In this study, the severity of slugging is assessed by predicting maximum slug lengths (MSL) quickly using the random forest (RF) algorithm based on the geometric features of well trajectories for a shale gas field. Severe slugging is one of the critical issues production engineering-wise because it causes operation shut-down. Thus it should be predicted for proactive measurements. A total of 5033 well trajectories were acquired from the northeastern area of British Columbia, Canada. The well trajectories are described using ten geometric features such as X, Y, and Z lengths in the Cartesian coordinate system, inclination, azimuth, and the other five. The 5033 well trajectories are grouped using the k-medoids clustering algorithm. The well trajectories in each group and the groups are compared visually to see if the ten features are representative enough to describe the geometric features of the well trajectories. The ten geometric features of the well trajectories are used as the input for RF, and MSL, which represents the severity of slugging, is used as the output for RF. The output data is simulation results by a pipe flow simulator, OLGA. The trained RF model provides the satisfactory prediction performance of MSL (R values, 0.866 and 0.857 for training and test data, respectively). In the trained RF model, X, Y, and Z lengths have the most significant importance among the ten geometric features. Because it is impractical to simulate all well trajectory scenarios by OLGA, the MSL values are projected onto a 3-dimensional map of which axes are X, Y, and Z lengths to visualize the trend of MSL. The 3-dimensional map showing the relation between MSL and the geometric features of well trajectories can be utilized as a quick reference to avoid severe slugging in designing well trajectories.

在这项研究中，通过基于页岩气田井轨迹几何特征的随机森林（RF）算法快速预测最大段塞长度（MSL）来评估段塞的严重程度。严重的段塞是生产工程方面的关键问题之一，因为它会导致操作关闭。因此，应该预测主动测量。从加拿大不列颠哥伦比亚省东北部地区共获取了 5033 条井轨迹。井轨迹使用十个几何特征来描述，例如笛卡尔坐标系中的 X、Y 和 Z 长度、倾角、方位角和其他五个。使用 k-medoids 聚类算法对 5033 口井轨迹进行分组。将每组和各组的井轨迹进行视觉比较，看这十个特征是否具有足够的代表性来描述井轨迹的几何特征。井轨迹的 10 个几何特征作为 RF 的输入，MSL 表示段塞的严重程度，作为 RF 的输出。输出数据是管道流量模拟器 OLGA 的模拟结果。经过训练的 RF 模型提供了令人满意的 MSL 预测性能（训练和测试数据的 R 值分别为 0.866 和 0.857）。在经过训练的 RF 模型中，X、Y 和 Z 长度在十个几何特征中具有最重要的重要性。因为通过 OLGA 模拟所有井轨迹场景是不切实际的，所以将 MSL 值投影到轴为 X、Y、和 Z 长度以可视化 MSL 的趋势。显示 MSL 与井轨迹几何特征之间关系的 3 维图可作为快速参考，以避免在设计井轨迹时出现严重的段塞。