基于SVM的套管最大von Mises应力预测方法

An Prediction Method for Determining the Maximum von Mises Stress in Casing Based on SVM

狄勤丰 吴志浩 王文昌 覃光煦 陈锋

DI Qinfeng WU Zhihao WANG Wenchang QIN Guangxu CHEN Feng

上海大学应用数学和力学研究所，上海 200072

Institute of Applied Mathematics and Mechanics, Shanghai University, Shanghai, 200072, China

为了预测非均匀地应力条件下不居中套管的最大应力，提高套管安全性，研究了基于支持向量机（SVM）的套管最大von Mises应力预测方法。首先确定了影响套管最大应力的关键因素，包括非均匀地应力、水泥环的弹性模量及泊松比、套管偏心距等8个因素；然后利用ANSYS软件构建了套管应力实验样本；最后建立了$\varepsilon {\rm{ - SVR}}$模型，实现了套管最大应力的预测。通过自学习，基于径向基核函数的SVM回归方法对于训练样本达到了很好的精度，5个测试样本的平均相对误差仅为1.32%，具有较好的预测精度，满足工程需求，且可以实现非均匀地应力条件下不居中套管最大应力的快速求解。研究结果为现场安全施工提供了理论依据。

In order to predict the maximum stress of uncentered casing under non-uniform in-situ stress and improve the safety of casing, a prediction method of casing’s maximum von Mises stress based on artificial intelligence SVM is studied. First, the key factors affecting the maximum stress of casing are determined, including non-uniform geologic stress, elastic modulus and Poisson's ratio of cement sheath, eccentricity of casing, etc. Then the "experimental" samples of casing stress are constructed by using ANSYS software. Finally the $\varepsilon {\rm{ - SVR}}$ model is established to realize the prediction of casing’s maximum stress. Through self-learning, the SVM regression method based on RBF kernel achieves good accuracy for training samples. For the five test samples, the average relative error is only 1.32%, which means that this method can meet the needs of engineering application. In particular, this method can be used to quickly solve the maximum stress of uncentered casing under non-uniform in-situ stress.The research results provide theoretical basis for site safety construction.

国家自然科学基金重点项目“超深井钻柱非线性动力学及动态安全性基础理论研究”（编号：U1663205）、国家自然科学基金青年项目“气体钻井中预弯底部钻具组合控斜的非线性动力学机制”（编号：51704191）和国家自然科学基金青年项目“超深井油套管螺纹接头三维力学特性分析及密封机理研究”（编号：51804194)联合资助

https://doi.org/10.11911/syztjs.2019065