基于数据驱动的深水钻井溢流智能识别与反演解释研究

The identification and interpretation of downhole influx, closely linked with the real-time detection of drilling information and the precise control of wellbore pressure, serve as the key technology to ensure the operational safety and efficiency in deepwater drilling. Currently, there exist many problems which are in urgent need to be solved, such as low utilization rate of the measured data, delayed identification and inaccurate interpretation of downhole influx, and the subsequent difficulty in wellbore pressure control accordingly. Therefore, this project will include the following researches: supported by the integration of both the real-time and historical data of deepwater drilling in the west of the south China sea, the influx identification model based on cluster analysis and deep neural network will be established with the aim to uncover the correlation between influx events, drilling condition, multi-source parameters and multi-characteristics, to further achieve rapid and accurate automatic identification of influx under different drilling conditions; the influx inverse interpretation method based on forward model of multiphase flow in wellbore under multiple factors and adaptive unscented kalman filter will be proposed to conduct real-time inversion of the three downhole parameters: influx location, influx rate and height of the gas front with the purpose of moving forward the real-time and accurate quantitative interpretation of the influx condition. Thus, this project, aiming at establishing a theory and method of identification and interpretation of the deepwater drilling influx to help reduce the well control risks and prevent major accidents such as blowout, will have great significance to promote the exploitation of deepwater oil and gas resources in China.

井下溢流识别与解释连接着钻井信息的实时检测和井筒压力的精确控制，是保障深水钻井安全高效作业的关键技术。目前深水钻井面临测量数据利用率低、溢流识别不及时、解释不准确、后期控制困难等瓶颈问题，亟待解决。本项目拟通过整合南海西部深水钻井数据，结合历史与实时数据，建立基于聚类分析与深度神经网络的深水钻井溢流识别方法，揭示溢流事件-钻井工况-多源参数多特征之间的相关性，实现不同钻井工况下溢流的快速、准确的自动识别；建立基于自适应无迹卡尔曼滤波与多因素耦合下井筒多相流正演模型的深水钻井溢流反演解释方法，实时反演溢流位置、溢流速率以及气体前缘高度三类井下参数，进一步实现溢流工况的实时准确的量化解释。研究成果旨在形成一套深水钻井溢流识别与解释理论与方法，有助于降低井控风险，预防井喷等重大事故的发生，对于推进我国深水油气资源开发具有重要意义。

随着钻井作业向深水复杂地层的不断深入，钻井风险、井下事故也在呈几何级数增长，这就要求我们能够快速、准确的识别井下复杂工况。其中，深水溢流发生频率高，处理时间长，危害尤为严重。目前深水溢流工况识别仍然面临着以下测量数据利用率低、数据挖掘不充分、解释不准确、后期控制困难等瓶颈问题，亟待解决。基于此，本项目采用理论研究、实验研究及数值模拟相结合的方式，以数据分析与挖掘、人工智能及反演理论为突破口，整合我国南海西部钻井数据，取得了以下创新成果：①结合深度学习和自编码器技术，建立了基于深度自编码器的钻井工况智能识别模型，对钻进、循环、起钻、下钻、倒划眼、接单根、钻水泥塞、短起下钻和复杂情况9类工况进行实时识别。②建立了综合长短期记忆神经网络和多头注意力机制的深水钻井溢流识别模型，通过自我反馈训练对模型参数进行优化，实现了溢流的提前预警和识别。③综合考虑气体组分与溶解性、高温高压对流体热物性参数的影响以及钻头连续破岩等因素，建立了基于移动边界的深水钻井井筒多相瞬态流动正演模型，揭示了深水井筒多相流动和传热耦合作用机理。④结合井筒多相流正演模型与自适应无迹卡尔曼滤波技术，建立了井下溢流实时反演解释模型，实时反演溢流位置、溢流速率、气体前沿高度等井下参数，实现了对于井下溢流工况的准确量化解释。研究成果有助于完善深水钻井溢流识别与解释理论，预防井喷等重大事故的发生，指导后期井控作业，为我国深水钻井安全高效作业提供技术支撑。