現在位置首頁 > 博碩士論文 > 詳目
  • 同意授權
論文中文名稱:利用深度學習探勘地物結構及異常體 [以論文名稱查詢館藏系統]
論文英文名稱:Deep Learning Application on Geophysical Structure and Abnormal Body [以論文名稱查詢館藏系統]
院校名稱:臺北科技大學
學院名稱:工程學院
系所名稱:土木工程系土木與防災碩士班
畢業學年度:106
畢業學期:第二學期
出版年度:107
中文姓名:鄭宇廷
英文姓名:Yu-Ting Cheng
研究生學號:105428051
學位類別:碩士
語文別:中文
口試日期:2018/07/18
論文頁數:106
指導教授中文名:楊元森
指導教授英文名:Yuan-Sen Yang
口試委員中文名:楊元森;張哲豪;陳建志
中文關鍵詞:地物構造探勘深度學習地物分類問題
英文關鍵詞:geophysical investigationdeep learninggeophysics classification
論文中文摘要:許多的重大工程、如地下大型工程、探測能源等,都需要先進行地物構造與異常體探勘,來進行行前規劃與可行性研判。地底構造無法直接得知,需要以電探、震測、磁力與重力常數等方式進行量測,並且以逆運算的方式進行構造的還原。但是真實情況是,地物構造與異常體探勘的分析,在工程上會遇到非常多的問題。三維(3D)的逆運算分析,時常會有收斂性與電腦運算量的問題。可能是數天至數週,而且結果還有可能不夠精確或是不合理。這樣的處理方式,造成工程規劃上有非常多的問題。
本研究利用深度學習(deep learning),一個近年非常新興的機器學習(machine learning)技術,來初步嘗試深度學習在處理地物結構探勘問題的可行性。我們的目標是利用這項技術,來為傳統三維逆運算分析的問題找到新的方法。我們將地物結構的逆運算問題,轉換為分類問題,並導入深度學習技術。以長遠的發展來看,透過深度學習在地物結構探勘的分析,有潛力能夠協助大型工程規劃的能力,並進一步提升我國在地球物理工程應用領域的技術。
論文英文摘要:Geophysical investigation is important project. Traditionally, geophysical investigation can be done by electromagnetic, seismic, magnetic, and gravitational and measuring the response at the Earth surface. Engineers execute practical inverse operation of 3-dimensional geophysical investigation will encounter difficulties such as unreasonable results or numerical divergence even after waiting days or weeks of computing time possibly because of ill-posed characteristics of the geophysical problems.
This research employs rapidly developing deep learning technique and applies it on geophysical investigation. We consider geophysical investigation as a classification problem, and verify the feasibility and performance of deep learning application. The result of this research indicates the potential of deep learning on improving planning capability and technology of large-scale projects in the geophysical engineering.
論文目次:摘要 i
ABSTRACT ii
誌謝 iii
目錄 iv
表目錄 vi
圖目錄 vii
第一章 緒論 1
1.1 研究動機 1
1.2 研究目的 1
1.3 文獻回顧 2
1.4 論文架構 3
1.5 研究範圍與方法 3
第二章 地層 5
2.1 地物探勘 5
2.2 順向運算 6
2.3 地層種類探討 7
2.4 參考的地層種類 7
2.5 地層設計 10
2.6 地層磁感強度圖 20
第三章 深度學習 31
3.1 基本介紹 31
3.2 神經網路 31
3.2.1 活化函數 35
3.2.2 損失函數 37
3.2.3 最佳化 38
3.3 神經網路細部設定 40
3.3.1 權重參數初始值 41
3.3.2 Batch Normalization 43
3.3.3 Overfitting 44
3.3.4 Dropout 45
3.3.5 Batch 46
3.4 卷積神經網路 47
3.4.1 卷積層 48
3.4.2 池化層 52
第四章 訓練神經網路 54
4.1 前言 54
4.2 產生訓練資料 54
4.3 設計與訓練卷積神經網路 64
4.4 結論 89
第五章 細部討論 100
5.1 關於訓練結果 100
5.2 關於地層 100
5.3 關於神經網路 101
5.4 關於訓練資料 101
5.5 往後的發展 102
第六章 結論與未來展望 103
6.1 結論 103
6.2 未來展望 104
參考文獻 105
論文參考文獻:1. P. Guillen, G. Larrazabal, G. González, D. Boumber, R. Vilalta , “Supervised learning to detect salt body”, SEG Technical ProgramExpanded Abstracts, 2015, pp. 1826-1829.
2. V. Addison, Artificial Intelligence Takes Shape in Oil, Gas Sector. Web site: https://www.epmag.com/artificial-intelligencetakes-shape-oil-gas-sector-846041#p=full, April, 2016.
3. M. Araya-Polo, T. Dahlke, C. Frogner, C. Y. Zhang, T. Poggio and D. Hohl, “Automated fault detection without seismic processing,” THE LEADING EDGE, March, 2017.
4. 陳建志, Multi_prism_CCC程式提供(未公開發表文獻), 2016.
5. B. B. Bougher, Machine learning applications to geophysical data analysis, Master of Science, The University of British Columbia, City of Vancouver, 2016.
6. Keras, Why use Keras?, Keras official website: https://keras.io/why-use-keras/, 2017.
7. M. A. Sandjideh, Announcing TensorFlow 1.0, Web site: https://developers.googleblog.com/2017/02/announcing-tensorflow-10.html, 2017.
8. Theano, Theano 1.0 Release, Web site: http://deeplearning.net/software/theano/, 2017.
9. Kavukcuoglu , DeepMind moves to TensorFlow, Google Research Blog, Web site: https://research.googleblog.com/2016/04/deepmind-movesto-tensorflow.html, 2016.
10. NOAA, Web site: https://www.ngdc.noaa.gov/geomag-web/#declination, 2018
11. 陳文山,臺灣地質概論,臺北 : 社團法人中華民國地質學會出版,2016。
12. 人工神經網路,Web site: https://zh.wikipedia.org/wiki/%E4%BA%BA%E5%B7%A5%E7%A5%9E%E7%BB%8F%E7%BD%91%E7%BB%9C, 2018.
13. 神經網路的運作原理, Web site: https://brohrer.mcknote.com/zh-Hant/how_machine_learning_works/how_neural_networks_work.html, March, 2017.
14. Tensorflow, MNIST For ML Beginners, Web site: https://www.tensorflow.org/versions/r1.1/get_started/mnist/beginners#evaluating_our_model, June, 2017.
15. D. P. Kingma, J. Ba, “Adam: A Method For Stochastic Optimization,” 3rd International Conference for Learning Representations, San Diego, 2015.
16. X. Glort, Y. Bengio, “Understanding the difficulty of training deep feedforward neural networks,” International Conference on Artificial Intelligence and Statistics, Sardinia, Italy , 2010, PMLR 9:249-256.
17. K. M. He, X. Y. Zhang, S. Q. Ren and J. Sum, “Delving Deep into Rectifiers: Surpass Human-Level Performance on ImageNet Classification,” The IEEE International Conference on Computer Vision (ICCV), Santiago, Chile, 2015, pp. 1026-1034.
18. S. Ioffe, C. Szegedy, “Batch Normalization: Acceleration Deep Network Training by Reducing Internal Covariate Shift,” arXiv, 1502.03167, February, 2015.
19. N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever and R. Salakhutditnov, Dropout: A simple way to prevent neural networks from overfitting. The Journal of Machine Learning Reasearch, pages 1929 – 1958, 2014.
20. K. Simonyan, A. Zisserman, Very Deep Convolutional Networks FOR Large-Scale Image Recognition. Published as a conference paper at ICLR, San Diego, 2015.
21. ILSVRC, Web site: http://www.image-net.org/challenges/LSVRC/, 2017.
22. C. Szegedy et al, “Going Deeper With Convolutions,” IEEE Conference on Computer Vision and Pattern Recognition(CVPR), Boston, Massachusetts, 2015.
23. K. M. He, X. Y. Zhang, S. Q. Ren and J. Sun, “Deep Resodual Learning for Image Recognition,” arXiv, 1512.03385, December, 2015.
論文全文使用權限:同意授權於2018-08-17起公開