基于改进BP神经网络的页岩地应力预测模型  

Shale In-situ Stress Prediction Model Based on Improved BP Neural Network

作　　者：尚福华[1] 王玮卿 曹茂俊[1] SHANG Fu-hua;WANG Wei-qing;CAO Mao-jun(School of Computer and Information Technology,Northeast Petroleum University,Daqing 163318,China)

机构地区：[1]东北石油大学计算机与信息技术学院,黑龙江大庆163318

出　　处：《计算机技术与发展》2021年第7期164-170,共7页Computer Technology and Development

基　　金：国家重大科技专项(2017ZX05019-005);黑龙江省自然科学基金(LH2019F004)。

摘　　要：目标地区的地应力计算模型构建是指导该地区压裂,井壁稳定分析的基础工作。针对模型构建问题,该文构建了一种页岩地应力计算模型,并基于改进的BP神经网络对模型中的岩石力学参数进行反演预测,进而进行地应力预测。首先基于黄式模型,考虑页岩的各向异性,层理倾角影响,构建了符合目标地区的地质概况的地应力计算模型。然后设计了基于BP神经网络的岩石力学参数反演方法,将实测应力值与岩石力学参数通过神经网络建立非线性映射关系,进行网络训练,输出的参数可用于模型计算,进行地应力计算。最后以长宁工区龙马溪组页岩的测井资料为基础数据,进行实际计算。实验结果表明,改进的神经网络算法更易于收敛,应用于地应力计算模型的计算值与实测值相比较误差较小。The construction of in-situ stress calculation model in the target area is the basic work to guide the fracturing and wellbore stability analysis in this area.In order to solve the problem of the model construction,we construct a shale in-situ stress calculation model,and based on the improved BP neural network,the rock mechanics parameters in the model are back predicted,and then the in-situ stress is predicted.Firstly,based on the Yellow model,considering the anisotropy of shale and the influence of bedding dip angle,the in-situ stress calculation model in line with the geological situation of the target area is constructed.Then,the back analysis method of rock mechanics parameters based on BP neural network is designed.The nonlinear mapping relationship between measured stress value and rock mechanical parameters is established through neural network for network training.The output parameters can be used for model calculation and in-situ stress calculation.Finally,based on the logging data of Longmaxi formation shale in Changning work area,the actual calculation is carried out.The experiment shows that the improved neural network algorithm is easier to converge,and the error between the calculated value and the measured value of the in-situ stress calculation model is smaller.

关 键 词：页岩 BP神经网络 各向异性模型 地应力 自适应学习率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]