Estimating oil and gas well life cycle using machine learning

在全球范围内，加拿大是第四大石油和天然气生产国，石油和天然气占加拿大GDP的5.6%。不列颠哥伦比亚省是加拿大第二大能源生产国，仅次于阿尔伯塔。卑诗省石油和天然气委员会（BCOGC）的目的是规范石油和天然气活动，以提供良好的发展。它审查许可证申请，以确保通过考虑对经济，环境，土著民族，林业和考古学的影响，符合公众的最佳利益。本项目的重点是预测石油和天然气威尔斯井何时将变得不活跃（由于产量下降、储量枯竭、价格下跌、运输系统中断或水产量过剩），以计划恢复现场。如果能够预测较长期的恢复速度，将改进风险预测，并提高委员会估计石油和天然气活动未来累积影响的能力。立法强制执行关闭休眠站点的时间表，BCOGC休眠和关闭法规旨在到2036年恢复10，000至11，000个休眠站点。此外，当一个遗址所有者不再可用时，例如由于破产，遗址恢复费用由该省承担（在过去5年中，不列颠哥伦比亚省约有700个遗址被指定为孤儿）。目前，BCOGC可以预测到2023年，所有休眠站点何时需要恢复，但无法预测当前活跃站点何时会进入休眠状态。该项目利用BCOGC数十年的数据以及UBC在机器学习和优化方面的专业知识，旨在创建新的预测模型，这些模型是可解释的，可解释的和公平的，同时保持可接受的准确度。一个关键目标是让所有利益相关者（BCOGC管理人员，土著社区，石油和天然气运营商）了解模型预测及其局限性。短期目标是通过在提交许可证申请时预测威尔斯井的生命周期，更好地管理该省的环境债务。长期目标是通过预测孤威尔斯井的数量来降低该省的金融风险。这样的预测将使BCOGC能够更好地管理与孤儿威尔斯井相关的省财政债务。

Globally, Canada is the 4th largest producer of oil and gas combined with oil and gas representing 5.6% of Canada's GDP. British Columbia is the 2nd largest producer of those energies in Canada after Alberta. The BC oil and gas commission (BCOGC) purpose is to regulate oil and gas activities to provide for a sound development. It reviews applications for permits to ensure they are in the public best interest by considering the impact on the economy, the environment, Indigenous Nations, forestry, and archaeology. The present project focuses on predicting when oil and gas wells will become inactive (due to decreased production, depleted reserves, depressed pricing, interruptions to transportation systems, or excess production of water) to plan the restoration of the site. Being able to project the rate of restoration over a longer term will both improve risk forecasting and the ability of the Commission to estimate the future cumulative effects of oil and gas activity. Legislation enforces timelines to shutdown dormant sites, and BCOGC dormancy and shutdown regulation aims at restoring 10,000 to 11,000 dormant sites by 2036. Moreover, when a site owners is no longer available, e.g. due to bankruptcy, site restoration costs falls on the province (over the past 5 years, around 700 sites were designated as orphan in BC). At this time, BCOGC can project when site restoration will be required for all dormant sites up to 2023, but cannot predict when currently active sites will become dormant. Leveraging BCOGC decades of data with UBC expertise in machine learning and optimization, the project aims to create new predictive models that are explainable, interpretable, and fair while maintaining an acceptable degree of accuracy. A critical objective is for all stakeholders (BCOGC managers, Indigenous communities, oil and gas operators) to understand the model predictions and its limitations. The short term goal is to better manage the province environmental debt by predicting wells life cycles when a permit application is submitted. The longer term goal is to reduce the province financial risk by predicting the number of orphaned wells. Such predictions will allow BCOGC to better manage the province financial debt associated with orphan wells.