开采沉陷岩体力学参数反演的BP神经网络方法

Rock Mass Mechanical Parameters Inversion of Mining Subsidence Using BP Neural Network

为获得可靠的开采沉陷数值模拟岩体力学参数,将覆岩按照不同的移动破坏特征分为冒落带、裂隙带、弯曲带和松散层带4部分。采用大量数值模拟结果进行BP神经网络训练,建立模拟地表移动与覆岩力学参数间的关系。以淮北某矿为例,根据正交设计思想进行了50次FLAC3D数值模拟实验。采用实验结果对BP神经网络模型进行训练,获得以地表下沉值为输入、以各层力学参数为输出的BP神经网络模型。采用实测地表下沉值进行力学参数反分析。以反演得到的力学参数进行数值计算,地表下沉数值模拟结果与实测结果吻合,表明所获得的力学参数正确可靠。采用BP神经网络和正交实验方法进行力学参数的正演反分析,可以用较少的实验次数建立地表移动变形量和岩层力学参数之间的非线性映射,获得与实测结果吻合的数值模拟结果,为开采沉陷数值分析中力学参数的选取提供了依据。

In order to obtain reliable rock mass mechanical parameters of mining subsidence in numerical simulation,the overburden rock is divided into four parts,namely,caving zone,fracture zone,curving zone and loose layer. Relationship between simulated surface movement and overburden mechanical parameters was built by BP neural network trained with a large number of numerical simulation results. A mining area in Huaibei mine was shown as an example,50 numerical simulation models designed according to the orthogonal design were done with FLAC3D. Trained with experimental results,a BP neural network model was established with surface subsidence as input and mechanical parameters as output. Mechanical parameters back analysis was done with field measured surface subsidence. Numerical simulation model was built with back analyzed mechanical parameters and the simulated mining subsidence results were fit for the measured,which show that the back analysis results are reliable. Nonlinear mapping relationship between simulated surface movement and overburden mechanical parameters can be built by BP neural network and orthogonal design with less experimental attempts,simulation results which fit for the measured can be obtained. The research provides a basis for selection of mechanical parameters in mining subsidence numerical simulation.