類神經網路建立地層滲透率模式之研究

Development of Stratum Permeability Models through Artificial Neural Network

謝秉志(Bieng-Zih Hsieh) ； 吳政達(Cheng-Da Wu) ； 林再興(Zsay-Shing Lin)

《鑛冶：中國鑛冶工程學會會刊》 47卷2期 (2003/06) Pp. 40-49

https://doi.org/10.30069/MM.200306_47(2).0005

摘要

本研究目的是利用地球物理井下電測資料及岩心滲透率資料，訓練類神經網路，建立滲透率計算模式，可為預測或計算滲透率之用。本研究利用所蒐集的三口井之岩心滲透率資料，當作網路輸出目標參數值。井測資料之各種井測，包含自然加瑪電測、感應式電阻電測、球聚式焦點電測、井徑、地層密度電測及聲波電測等，作為網路的輸入參數。本研究建立三種滲透率預測模式（三種不同的輸入參數個數及隱藏層處理單元個數之組合），其中以『4-5-1』（含4個輸入參數，5個隱藏層處理單元及1個輸出參數）之模式為最好。本研究發展之『4-5-1』網路預測模式比基因演算法之最佳化滲透率模式更能去擬合資料點。

關鍵字

井下電測 ； 類神經網路 ； 地層滲透率

並列摘要

The purpose of this study is to develop a prediction model for predicting and calculating the permeability by training the back propagation neural network with the geophysical well logging data and the core permeability data. In this study, the collected core permeability data from three wells were used as the target values of the output layer parameter. The well logs data including the natural gamma ray log, the deep induction resistivity log, the spherically focused resistivity log, the caliper log, the compensated formation density log and the compensated borehole sonic log were used as the input layer parameters. Three permeability prediction model (three different combinations of input layer parameters, hidden layer neurons, and output layer parameter) had been constructed. The best prediction model was the 『4-5-1』 model which included 4 input layer parameters, 5 hidden layer neurons, and 1 output layer parameter. The Prediction ability of the neural network model is batter than that of the optimal genetic algorithm model.

並列關鍵字

Well Logs ； Neural Networks ； Permeability