Extraction algorithm of mining subsidence information on water area based on support vector machine

Information on subsidence in a water area caused by mining has a great value on the research of the mining subsidence law of a mining area with a high groundwater level in Eastern China. Based on the measured data of the subsidence area without water, the data extraction of subsidence in a water area is studied in this paper, with a support vector machine, as subsidence in such an area is difficult to measure. Research shows that the training sample number and dimension should be strengthened by increasing the measuring times or using the interpolation method to obtain ideal prediction accuracy. The ε-Support Vector Regression model with three parameters optimized by the genetic algorithm or the particle swarm optimization algorithm is suitable to extract subsidence information in a water area caused by mining, and the algorithm is accomplished on Matlab. Data analysis showed that when the water is deeper than 1.8 m and the distance is over 60 m from the measured points, the prediction error of test samples will exceed 10 % out of all measured results, meaning that practicability is relatively poor; while water depth is <0.8 m or the distance is lower than 60 m from the measured points, the prediction error of test samples will be calculated to <5 % of the measured results, the prediction results can be used.