基于DNN的煤矿富水区探测反演方法研究

Research on Inversion Method of Coal Mine Water-rich Area Detection Based on DNN

提出了一种基于DNN的煤矿富水区探测反演算法,该算法可以快速准确地实现煤矿富水区二维分布模型的重建。首先,利用时域有限差分方法(FDTD)获得不同分布模型的数值解;随后,依据数据样本搭建网络框架,网络的输入主要为电场分量,输出为相应的模型电导率参数。通过对神经网络进行训练,得到网络的最优系数,随后对富水区分布进行反演预测;结果表明:DNN算法在单个小目标异常体反演中,可以有效克服BP神经网络模型失效的问题,且对于多目标异常体的反演效果更加准确。另外,相同数据集下,DNN的训练耗时与预测耗时也少于BP神经网络。实验结果表明,该算法可以有效提高煤矿富水区探测效率。

A DNN-based coal mine water-rich detection inversion algorithm is proposed,which can quickly and accurately realize the reconstruction of the two-dimensional distribution model of the water-rich area of coal mine.Firstly,the time-domain finite difference method(FDTD)is used to obtain numerical solutions for different distribution models.Then,the network framework is built according to the data samples,the input of the network is mainly the electric field component,and the output is the corresponding model conductivity parameter.By training the neural network,the optimal coefficient of the network is obtained,and the distribution of water-rich areas is inverted and predicted;the results show that the DNN algorithm can effectively overcome the problem of BP neural network model failure in the inversion of a single small target anomaly,and the inversion effect of handling multi-target anomalies is more accurate.In addition,the training time and prediction time of DNN neural network are also less than that of BP neural network under the same dataset.The experimental results show that this algorithm can effectively improve the detection efficiency of water-rich areas of coal mines.