Abstract: As hydrocarbon exploration and development extends into deep formations and unconventional reservoirs, the increasingly complex downhole conditions have posed greater challenges to the safety of drilling operations. To achieve safe and efficient well drilling, real-time acquisition of wellbore engineering parameters is imperative for accurate identification of complex downhole conditions, thereby guiding the prevention and remediation of drilling accidents. Based on the demand for quantitative detection of downhole gas kick, a self-developed miniaturized, low power, high-temperature and high-pressure rated downhole mobile micro-measurer is applied for full-borehole trajectory data acquisition; based on changes in its dynamic signals, the study classifies and quantifies the bubble flow regimes across different well sections following gas kick. Further, a hydrodynamic model for the micro-measurer moving in fluid media was established. Through analysis of hydrodynamic characteristics under the impact of different bubble flows, this study reveals the correlation between the micro-measurer's dynamic response and downhole bubble flow intensity. On this basis, gas kick experiments were conducted using a self-developed wellbore device, and deep learning-based analysis was performed on the measured dynamic data of the micro-measurer. Finally, a two-stage end-to-end neural network model integrating classification and regression was constructed, enabling the quantitative analysis of downhole gas influx volumes.

Key words: micro-measurer, hydrodynamic analysis, deep learning, gas kick, quantitative detection

摘要： 随着中国油气勘探开发逐步向深部地层和非常规油气层挺进,井下愈加频发的复杂工况给钻井作业安全带来了极大挑战。为实现安全高效钻井,需要及时获取井筒工程参数,进而对井下复杂工况进行准确识别,以指导井下复杂工况的预防和处理。基于井下气侵定量检测的需求,采用自研的小型化、低功耗、耐高温高压的井下移动式微型测量器进行全井筒的沿程数据采集,并基于其动力学信号的变化进行气侵后不同井段气泡流状态的分类与统计。构建了微型测量器在流体中运移时的水动力学模型,并通过对微型测量器在不同气泡流冲击下的水动力学特征分析,得到了微型测量器动力学响应与井下气泡流强度之间的对应关系;在此基础上,利用自研的井筒装置开展了气侵实验测试,并对微型测量器的实测动力学数据进行深度学习;最终,构建了包含分类和回归的两级端到端神经网络模型,实现了对井下气侵量的量化分析。

关键词: 微型测量器, 水动力学模型, 深度学习, 气侵, 定量检测