Mining Subsidence Prediction by Combining Support Vector Machine Regression and Interferometric Synthetic Aperture Radar Data

Lichun Sui,Fei Ma,Nan Chen

doi:10.3390/ijgi9060390

Abstract

Mining subsidence is time-dependent and highly nonlinear, especially in the Loess Plateau region in Northwestern China. As a consequence, and mainly in building agglomerations, the structures can be damaged severely during or after underground extraction, with risks to human life. In this paper, we propose an approach based on a combination of a differential interferometric synthetic aperture radar (DInSAR) technique and a support vector machine (SVM) regression algorithm optimized by grid search (GS-SVR) to predict mining subsidence in a timely and cost-efficient manner. We consider five Advanced Land Observing Satellite (ALOS)/Phased Array type L-band Synthetic Aperture Radar (PALSAR) images encompassing the Dafosi coal mine area in Binxian and Changwu counties, Shaanxi Province. The results show that the subsidence predicted by the proposed InSAR and GS-SVR approach is consistent with the Global Positioning System (GPS) measurements. The maximum absolute errors are less than 3.1 cm and the maximum relative errors are less than 14%. The proposed approach combining DInSAR with GS-SVR technology can predict mining subsidence on the Loess Plateau of China with a high level of accuracy. This research may also help to provide disaster warnings.

Highlights

Geo-hazards and structural damage caused by coal mining have attracted much attention in research over the past few years
We propose a novel approach that combines interferometric synthetic aperture radar (InSAR)-derived deformation with an support vector machine (SVM) regression algorithm optimized by grid search for predicting mining-related subsidence
Based on the above experiment and discussion, differential interferometric synthetic aperture radar (DInSAR) technology is highly efficient in mining surface monitoring, and the grid search (GS)‐support vector machine regression (SVR) prediction algorithm is valid for mining subsidence prediction

Summary

Introduction

Geo-hazards and structural damage caused by coal mining have attracted much attention in research over the past few years. The monitoring methods may include measurements obtained by using total station instruments, automatic levels, Global Navigation Satellite Systems (GNSS), and/or terrestrial laser scanners (TLS). These methods establish monitoring stations along the main plane of a deformation basin and measure the observed subsidence rate over various periods. These approaches exhibit shortcomings such as low spatial resolution, high cost, low efficiency, and high labor intensity, besides being a life-threatening activity [3,4]. In the specific case of China’s Loess Plateau, the practical applications of the traditional approaches for modeling mining subsidence are dramatically limited due to the difficulty in establishing these observation lines. The ravines and gullies of the Loess Plateau make establishing monitoring stations highly inconvenient

Methods

Results

Conclusion