SA-SVM模型预测泥质含量研究  

Study on Predicting Mud Content by the SA-SVM Model

作　　者：刘维凯[1] 白婷婷 LIU Wei-kai;BAI Ting-ting(Northeastern Petroleum University,Daqing Heilongjiang 163318,China)

机构地区：[1]东北石油大学,黑龙江大庆163318

出　　处：《当代化工》2022年第1期119-122,128,共5页Contemporary Chemical Industry

基　　金：国家自然科学基金(项目编号:52004064)。

摘　　要：泥质含量计算的准确性直接影响解释储层、分析储层岩性的精确度。常规测井曲线解释泥质含量需要多种测井曲线相结合,工作效率低,不可控因素多。通过分析测井资料与泥质含量的相关性,选取自然电位、自然伽马、声波时差和密度测井为特征参数,运用模拟退火算法优化支持向量机回归模型中的超参数,提高算法全局搜索能力。利用测井资料数值进行训练测试,SA-SVM模型预测孔隙度训练集和测试集的均方根误差分别为0.080 5和0.065 9,而SA-SVM模型预测的准确度分别为0.901 2和0.965 7。结果表明:相较于SVM模型,SA-SVM模型训练集和测试集均方误差分别降低了5.5%和25.6%,泥质含量预测的准确性提高了3.9%和7.1%。模拟退火优化算法不仅支持向量机的最佳参数,而且有效地避免陷入局部极小点,扩大全局搜索空间。The accuracy of mud content calculation directly affects the accuracy of reservoir interpretation and analysis of reservoir lithability. Routine logging curve explaining mud content requires a combination of multiple logging curves with low working efficiency and many uncontrollable factors. By analyzing the correlation between logging data and mud content, selecting natural potential, natural gamma, acoustic time difference and density measurement as the characteristic parameters, the hyperparameters in the support vector machine regression model were optimized by using the simulated annealing algorithm to improve the global search capability of the algorithm.Training tests using logging data values showed a root mean square error of 0.0805 and 0.0659, respectively,whereas the accuracy of the SA-SVM model was 0.901 2 and 0.9657, respectively. The results showed that the mean square error of the SA-SVM model decreased by 5.5% and 25.6%, respectively, and the accuracy of mud content prediction was increased by 3.9% and 7.1% compared to the SVM model.The simulated annealing optimization algorithm not only supports the best parameters of the vector machine, but also effectively avoids falling into local minima and expanding the global search space.

关 键 词：泥质含量 模拟退火算法 支持向量机 测井资料 

分 类 号：TE242[石油与天然气工程—油气井工程]