摘要

由電測資料計算地層參數時，需要利用或者是選用一些含有常數的方程式或經驗式，為避免這些常數的推求以及經驗式的選用，可以採用類神經網路的訓練模式來建立其間的關係式。因此，本研究的研究目的是利用類神經網路訓練井測資料，建立分析及計算地層參數之模式，以計算地層參數。

本研究所蒐集的資料包括：台西二號井、台西四號井、台西五號井及台西八號井，電測資料深度範圍約介於30-1500公尺之間。本研究所建立類神經網路模式，其輸入變數包括：深度、加瑪電測、聲波電測、地層密度電測及電阻電測，輸出變數包括：頁岩質含量、孔隙率及地層電阻。利用中國石油公司內部所計算的台西五號井之結果(頁岩質含量、孔隙率及地層電阻)作為訓練資料建立計算模式，將所得的類神經網路模式來分析及計算台西地區其他井(台西二號井、台西四號井及台西八號井)的地質參數。

由類神經網路模式分析(預測)台西二號井的地層深度約介於61公尺至1537公尺之間而所得的結果是：各個地層分層的頁岩質含量介於12.26％至40.90％，岩性為頁岩質砂岩。平均孔隙率介於18.84％至37.51％。地層電阻介於0.691 ohm-m至95.79 ohm-m。台西四號井的地層深度約介於182公尺至1347公尺之間而所得的結果是：頁岩質含量介於10.36％至44.42％，岩性皆為頁岩質砂岩。平均孔隙率介於26.44％至37.19％。地層電阻1.32 ohm-m至79.99 ohm-m。台西八號井的地層深度約介於310公尺至1448公尺之間而所得的結果是：頁岩質含量介於14.02％至33.83％，其岩性也皆為頁岩質砂岩。各個地層之平均孔隙率介於12.87％至49.43％。各個地層之地層電阻介於1.86 ohm-m至30.94 ohm-m。

本研究中所計算(預測)的深部地層之孔隙率值高達30％-41％，其原因可能是，在類神經網路所用的訓練資料是淺層地層資料，其地層鬆軟且未固結，以及孔隙率值沒有經過修正。所以，會對深部地層的孔隙率值造成影響，所以，在深部地層的孔隙率值的準確性值得商議。

Abstract

In well log analysis, some empirical formula or parameters, depending on regional geology, are required. In order to avoid using these empirical formula or parameters, a neural network model may be used. So the purpose of this study is to develop a neural network, using well logging data, for estimating formation parameters.

For developing a neural network, the data collected in this study includes: Gamma Ray Log(GR)、Deep Induction Log (ILD)、Borehole Compensated Sonic Log (BHC)、and Formation Density Log(FDC) from the wells of THS-2, THS-4, THS-5, and THS-8, for the depth between 30-1500 meters from Tai- His area in Taiwan. The input variables of the neural network developed in this study are: depth, GR, BHC, FDC, and ILD. And the output variables include: shale content, porosity, and formation resistivity. The trained data (shale content, porosity, and formation resistivity) for the neural network are from the results of well log analysis by Chinese Petroleum Corporation for THS-5 well. Then the neural network model developed is used to analyze and calculate the formation parameters of the other wells in Tai- His area (THS-2、THS-4、 and THS-8 wells).

The results from neural network analysis of the THS-2 well (depth between 61m to 1537m) are as follows: the shale content in each layer of formation is between 12.26% and 40.90%. The formation is shaly sand. The average porosity is between 18.84% and 37.51%. And the formation resistivity is between 0.691ohm-m and 95.79ohm-m. The formation parameter for the THS-4 well (depth between 182m to 1347m) is as follows: the shale content in each layer of formation is between 10.36% and 44.42%. The formation is shaly sand. The average porosity is between 26.44% and 37.19%. And the formation resistivity is between 1.32ohm-m and 79.99ohm-m. For THS-8 well (depth between 310m to 1448m) the shale content in each layer of formation is between 14.22% and 33.83%. The formation is shaly sand. The average porosity is between 12.87% and 49.43%. And the formation resistivity is between 1.86ohm-m to 30.94ohm-m.

In this study, the calculated porosity is pretty high (between 30% and 41%) in deep formation (for the depth greater than 700m). The reason for the high porosity in deep formation may be because the porosity of training data is from shallow formation which is the weak and unconsolidate.

參 考 文 獻
Ali, M., Chawathe, A.,〝Using Artificial Intelligence to Predict Permeability from Petrographic Data,〞 Computers & Gosciences ,915-925(September2000),.
Chun Che Fang , Kok Wai Wong, Halit Eren, Robert Charlebois and Hugh Crocker,〝Modular Artificial Neural Network For Prediction of Petrophysical Properties from Well Log Data,〞,IEEE，Brussels, Belgium（1992）.
Crain, E.R.,〝The Log Analysis Handbook Quantitative Log Analysis Methods vol.1,〞PennWell Publishing Company, Tulsa , Oklahoma, (1986).
Dresser Atlas,〝Log Interpretation Fundamentals,〞Dresser Industries Inc., Houston, Texas, (1975).
Haykin , Simon, Neural Network—A Comprehensive Foundation, Prentice Hall, New Jersey,(1999).
Hornik, K., Stinchcombe, M. and White, H. 〝Multilayer Feedforward Networks are Universal Approximators ,〞Neural Networks, vol.2, (1989).
Lin, L.H.,Lin,H.R.,Ke,A.H.W.,and Chou,T.H.,〝Potential of the Pre-miolene Formations, in the Chianan Plain, Taiwan,〞Petroleum Geology of Taiwan,vol.27,117-197(1992).
Matlab, Neural Network Toolbox, User's Guide, The Math Work Inc., (1998).
Mohaghegh, S., Arefi, S., Ameri, S and Rose,D., 〝Design and Development of an Articial Neural Network for Estimation of Formation Permeability,〞 SPE-28237 , Rewservoir Engineering(September 1995).
Mohaghegh, S.，Arefi, R.，Ameri, S.，Aminiand, K. and Nutter,R.〝Petroleum Reservoir Characterization with the Aid of Artificial Neural Network,〞Journal of Petroleum Science and Engineering ,Volume：16,263-274,(1996a).
Mohaghegh, S., Arefi, R., Ameri, S., Aminiand, K and Nutter, R.,〝Petroleum Reservoir Characterization with the Aid of Artificial Neural Networks,〞 SPE , Rewservoir Engineering(September 1996)263-274.
Mohaghegh, S., McVey, D., Aminian, K. and Ameri, S., 〝Predicting Well- Stimulation Results in a Gas-Storage Field in the Absence go Reservoir Data with Neural Networks,〞SPE, Rewservoir Engineering(November 1996)268-272.
Rogers, S. J., Fang, J. H. , Karr, C. L. ,and Stanley, D. A.,〝Determination of Lithology，From Well Logs Using a Neural Network,〞AAPG Bull. ,vol.76,pp.731-739,（1992）.
Schlumberger, Log Interpretation ,Volume I —Principle , Schlumberger Limited, New York ,N.Y.,(1972).
Weiss ,W. W. , Wo, S. , Balch, R. S.,“Integrating Core Porosity and Sw Measurement with Log Values,”SPE , Gillette , Wyoming（1999）。
White, A. C., Molnar, D., Aminian, K., Mohaghegh, S. Ameri, S. and Esposito, P., "The Application of ANN for Zone Identification in a Complex Reservoir, " Paper SPE 30977 presented at the SPE Eastern Regional Conference & Exhibition, Morgantown, West Virginia, September 19-23,(1995).
Wong P. M., Henderson D. J. and Brooks L. J.,〝 Permeability Determination Using Neural Networks in the Ravva Fiels, Offshore India,〞 SPE Reservoir Evaluation & Engineering, (1998).
Wong, P. M., Gedeon, T. D., and Taggart, I. J., “An Improved Technique in Porosity Prediction: A Neural Network Approach, ” IEEE Transactions on Geoscience and Remote Sensing, vol.33, NO.4, ( 1995).
中國石油學會及中國石油股份有限公司〝電測解釋作業程序(F1-76-P06-021)，〞台北，(1987)。
王崇興，〝類神經網路評估已開發礦區蘊藏量之研究，〞國立成功大學資源工程研究所碩士論文，(1998)。
田川昇、謝秉志、林再興，〝利用基因演算法建立滲透率計算模式，〞 石油季刊，第36卷，第4期，第1-8頁，(2001)。
林人仰，〝類神經網路在地層辨識的應用，〞石油鑽採工程，第四十期，第104-120頁，（1999）。
林再興、謝秉志，〝利用井測資料估算地層強度之研究，〞礦冶，第44卷，第3期，第49-62頁，（2000）。
林再興，〝利用類神經網路及灰色理論預測以生產礦區之油氣生產量及蘊藏量之研究，〞國科會產學技術合作研究計畫，（2000）。
林再興 、 陳昶旭，〝利用類神經網路及灰色理論預測已生產礦區之油氣生產量及蘊藏量之研究，〞石油暨石化科技產業科技學術合作八十九年度期末報告，(2001)。
林再興、謝秉志、湯于德、蔡孟原、李振誥及陳時祖，〝雲林地層下陷區水文特性之計算-井下電測資料分析，〞第二屆地下水資源及水質保護研討會，853-864頁，(1997)。
吳忠益，〝利用井測資料與基因演算法研究雲林沿海地區深部含水地層特性，〞國立成功大學資源工程研究所碩士論文，(2000)。
陳昶旭、林再興，〝利用類神經網路估算地層滲透率之研究，〞環球學報，(2000)。
葉怡成，〝應用類神經網路，〞儒林圖書公司，台北，（1997）。
葉怡成，〝類神經網路模式應用與實作，〞儒林圖書公司，台北，（1998）。
葉怡成，〝應用類神經網路模式，〞儒林圖書公司，台北，(2001)。
周錦德，〝石油地質學（上冊），地物、地化與地質探勘之整合，〞台北，(1989)。
孫習之，〝台灣海域的第三紀沉積盆地〞，From the Proceedings of the Second ASCOPE Conference and Exhibition，(1982)。
鄧屬予，〝台灣的沈積岩，〞經濟部中央地調所，台北，(1997)。