【作者】 石林；

【导师】 陈冬；

【作者基本信息】 中国石油大学(北京) ， 油气井工程， 2022， 硕士

【摘要】 地层岩性的准确识别是地质参数确定的前提,也是储层特征研究、储量估算和地质建模的基础。随着测井技术的发展,传统的地层岩性识别方法已经越来越难以满足高效、快速以及高精度的需求。人工智能与大数据技术的崛起,使得机器学习技术开始在油气田勘探开发过程中逐渐应用。本文针对测井数据存储与处理的自动化和地层岩性感知智能化问题,利用机器学习算法开展了测井数据预处理的系统化流程研究,搭建NLDIW-PSO-XGBoost模型实现了基于测井数据的地层岩性智能识别,并基于石油云平台开发了地层岩性智能感知模块。主要取得的成果和认识如下:(1)建立了储存地质数据的数据库,并设计了一套对于测井数据处理的全系列流程,包括井位聚类分析、测井数据挖掘及可视化和异常值检测。初步实现了测井数据处理过程中的自动化与智能化。(2)基于门控循环神经网络(GRU)开展了测井曲线补全实验和生成实验,利用该方法生成缺失的测井曲线,为后续建模提供了更高质量的数据集,可以作为真实地层解释的参考依据。(3)提出了利用非线性递减惯性权重改进粒子群算法,然后建立NLDIWPSO-XGBoost模型对地层岩性进行识别,并将模型预测结果与决策树、支持向量机、多层感知机、随机森林、梯度提升树模型进行对比。实验结果表明,优化的XGBoost模型在盲井测试中表现出了最优的模型性能,而且还表现出了较强的泛化能力。(4)基于石油云平台搭建了地层岩性智能感知云端应用模块,该模块可以自主定义训练参数,并从交互界面直观展示不同模型的预测结果,形成了一套地层岩性智能识别的高效流程。基于以上结果,本论文可为智能油气井提供技术储备,对测井数据自动化处理流程和地层岩性智能感知模型的设计具有指导意义,推动了地层解释相关领域的数字化与智能化发展。更多还原

【Abstract】 The accurate classification of lithology is not only the premise of calculating geological parameters,but also the basis of reservoir characteristics research,reserve estimation and geological modeling.With the development of logging technology,the traditional lithology classification method has become more and more difficult to meet the needs of high efficiency and precision.With the rise of artificial intelligence and big data technology,machine learning technology has been gradually applied in the process of oil and gas field exploration and development.Aiming at the problems of automatic well log data storage and processing and intelligent formation lithology classification,this thesis uses machine learning algorithms to carry out systematic process of well log data preprocessing,and builds NLDIW-PSO-XGBoost model to realize intelligent formation lithology classification based on well log data.Finally,the formation intelligent lithology classification module is developed based on petroleum cloud platform.The main research contents and results are as follows:(1)A database for storing geological data is established,and a whole series of processes for well log data processing are proposed,including well location cluster analysis,well log data mining,visualization and outlier detection.The automation and intellectualization of well log data processing are preliminarily realized.(2)The gated recurrent neural network(GRU)is used to carry out the well log completion and generation.This method provides higher quality data sets for subsequent modeling by generating missing well logs and serves as a reference for real formation interpretation.(3)An improved particle swarm optimization algorithm using nonlinear decreasing inertia weight is proposed,and then the NLDIW-PSO-XGBoost model is established to identify the lithology.The prediction results of the model are compared with decision tree,support vector machine,multi-layer perceptron,random forest and gradient boosting tree model.The experimental results show that the optimized XGBoost model not only achieves the best effect in blind well test,but also shows strong generalization ability.(4)The cloud application module of intelligent lithology classification is developed based on the petroleum cloud platform.The module can independently define training parameters and visually display the prediction results of different models from the graphical user interface.Finally,a set of efficient process for intelligent lithology classification is formed.Based on the above results,this thesis can provide technical reserves for intelligent drilling,and has guiding significance for the design of automatic well log data processing and intelligent lithology classification method,which promotes the development of digitalization and intelligence in the field of formation interpretation.更多还原