基于BP神经网络的油气储量价值等级划分  

Applying BP Neural Network to Grade Reserve Value of Oil and Gas

作　　者：王化增[1] 迟国泰[1] 程砚秋[1] 

机构地区：[1]大连理工大学管理学院,辽宁大连116024

出　　处：《中国人口·资源与环境》2010年第6期41-46,共6页China Population,Resources and Environment

基　　金：国家自然科学基金项目(No.70471055);高等学校博士学科点专项科研基金项目(No.20040141026)资助

摘　　要：在广泛选取原始指标的基础上,从可采储量、油气价格、开发投资、经营成本4个方面,构建了基于主成分分析法的油气储量价值等级划分指标体系,建立了基于BP神经网络的油气储量价值等级划分模型,并对胜利油田的数据进行实证分析。本文的创新及特色一是通过用7个主成分保留了95%的原始信息建立指标体系,避免了指标间相关性对后期评价的影响,提高了后期评价的准确性。二是通过设置初始权重、学习率、动态系数等参数使基于BP神经网络的油气储量价值等级划分模型的精度高达96.61%,避免了传统评价中模糊随机因素和人为主观因素的影响,提高了评价的准确性和科学性。结果表明,采收率、储量丰度、储量规模、储层埋深、凝固点等5个指标是影响油气储量价值等级的关键因素。储量价值越高,采收率越大、储量规模越大、储量丰度越大、储层埋深越小、凝固点越低。On the ground of the widely chosen efficiency indexes of reserve value of oil and gas,an index system of reserve value of oil and gas is established based on principal component analysis.The index system is formed through four aspects,such as recoverable reserves,oil-gas price,development investment and operating cost.Then,BP neural network is applied to grade reserve value.The contribution of this article came from the following three aspects. Firstly,the new index system is built based on improved principal component analysis which keeps 96.50% of the original information with 7 indexes and avoids the influence of the correlation between different indexes on later appraise.Secondly,BP neural network is established which avoids the influence of the subjective factors.By setting the initial weights,learning rate,the dynamic coefficient of parameters,the accuracy of the model is 96.61%.Thirdly,the result reflects that the major factors that affect the grade reserve value of oil and gas are the recovery factor,the reserve abundance,the reserve scale,the reserve depth and freezing point.

关 键 词：BP神经网络 储量价值 主成分分析 价值评价 

分 类 号：TE122[石油与天然气工程—油气勘探]