Abstract Gas leakage through the cemented section of a gas well, from a producing zone to other zones or to the open air, poses serious threats to safety and the environment. A number of design variables during drilling and cementing jobs may possibly contribute to such leakage. Decisions on these variables are best made during the design and well construction phase, as remedial operations after the well begins production have limited success rate. Therefore an approach that avoids the problem by ensuring robust zonal isolation during well construction jobs is more suitable. Such an approach involves decisions on a fairly large number of design variables. Building a model based on first principles to predict the effect of all of these variables on leakage is a formidable task. An alternative examined in this paper relies on using multivariate statistics to build an empirical model from available data. The model can then be used to make decisions on design variables such that leakage is avoided. The proposed approach is explained using data from 105 gas wells. The model built predicts leakage with about 75% accuracy in cross-validation tests. In addition, it ranks decision variables in the order of importance and suggests which ones need to receive more attention. The approach presented can be extended to include additional variables for which data is available.

中文翻译：

---

胶结气井分区隔离的数据驱动建模方法

摘要 气井固井段的气体泄漏，从产区到其他区或露天，对安全和环境构成严重威胁。钻井和固井作业期间的许多设计变量可能会导致这种泄漏。关于这些变量的决定最好在设计和建井阶段做出，因为井开始生产后的补救操作成功率有限。因此，通过在建井作业期间确保强大的区域隔离来避免该问题的方法更合适。这种方法涉及对大量设计变量的决策。建立基于第一性原理的模型来预测所有这些变量对泄漏的影响是一项艰巨的任务。本文中研究的替代方案依赖于使用多变量统计从可用数据构建经验模型。然后可以使用该模型对设计变量做出决策，从而避免泄漏。建议的方法使用来自 105 口气井的数据进行解释。构建的模型在交叉验证测试中以约 75% 的准确率预测泄漏。此外，它还按重要性对决策变量进行排序，并建议哪些需要受到更多关注。所提出的方法可以扩展到包括可用数据的其他变量。构建的模型在交叉验证测试中以约 75% 的准确率预测泄漏。此外，它还按重要性对决策变量进行排序，并建议哪些需要受到更多关注。所提出的方法可以扩展到包括可用数据的其他变量。构建的模型在交叉验证测试中以约 75% 的准确率预测泄漏。此外，它还按重要性对决策变量进行排序，并建议哪些需要受到更多关注。所提出的方法可以扩展到包括可用数据的其他变量。