基于机器学习的平面剩余油分布预测方法

Prediction methods of remaining oil plane distribution based on machine learning

老油田在长期开发过程中积累了大量的数据资源,为机器学习技术应用提供了基础。以深入挖掘数据资源内在关系为目的,提出基于机器学习的剩余油分布预测新方法。首先以测井解释成果、油藏工程理论计算和多套油藏数模结果为基础数据,开展数据融合和处理,给出12个维度参数的具体计算方法,形成样本资料库;利用支持向量机和长短期记忆神经网络模型分别开展见水波及识别和剩余油分布预测训练,搭建剩余油预测模型,实现在输入储层物性参数、油水流动特征参数和生产参数的情况下,简单快速预测油藏平面剩余油分布的目的。测试表明,新预测模型计算的剩余油饱和度与数值模拟计算结果相比,预测准确率达到96%。

A large amount of geological and production data can be accumulated in mature oilfields during a long-term development process, which can provide a foundation for the application of machine learning method. In this paper, a method for the prediction of remaining oil distribution was proposed based on machine learning. First of all, well logging data, reservoir engineering calculation and numerical simulation results were comprehensively used to carry out data fusion, and a specific calculation method for 12 dimension parameters was given to obtain a basic sample database for the prediction of remaining oil distribution. Then the support vector machine and long-short term memory neural network models were used to conduct the training of water breakthrough identification model and remaining oil distribution prediction model respectively, in order to realize a rapid prediction of the change of remaining oil distribution in the reservoir with the provided petrophysical, oil-water flow and production data. The comparison between the remaining oil saturation calculated by the proposed model and the widely used reservoir simulation software shows that the prediction accuracy of the proposed model can be over 96%.