基于GSO-GPR算法的岩层移动角预测模型及其应用

为快速准确获取金属矿地下开采岩层移动角，提出基于群搜索优化（GSO）算法的改进高斯过程回归（GPR）理论。以矿体上下盘围岩性质、上下盘围岩稳固程度、地下水情况、矿体走向长度、矿体倾角、开采厚度及深度9个影响因素作为判别指标，结合35组实测数据建立金属矿充填开采岩层移动角学习预测模型。将该模型应用于三山岛金矿岩层移动角预测，利用UDEC数值模拟结果进行对比验证，同时分析三山岛金矿海下开采对上盘竖井的影响。研究结果表明：（1）GSO-GPR预测模型对地下开采岩层移动角的预测效果良好，预测精度在5%以内；（2）三山岛金矿上、下盘岩层移动角分别为72°和68°；（3）当前矿山开采未对竖井造成影响，但随着开采深入至-658 m水平，竖井将进入塌陷范围。

GSO-GPR Model for Strata Displacement Angle Predicting and Its Application

The strata displacement angle of underground metal mine is the key parameter for analyzing rock movement caused by underground mining.It is usually used in the design of security pillars， the delineation of dangerous moving boundaries， and the delineation of surface protective buildings.Therefore， to quickly and accurately obtain the strata displacement angle of underground metal mine is of great significance for the safe and efficient mining of underground metal mines.The Gaussian Process （GP） is a new machine learning technology that was developed on the basis of strict statistical foundation.It has outstanding advantages in dealing with complex classification and regression problems such as high dimensionality， small sample， and nonlinear.However， the optimization effect of its performance parameters has the disadvantages of strong dependence on initial value， difficulty in determining the number of iterations， and local optimization.Group Search Optimizer （GSO） algorithm draws on the behavior of group behavior and has strong optimization search ability.To quickly and accurately obtain the strata displacement angle of underground metal mine， the GSO was chosen to take the place of conjugate gradient method to search for optimal hyper parameters， therefore， an improved Gaussian Process Regression （GPR） theory based on the Group Search Optimization （GSO） algorithm was proposed.According to the actual situation of underground mining， the influence factors selected by relevant scholars and relevant national standards， nine influence factors which including characteristics and stability of surrounding rock on upper-wall and foot-wall， geological structure， underground water， ore-body length along its trend， ore-body dip angle， mining thickness and depth were chosen as the evaluation indexes. Combining with 35 groups of measured data to establish, the learning model was established for predicting displacement angle of strata in metal mine adopting backfilling methods. Sanshandao gold mine is the only large-scale submarine gold mine in the world.The range of strata displacement and deformation gradually expands with the deepening of mining.It is likely to affect the stability of the coastal shaft in the later stage.The improved Gaussian Process Regression （GPR） theory model was then applied to predict the strata displacement angle of Sanshandao gold mine， and the results was compared to the simulation results by UDEC.The FLAC software was used to establish the underground mining geometric model.The UDEC software was used to simulate the rock movement caused by submarine mining， and then the strata displacement angle of the upper rock stratum of the Sanshandao gold mine was calculated and verified.Meanwhile， the influence of mining on the upper-wall shaft was analyzed.The results show that：（1） The GSO-GPR model has a significant effect on the prediction of strata displacement angle of underground mines with the prediction accuracy within 5\mathrm{\%}；（2） The strata displacement angles of upper-wall and foot-wall of Sanshandao gold mine are 72° and 68° respectively, through the UDEC numerical simulation， the law of rock stratum subsidence in the mining is analyzed， and the moving angle of the upper plate is stable at around 72° after the mining in the -600 m level. （3）Applying the strata displacement angle of the upper rock to the division of the subsidence range，the current mine mining has no effect on the shaft， but as the mining goes deeper， the shaft will fall into the collapse range when mining the level of -658 m.