基于GMS的地层三维结构可视化模型及神经网络预测

 通过收集到的长春市及周边地区各类钻孔资料,运用软件GMS建立了长春及周边地区的三维地层结构可视化模型,与实际地质(地形)情况较为吻合,清晰地反映出长春地区地层结构情况,通过软件还可观察地层任意位置的剖面情况。将神经网络引入其中,当输入钻孔坐标(x,y,z)、地层厚度及地层深度时,能够较为准确地预测出对应地层的地质时代和岩性,采用结构为5-13-5的BP神经网络(单隐含层)预测结果的平均相对误差为11.12%,其最小误差为7.50%、最大误差为15.71%;采用改进后的结构为5-11-7-5的BP神经网络(双隐含层),预测结果的平均相对误差为4.64%,其最小误差为3.63%、最大误差为6.59%,完全满足预测精度要求。

Visualization Model of the Stratum Three-dimensional Structure Based on GMS and the Prediction of the Neural Network

Through the collection of various typed drilling data in the City of Changchun and the surrounding areas, the visualization model of three-dimensional stratum structure in Changchun and the surrounding areas is established by using the software of GMS. The model fits with the actual geology (topography) quite well, clear reflecting the stratum structure of Changchun, by means of the software, the profile situation of any stratum locations could be also observed. The neural network is introduced, by using hole coordinates (x, y, z), the depth of the stratum, and the thickness of the stratum as input, the corresponding geological age and the lithology (in Chinese and English) is able to be accurately predicted. Using the 5-13-5 structure of the BP neural network (single hidden layer), the average relative error of prediction is 11.12% (among them, minimum error is 7.50%, maximum error is 15.71%); using the improved 5-11-7-5 structure of the BP neural network (two hidden layers), the average relative prediction error is 4.64% (among them, minimum error is 3.63%, maximum error is 6.59%), the requirement for forecast accuracy is fully met.