類神經網路評估已開發天然氣礦區蘊藏量之研究

• Main Authors: Wang Chung-Shing, 王崇興
• Other Authors: Zsay Lin
• Format: Others
• Language: zh-TW
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/24335379143003175121

碩士 === 國立成功大學 === 資源工程學系 === 87 === ABSTRACT The purposes of this study are to establish and train a neural network for a gas field in Taiwan,and then to predict future production rate as well as to estimate remaining reserves. The gas reserves estimated from neural network are compared with those from decline curve analysis. Field data, including gas production rate, water production rate, gas price, and gas demand, from T-gas field in Taiwan are collected and used in this study. Five different kinds of neural network are derived from various combinations of four category data. Back propagation neural networks of three layers(with one hidden layer) are established in this study. Training and validating of these five neural networks have been conducted. Only two neural networks with good validation results are selected for future gas prediction and reserves estimation. Analysis of production data shows that production rate was probably affected by injecting wells in the field. The reserves estimates from neural networks for T-gas field ranges from 1.5 billion and 6.0 billion SCM (standard cubic meters). For comparison, decline curve analysis, which including conventional, probabilistic, and stochastic analysis, are performed to estimate reserves for the T-gas field. The reserves estimates from decline curve analysis are between 1.7 billion SCM and 4.4 billion SCM.