基于BP神经网络集成的钻井液粒度分布预测  

Drilling fluid particle size distribution prediction based on BP neural network ensemble

作　　者：刘鹭[1] 蒲晓林[1] 陶怀志[2] 罗霄[1] 黄桃[1] 

机构地区：[1]油气藏地质及开发工程国家重点实验室(西南石油大学),四川成都610500 [2]川庆钻探工程有限公司钻采工程技术研究院,四川广汉618301

出　　处：《计算机与应用化学》2014年第4期421-425,共5页Computers and Applied Chemistry

基　　金：国家自然科学基金(51304169);油气藏地质及开发工程国家重点实验室开放课题(西南石油大学)(PLN1136)

摘　　要：利用BP神经网络模拟了钻井液的粒度分布与钻井液处理剂种类及加量之间的关系,并采用神经网络集成的方法提高模型的泛化能力。利用激光散射粒度分析仪测定了NaCl、CaCl_2、钻井液用高温抗盐降滤失剂SPNC、磺甲基酚醛树脂(SMP-2)4种处理剂对钻井液粒度分布的影响,选用60组实验数据作为试验样本,建立了钻井液粒度分布模型,并利用19组实验数据验证其精度。结果表明:该模型具有良好的预测精度,集成网络输出结果的平均误差率和最大误差率均小于单个神经网络子网,表现出良好的泛化能力;并利用该模型研究了单一处理剂对钻井液粒度的影响。BP neural network ensemble was proposed to simulate complex nonlinear relationship exists between drilling fluids particle size distribution and the types and dosage of additives, and neural network ensemble was used to enhance generalization ability. LA-950 Laser Scattering Particle Size Analyzer was used to test the effect ofNaC1, CaCI2, high temperature salt fluid loss additive SPNC, sulfomethylated phenolic resin （SMP-2）, four kinds of agents on the particle size distribution of drilling fluid. 60 groups of data were chosen for test samples, to establish drilling fluid particle size distribution model, 19 groups of data for test. The results showed that the output Ensemble ANN showed good generalization ability, whose average error and the maximum error were less than single neural network subnets. The model was also used to study one single agent effect on the particle size.

关 键 词：BP神经网络 钻井液粒度分布 神经网络集成 泛化能力 

分 类 号：TE39[石油与天然气工程—油气田开发工程] TP391[自动化与计算机技术—计算机应用技术]