基于小样本机器学习的海上低渗储层横波速度预测

Prediction of shear wave velocity in offshore low permeability reservoir based on small sample machine learning

横波速度信息对于海上古近系扇三角洲低渗储层精细预测极为重要,但由于成本和采集技术的局限性,有效获取横波速度信息成为低渗储层精细刻画中亟需解决的问题之一。实际低渗有效储层的多样性和沉积环境的复杂性导致基于数据驱动的经验公式法和物理规律驱动的岩石物理模型法进行海上小样本横波速度预测都存在不足。为此,开展基于高斯过程回归机器学习算法研究,其算法具有训练数据需求小、预测精度高和可实现对结果进行不确定性评价等优点。以渤海M油田古近系低渗区的测井数据曲线为应用对象,结果表明基于高斯过程回归机器学习算法可快速实现弹性波速度预测,并可实现对预测结果不确定性的量化分析。

The shear wave velocity information is very important for the precise prediction of low permeability reservoirs in offshore Paleogene fan delta.However,due to the limitations of cost and acquisition technology,effective acquisition of shear wave velocity information has become one of the urgent problems to be solved in the fine characterization of low permeability reservoirs.Meanwhile,due to the diversity of low permeability reservoirs and the complexity of sedimentary environment,lead to deficiencies in predicting small samples shear wave velocity based on the data-driven empirical formula method and the physical-law-driven petrophysical model method.Therefore,the research on machine learning algorithm based on Gaussian process regression is carried out.The algorithm has the advantages of small training data demand,high prediction accuracy and uncertainty evaluation of results.Taking the log data curve of Paleogene low permeability area in Bohai M oilfield as the application object,the results show that the machine learning algorithm based on Gaussian process regression can quickly predict the elastic wave velocity,and can realize the quantitative analysis of the uncertainty of the prediction results.