基于集成学习的井漏智能预警模型及智能推理方法

Intelligent early warning model and intelligent reasoning method based on integrated learning for loss circulation

张矿生 宫臣兴 陆红军 欧阳勇 辛庆庆

ZHANG Kuangsheng GONG Chenxing LU Hongjun OUYANG Yong XIN Qingqing

中国石油长庆油田公司 中国石油长庆油田公司油气工艺研究院 中国石油长庆油田公司低渗透油气田勘探开发国家工程实验室

Changqing Oilfield Company, PetroChina, Xi’an 710018, Shaanxi, China Research Institute of Oil and Gas Technology, Changqing Oilfield Company, PetroChina, Xi’an 710018, Shaanxi, China National Engineering Laboratory for Exploration and Development of Low Permeability Oil and Gas Fields, Changqing Oilfield Company, PetroChina, Xi’an 710018, Shaanxi, China

水平井优快钻井技术加快了长庆油田致密油气大规模开发速度，但由于部分区域地层孔隙和微裂缝发育，承压能力低，极易发生井漏风险，严重制约了水平井提速提效。为进一步降低井漏风险，提高漏失预警的时效性和准确率，提出了一种因果关系约束下的井漏智能预警及漏失原因推理方法。基于漏失产生机理，分析了漏失风险的表征参数及其变化规律，将其作为输入参数约束条件，利用工况识别模型和特征变化规律准确定位井漏时间，基于BP神经网络和LSTM长短时记忆网络建立漏失风险预警模型，利用因果推断算法解释模型预警原因，结合风险机理实现警报约束。研究结果表明，LSTM集成网络井漏预警准确率达95.6%，基于集成学习的智能推理方法能够准确解释预警发生原因，对钻井现场及时采取井漏防范措施，保障水平井优快钻井具有重要意义。

The optimal and fast drilling techniques for horizontal wells has accelerated the large-scale development of tight oil and gas in Changqing Oilfield. However, due to the development of formation pores and micro-fractures in some areas, the pressure bearing capacity of the formation is low, and the risk of lost circulation is extremly high, which seriously restricts the speed and efficiency when drilling horizontal wells. In order to further reduce the risk of loss circulation, and improve the timeliness and accuracy of loss circulation warning, when lost circulation occurs, a method for intelligent early warning and reasoning under the constraint of causality was proposed. Based on the mechanism of loss circulation, the characteristic parameters of loss circulation risk and their changing rules are analyzed, which are used as input parameter constraints to accurately locate the occurrence time of loss circulation by using the working condition identification model and characteristic changing rules. Based on BP neural network and long-short-term memory network (LSTM), a risk early warning model for loss circulation was established, and the causes of the early warning were explained by using the causal inference algorithm, realizing warning constraint combining with the risk mechanism. The results show that the early warning accuracy rate of the LSTM integrated network on loss circulation reaches 95.6%, and the intelligent reasoning method based on integrated learning can accurately explain the causes of the early warning, which is of great significance for timely adopting measures to prevent loss circulation at the drilling site, and of great significance for ensuring the optimal and fast drilling of horizontal wells.

10.13639/j.odpt.2023.01.007