,发现WOA-SVR最大相对误差均控制在2%以下,RMSE与MAE较传统模型分别最大提升93%与92%,说明WOA算法具有较强的参数寻优能力且WOA-SVR预测模型在矿区沉降预测方面具有良好的应用前景。

In response to the difficulties in obtaining long-term monitoring results and the low accuracy of subsidence prediction models in mining areas,this study employs SBAS-InSAR technology for temporal monitoring of subsidence during mining activities in the region.Additionally,a novel mining-induced subsidence prediction model is proposed,integrating the Whales Optimization Algorithm with Support Vector Machine(WOA-SVR)in conjunction with time-series InSAR.A total of 45 Sentinel-IA images from October 2021 to March 2023 were selected to monitor subsidence in some mining areas in Guizhou Province.The study yielded annual average subsidence rates and cumulative subsidence results for the research area during this period.The monitoring results indicate that the highest annual average subsidence rate within the mining area reaches 373 mm/year,with a maximum cumulative subsidence of 600 mm.These monitoring results were used as input for training and testing the predictive model.When compared to traditional SVR models and GM(1,1)models,the WOA-SVR model demonstrated a maximum relative error of less than 2%.The Root Mean Square Error(RMSE)and Mean Absolute Error(MAE)showed an improvement of up to 93%and 92%,respectively,over traditional models.This suggests that the WOA algorithm possesses strong parameter optimization capabilities»and the WOA-SVR predictive model holds promising applications in subsidence prediction for mining areas.