Abstract Gas kick is generally difficult to discover in time using traditional surface detection methods, which results in a significant wastage of time and money. Owing to the restriction of the low data-transmission speed of measurement-while-drilling system, downhole measured data is usually ignored in gas-kick detection. Furthermore, surface detection methods comprising the use of pressure and flow-rate sensors require professional knowledge and many input parameters, some of which are required to be assumed. In this study, we used downhole dual measurement points for detecting gas kick without the use of other surface input parameters. Firstly, we developed an end-to-end supervised neural network to determine the still and circulation working conditions, which were used for calculating the drilling fluid density and viscosity. Secondly, an unscented Kalman filter was applied to perform a backward gas fraction calculation dynamically. However, this downhole calculation method cannot be used in highly deviated and horizontal wells. Because there is a downhole fluctuating pressure generated during the rock breaking, we proposed an auxiliary gas-kick detection method based on the theory of pressure wave attenuation. This method can be applied to all well types. To evaluate the proposed gas-kick detection method, we used a gas-liquid flow simulation model combined with a pump rate model, screw-drilling-tool pressure-consumption model, rock-breaking model, and formation permeability model to generate transient data with the highest possible accuracy. The advection upstream splitting model was used as the numerical scheme. The accuracy of the simulation model was successfully validated using two field experimental data sets. Finally, we generated a set of vertical and horizontal well data each with the simulation model to test the gas-kick detection method. The experiment results showed that the proposed gas-kick detection model was successful in detecting gas kick and obtaining the accurate gas fraction.

中文翻译：

---

水基钻井液井下双测点井下气爆瞬态模拟与检测

摘要 使用传统的表面检测方法通常很难及时发现气涌，这导致了大量的时间和金钱浪费。由于随钻测量系统数据传输速度低的限制，在气爆检测中通常忽略井下测量数据。此外，包括使用压力和流量传感器的表面检测方法需要专业知识和许多输入参数，其中一些需要假设。在这项研究中，我们使用井下双测量点来检测气涌，而不使用其他地面输入参数。首先，我们开发了一个端到端的监督神经网络来确定静止和循环工作条件，用于计算钻井液密度和粘度。第二，应用无迹卡尔曼滤波器来动态执行反向气体分数计算。但是这种井下计算方法不能用于大斜度井和水平井。由于破岩过程中会产生井下脉动压力，我们提出了一种基于压力波衰减理论的辅助气爆检测方法。该方法适用于所有井型。为了评估所提出的气爆检测方法，我们使用气液流动模拟模型结合泵速模型、螺旋钻具压力消耗模型、破岩模型和地层渗透率模型来生成瞬态数据。尽可能高的准确度。对流上游分裂模型用作数值方案。使用两个现场实验数据集成功验证了模拟模型的准确性。最后，我们生成了一组垂直和水平井数据，每个数据都带有模拟模型来测试气爆检测方法。实验结果表明，所提出的气涌检测模型能够成功地检测到气涌并获得准确的气馏分。