Integration of D-InSAR technology and PSO-SVR algorithm for time series monitoring and dynamic prediction of coal mining subsidence

Abstract

Subsidence of the ground surface caused by coal mining is a serious environmental problem in many countries. Therefore, an effective monitoring and prediction system must be established in coal mining areas to protect nearby property and the surrounding environment. In this paper, a model is proposed that integrates differential interferometry synthetic aperture radar (D-InSAR) technology and the support vector regression (SVR) algorithm to monitor and dynamically predict mining subsidence. D-InSAR technology is first used to monitor the range of influence and the development trend of mining subsidence, thus obtaining the law of surface subsidence. Based on the monitoring results obtained by D-InSAR technology, the SVR algorithm is used to describe the nonlinear function correlativity between the monitored data and future subsidence. As the performance of the SVR algorithm depends largely on the choice of relevant parameters, the particle swarm optimisation (PSO) algorithm is introduced to select the optimal parameters for the SVR algorithm. Finally, a method of rolling prediction based on the optimised SVR parameters is adopted to update the training and learning samples of SVR, thus allowing the algorithm to use the latest monitored data to dynamically predict future mining subsidence. To verify the applicability of the proposed methodology, it was applied to a coal mining area in Neimeng, China, where thirteen TerraSAR-X images were acquired from 21 November 2012 to 2 April 2013. The experimental results show that the monitoring results very accurately reflect the range of influence and the trend in the development of mining subsidence and also that the PSO-SVR algorithm provides high-accuracy prediction results with a maximum absolute error (MAE) of 29 mm and a maximum relative error (MRE) of 6·5%, thus demonstrating the accuracy and feasibility of the proposed model.