In the process of shale gas production, it is of great significance to select an appropriate mathematical method to accurately predict the gas flow rate of wellhead choke for the rational formulation of shale gas well production plan. Gas-liquid two-phase flow occurs in most of the time from the flowback to the production period in shale gas wells. Wellhead chokes play key roles in regulating the flowing rates of both the flowback fluid and shale gas. Therefore, it is important to study the law of two-phase choke flow clearly so as to accurately predict gas flow rate through wellhead chokes. Up to now, previous studies have proposed a variety of applicable empirical methods, including Gilbert-type correlation (GC), artificial neural network (ANN) and support vector machine (SVM). The analysis of training data and the establishment of accurate prediction models determine the accuracy of prediction. In this study, Gaussian process regression (GPR) was adopted to learn and predict the behavior of gas-liquid two-phase flow through wellhead chokes, and huge amounts of data collected from Chuannan Shale gas wells were used to verify the effectiveness of the GPR method. The prediction accuracy of the GPR method was compared with those of other methods like GC, ANN and SVM. In addition, we also compared the prediction accuracy of different kernel functions to select the best kernel function for GPR. The kernel functions considered are exponential function, squared exponential function, rational quadratic function and Matérn function. The results showed that GPR method is accurate and applicable for analyzing the behavior of gas-liquid two-phase flow through wellhead chokes, and GPR method with exponential kernel function could achieve greater prediction accuracy than other kernel functions.

在页岩气生产过程中，选择合适的数学方法准确预测井口节流阀的气体流量，对于合理制定页岩气井生产计划具有重要意义。页岩气井从返排到生产期大部分时间都发生气液两相流动。井口节流阀在调节回流流体和页岩气的流速方面起着关键作用。因此，深入研究两相节流流动规律，对准确预测通过井口节流的气体流量具有重要意义。迄今为止，已有研究提出了多种适用的实证方法，包括吉尔伯特型相关（GC）、人工神经网络（ANN）和支持向量机（SVM）。训练数据的分析和准确预测模型的建立决定了预测的准确性。本研究采用高斯过程回归（GPR）学习和预测气液两相流通过井口节流阀的行为，并利用川南页岩气井采集的大量数据验证了探地雷达的有效性。方法。将 GPR 方法的预测精度与 GC、ANN 和 SVM 等其他方法的预测精度进行了比较。此外，我们还比较了不同核函数的预测精度，以选择最佳的 GPR 核函数。考虑的核函数有指数函数、平方指数函数、有理二次函数和 Matérn 函数。