基于差分进化优化的随机森林产能预测  

Productivity Prediction Based on Random Forest Optimized by Differential Evolution

作　　者：毛靖 陈仲旭 刘加元 赵行 MAO Jing;CHEN Zhongxu;LIU Jiayuan;ZHAO Xing(Tazhong Oil and Gas Management Department of Tarim Oilfield,Kuerle 841000,China)

机构地区：[1]塔里木油田塔中采油气管理区,新疆库尔勒841000

出　　处：《河南科技》2024年第21期32-37,共6页Henan Science and Technology

摘　　要：【目的】为了解决传统的油藏产能预测方法存在考虑因素少、耗时长、计算过程复杂、在复杂地质条件下预测精度低等问题,有必要对基于差分进化优化的随机森林产能预测方法进行研究。【方法】采用机器学习的方法,建立基于差分进化算法优化的随机森林产能预测模型。以某油藏为例,根据油田实际开发情况,从地质、开发和工程等方面选择对产能影响比较大的6个因素,采用Person相关系数法分析各个影响因素之间的线性相关性,运用随机森林算法计算各个因素对产能的影响程度并进行主控因素分析。【结果】研究结果表明,孔隙度和含油饱和度之间的正相关性最强;对产能影响的程度从高到低分别为生产压差、射孔段厚度、渗透率、初始含油饱和度、油层有效厚度、孔隙度。支持向量机、多元线性回归、基于网格搜索优化的随机森林等方法中,采用基于差分进化优化的随机森林方法的预测精度最高。【结论】研究成果为复杂油藏产能预测提供了新的思路。[Purposes]In order to solve the problems of traditional reservoir productivity prediction methods,such as less consideration,long time consuming,complicated calculation process,low prediction accuracy under complex geological conditions,it is necessary to study the random forest productivity prediction method based on random forest optimized by differential evolution.[Methods]This paper adopted machine learning approach and proposed a method of productivity prediction based on random forest optimized by differential evolution.A reservoir was selected as the example.According to the actual development situation of the oilfield,six factors which have greater impacts on productivity were selected from aspects of geology,development,and engineering.Person correlation coefficient method was used to calculate the linear correlation among influencing factors while random forest algorithm was applied to calculate the influence degree of each factor on productivity and analyze the main controlling factors.[Findings]The results showed that the positive linear correlation between porosity and initial oil saturation was the strongest.The degree of influence on productivity from high to low was producing pressure differential,perforation thickness,permeability,initial oil saturation,effective oil layer thickness and porosity.Compared with support vector machine,multiple linear regression and random forest optimized by grid search,random forest optimized by differential evolution has highest prediction accuracy.[Conclusions]This paper provided other theoretical method for predicting the productivity of complex reservoir.

关 键 词：机器学习 产能预测 随机森林 差分进化 

分 类 号：TE358.5[石油与天然气工程—油气田开发工程]