摘要
压裂工艺是低渗透油田开发的一项关键技术。国内每年实施压裂近万井次,为压裂分析积累了大量的实践经验。传统的建模方法是基于力学理论,模型复杂,对专业要求非常高。用机器学习技术,可以从已有的数据中抽取有用的信息,建立有效的数学模型。传统的机器学习算法只有在已知样本数无限多时才有效,而实际应用中压裂分析的已知样本数非常有限。目前常用的机器学习算法有拟合、聚类、决策树、神经网络、遗传算法、支持向量机等,本文针对其中的四种算法开展了现场应用,并对各种算法进行了分析,指出了算法的特点和应用条件。通过对机器学习算法的分析表明,各种算法可信度高,使用方便,可以进行选井选层、压裂效果预测、优化压裂设计。
Hydraulic fracturing plays important role in exploration of low permeability oilfield. Every year there are nearly ten thousands fracturing wells, which accumulate many practical experiences for fracturing analysis. Traditional fracturing modeling is based on.mechanics theory with the disadvantage of complicated model and special subject which demands a very high professional. Machine learning technology is theory that extracts useful information from historic data to build effective mathematical model. Traditional machine learning algorithm is applicable for very large samples while limited data can be acquired in reality. Recently several machine learning algorithm is used widely in petroleum industry such as fitting, cluster, decision tree, neural network, genetic algorithm, support vector machine. In this paper four algorithms are analyzed and applied. With comparison every algorithm＇s character and applied condition is defined. Machine learning algorithm has high credibility and easy operation which can be uses in wells and formation selection, postfracture response prediction, fracturing design optimization.
出处
《内蒙古石油化工》
CAS
2010年第2期17-20,共4页
Inner Mongolia Petrochemical Industry
关键词
压裂
机器学习
神经网络
遗传算法
支持向量机
Fracturing
Machine learning
Neural network
Genetic algorithm
Support vector machine
