碩士 / 國立成功大學 / 資源工程學系碩博士班 / 90 / ABSTRACT
Permeability information is a very important parameter in the reservoir engineering and underground water study. In the process of deriving the traditional permeability models, the key steps are the selection of model and the determination of constant parameters in the model. These key steps can be easily implemented by developing an artificial neural network. Therefore the purpose of this study is to use geophysics well logging data and core permeability data to derive permeability models (traditional models) in which the parameters in the model are estimated by genetic algorithm. At the same time, we also use the known data to train the artificial neural network for calculating the permeability.
In this study, geological well logs from three wells (X-9、X-10 and X-11 , well depths between 3244 m to 3450 m ) in southern Taiwan area are collected and analyzed. The collected well logging data include gamma ray (GR)、resistivity induction log deep (ILD)、resistivity of spherically focused log (SFL)、resistivity of microspherically focused log (MSF)、caliper log (CAL)、compensated neutron log (CNL)、formation density log (FDC) and borehole compensated sonic log (BHC). Three permeability models, such as, Wylie - Rose model、Coates-Dumanoir model and Porosity model, are selected. These three models are derived from using collected data, the well log data and core data, in the interval of Oligocene sand formation. Also, the data in NP24B and NP24A, which are in the the Oligocene sand formation, are used separately to derive the Wylie - Rose model、Coates-Dumanoir model and Porosity model. The Wylie - Rose model is the best of among those models.
Besides, three artificial neural network is also developed from using the data collected. In the first kind of neural network structures, the input parameters are：DEPTH、CAL、BHC、GR、ILD、SFL and FDC；and the output data is formation permeability. In the second kind of neural network, the input parameters are : CAL、BHC、GR、ILD、SFL and FDC；and output data is formation permeability. In the third kind of neural network, the input parameters are：porosity、water saturation、formation resistivity and formation water resistivity；and output data is formation permeability .The result of the first kind of artificial neural network model is the best of the three models ,because of the Mean Square Error being the minimum, and coefficient of determination being the maximum .The permeability calculated from using artificial neural network is better than using genetic algorithm .
Keywords：well logging ; genetic algorithm ; artificial neural network；formation permeability
| Identifer | oai:union.ndltd.org:TW/090NCKU5397007 |
| Date | January 2002 |
| Creators | Chuan-Sheng Tian, 田川昇 |
| Contributors | Zsay-Shing Lin, 林再興 |
| Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
| Language | zh-TW |
| Detected Language | English |
| Type | 學位論文 ; thesis |
| Format | 112 |
Page generated in 0.0258 seconds