成都理工大学毕业设计(论文)I测井时间序列的支持向量机回归预测摘要统计学习理论是针对小样本情况下的机器学习理论,其核心思想是通过控制学习机器的复杂度实现对学习机器推广能力的控制。支持向量机能够尽量提高学习机的推广能力,即使由有限数据集得到的判别函数对独立的测试集仍能够得到较小的误差。因此,本文把支持向量机用于测井时间序列的回归预测。首先,介绍了时间序列和支持向量机的基础理论。其次,详细介绍了支持向量机的回归原理和算法。最后,本文根据石油地质勘探的实际问题,将支持向量机运用测井曲线预测储层参数——孔隙度。结果表明,该方法预测精度高,方法稳定有效。支持向量机较好的解决了小样本测井勘探的实际问题。关键词:支持向量机;时间序列;回归预测成都理工大学毕业设计(论文)IILoggingtimeseriessupportvectormachineregressionAbstract:Statisticaltheoryisacaseofmachinelearningtheorywhichisbasedonsmallsample.It’scoreideaisthemachinebycontrollingthecomplexityoflearningtoachievethepromotionoftheabilityoflearningmachinecontrol.Supportvectormachinetomaximizethegeneralizationabilityoflearningmachine,evenifalimiteddatasetobtainedfromthediscriminantfunctionontheindependenttestsetwillbesmallerstillerror.Therefore,thesupportvectormachineisusdtologgingtimeseriesregression.Firstofall,thisarticleintroducesthetheoryofthetime-seriesandthebasisofsupportvectormachine.Second,itintroducesdetailedinformationonthereturnofsupportvectormachinetheoryandalgorithm.Finally,thisarticleinaccordancewiththeactualgeologicalexplorationofoilwillbetheuseofsupportvectormachinepredictionofreservoirparameterslogging-porosity.Theresultsshowthathighpredictionaccuracyofthemethod,astableandefficientmethod.Supportvectormachinetoresolvebetterthesmallsampleofthepracticalproblemsloggingexploration.Keywords:supportvectormachines;timeseries;regression成都理工大学毕业设计(论文)III目录第1章前言............................................................................................................11.1选题意义............................................................................................................11.2研究现状............................................................................................................11.3论文内容............................................................................................................2第2章测井时间序列................................................................................................32.1时间序列概述....................................................................................................32.2时间序列的预测方法........................................................................................42.2.1时间序列线性预测方法.............................................................................42.2.2时间序列的非线性预测方法.....................................................................52.2.3自回归移动平均(ARMA)模型..................................................................62.2.4季节型模型...............................................................................................10第3章支持向量机的原理和方法..........................................................................113.1SVM的基本思想............................................................................................113.1.1最优分类面...............................................................................................113.1.2广义的最优分类面...................................................................................133.2支持向量回归..................................................................................................143.2.1SVM回归原理.........................................................................................143.2.2线性支持向量回归...................................................................................143.2.3非线性支持向量回归...............................................................................153.2.4支持向量回归....................................................................................163.2.5v-支持向量回归.......................................................................................183.2.6时间序列分析...........................................................................................193.3支持向量算法..................................................................................................203.3.1支持向量机的训练算法...........................................................................203.3.2C-SVM算法及其变形算法.....................................................................21成都理工大学毕业设计(论文)IV第4章测井时间序列的支持向量机回归预测........................................................254.1引言..................................................................................................................254.2应用实例..........................................................................................................26结论..........................................................................................................................42致谢..........................................................................................................................43参考文献......................................................................................................................44成都理工大学毕业设计(论文)1第1章前言1.1选题意义本课题的主要目的是研究支持向量机预测储层岩性参数问题。在估计孔隙度的过程中,测井的数目往往是固定且有限的,支持向量机在解决小样本问题中表现出许多特有的优势SVM方法的几个主要优点有:1.是专门针对有限样本情况的,其目标是得到现有信息下的最优解而不仅仅是样本数趋于无穷大时的最优值;2.算法最终将转化成为一个二次型寻优问题,从理论上说,得到的将是全局最优点,解决了在神经网络方法中无法避免的局部极值问题;3.算法将实际问题通过非线性变换转换到高维空间,在高维空间中构造线性逼近函数来实现原空间中的非线性逼近函数,特殊性质能保证学习机有较好的推广能力,同时,它巧妙地解决了维数问题,使其算法复杂度与维数无关。对于小样本的分类问题,SVM具有调