基于智能算法的油气田地应力三维预测

Three-dimensional Prediction of In-situ Stress in Oil and Gas Field Based on Intelligence Algorithms

袁多 吴超 卢运虎 张东清

YUANDuo WUChao LUYunhu ZHANGDongqing

页岩油气富集机理与有效开发国家重点实验室, 北京 昌平 102206 中国石化石油工程技术研究院, 北京 昌平 102206 中国石油大学(北京)石油工程学院, 北京 昌平 102249

State Key Laboratory of Enrichment Mechanism and Effective Exploitation of Shale Oil and Gas, Changping, Beijing 102206, China SINOPEC Research Institute of Petroleum Engineering, Changping, Beijing 102206, China School of Petroleum Engineering, China University of Petroleum, Changping, Beijing 102249, China

为了解决复杂沉积构造环境导致未钻区域的地应力定量预测难度大的问题，根据层速度、地应力、叠后地震信息之间的定量关系，运用BP神经网络、模拟退火等智能算法提出了用于不同工况条件的两种油气田地应力三维预测方法。在完钻井数量较多、实测信息较丰富的工区使用BP神经网络算法，利用地震数据空间速度信息与岩石力学方法建立地应力三维数据体；在实测数据较少的工区，运用模拟退火算法直接搜寻合成与实际地震记录达最优匹配下的地应力解向量。该技术在东部某油田的主要工区进行了现场应用，得到了具有较高精度和分辨率的地应力预测结果，验证了基于智能算法的油气田地应力三维预测方法的可行性。

Complex sedimentary and tectonic environment results in the difficulty of predicting in-situ stress quantitatively. Two new three-dimensional in-situ stress prediction methods for different working conditions based on intelligence algorithm are proposed using the relationships between interval velocity, in-situ stress and post stack seismic information. With abundant well logging and test data, in-situ stress is predicted by the rock physics and mechanics method based on spatial velocity information obtained using BP neural network algorithm. With sparse well points and a small number of data, simulated annealing algorithm is utilized to directly search for the stress solution vector with the best match of synthetic and actual seismic records. These methods have been applied in an oilfield in eastern China and achieves high accuracy and resolution prediction results, which proves the feasibility of these methods.

10.11885/j.issn.1674-5086.2019.03.27.03