基于DBN的隧道地质超前预报自动解释算法

发布时间：2019-03-22 11:04

【摘要】：针对隧道地质超前预报过程中,探地雷达(ground penetrating radar,GPR)线测图解释仅依靠专家经验,且存在准确率不高的问题,通过对GPR成像原理和隧道地质特性的研究,以及对深度置信网络(deep belief networks,DBN)计算复杂度的分析,提出一种改进的基于压缩感知和DBN的GPR线测图分类解释模型。该模型首先利用压缩感知技术对原始GPR线测图进行压缩处理,通过选择图像压缩比得到合理的压缩图像;然后将压缩后的图像送入DBN模型进行分类,根据分类结果对原始GPR线测图进行解释;最后利用广西六宜(六寨—宜州)高速公路隧道实测数据对模型的有效性进行验证,试验数据共20 000幅GPR图像,包括6种隧道地质类型,其中15 000幅图像作为训练样本集,5 000幅图像作为测试样本集。研究结果表明:当GPR线测图压缩比为256,反向微调数据为1 000幅图像,DBN模型迭代次数为30时,模型对测试数据中6类探地雷达线测图的分类准确率达100%,单次训练时间降低为原DBN模型的8%左右;大量仿真试验发现GPR线测图的合理图像压缩比区间为64~1 024,在此区间压缩的图像能最大限度地降低图像维度并且保留原始图像信息。该模型具有解释准确率高、训练速度快等优点,可为制定隧道施工和开挖计划提供合理依据。
[Abstract]:In the process of ahead prediction of tunnel geology, the interpretation of ground penetrating radar (ground penetrating radar,GPR) line mapping only depends on the experience of experts, and the accuracy is not high. This paper studies the principle of GPR imaging and the geological characteristics of tunnel. Based on the analysis of computing complexity of depth confidence network (deep belief networks,DBN), an improved classification and interpretation model of GPR line mapping based on compressed sensing and DBN is proposed. Firstly, the compression sensing technology is used to compress the original GPR line map, and a reasonable compression image is obtained by selecting the compression ratio of the image. Then the compressed images are put into the DBN model for classification, and the original GPR line maps are interpreted according to the classification results. Finally, the validity of the model is verified by using the measured tunnel data of Liuyi (Liuzhai-Yizhou) Expressway in Guangxi. The experimental data are 20,000 GPR images, including 6 tunnel geological types, of which 15,000 images are used as training sample set. Five thousand images were used as a test sample set. The results show that when the compression ratio of GPR line mapping is 256, the reverse fine-tuning data is 1,000 images, and the iterative times of DBN model is 30, the classification accuracy of GPR line map of 6 kinds of GPR lines in test data is 100%, and the classification accuracy of GPR line mapping is 100% when the number of iterations of GPR model is 30. The single training time was reduced to about 8% of the original DBN model. A large number of simulation experiments show that the reasonable image compression ratio interval of GPR line mapping is 64 ~ 1024, in which the compressed image can minimize the dimension of the image and retain the original image information. The model has the advantages of high interpretation accuracy and fast training speed. It can provide a reasonable basis for making tunnel construction and excavation plan.
【作者单位】：西北工业大学航海学院;长安大学信息工程学院;西安公路研究院;
【基金】：国家自然科学基金项目(61271143)
【分类号】：U452.11

【相似文献】