摘要
近年来,大庆油田南一区出现了严重的套损问题,产生了巨大的经济损失。本文根据大庆油田南一区西西区块地质、开发及工程部分数据,首次采用混合蒙特卡洛(HMC)算法学习的贝叶斯神经网络方法(BNN)来对套管状况进行预测。并与BP神经网络算法和Laplace算法学习的BNN算法进行预测分析比较,HMC算法学习的BNN的预测结果比BP神经网络算法和Laplace算法学习的BNN算法的误差值小。结果表明HMC算法学习的BNN具有较高的预测精度和较强的泛化能力,为套损预测提供有力的手段,具有较好的应用前景。
In recent years,serious casing damage occurs in NanYi Block of Daqing oilfeild,which inflicts great economic harm.Based on data about geological,development and engineering factors of some wells from XiXi Block in Nanyi Block of Daqing oilfield,it is the first time to apply Bayesian neural network learned by HMC algorithm to casing damage forecasting,and compare the result with these of BP neural network and BNN learned by Laplace algorithm.The error of the former is smaller than these of the latter two.Hence,the model based on BNN learned by HMC algorithm has higher prediction accuracy and stronger generalization ability.It provides a powerful means to casing damage forecast,and should be widely used.
作者
王璐
孟凡顺
张旭
伊天宇
WANG Lu;MENG Fan-Shun;ZHANG Xu;YI Tian-Yu(Ocean University of China.Teaching Center of Fundamental Courses,Qingdao 266100;Ocean University of China.,College of Marine Geoscience Qingdao 266100)
出处
《内蒙古石油化工》
CAS
2020年第3期9-12,共4页
Inner Mongolia Petrochemical Industry
关键词
贝叶斯神经网络
套损预测
蒙特卡洛方法
哈密顿动态系统
Bayesian Neural Network
Casing Damage Forecast
Monte Carlo Algorithm
Hamiltonian Dynamics
