丹佛的数据从各种来源涌入，公司越来越多地将数据分析嵌入到石油行业的工作流程中。

发言人在“将数据分析与石油工程工作流程集成：需要什么？”会议上表示，数据可用于包括地质导向在内的多种行业运营，但在使用数据之前，必须确认数据的质量。 6 月 13 日在“非常规资源技术会议”(URTeC) 上享用午餐。同样，人工智能 (AI) 在工作方式方面也发挥着更大的作用。雪佛龙公司的五年愿景是让工作流程得到人工智能辅助。

标准普尔全球上游工程专家总监 Alireza Haghighat 表示，通常钻井工程师想要利用数据分析做的第一件事就是预测。

“他们想要缩短周期时间。他们希望减少非生产时间并提高效率，”他说。

最终，团队的每个部分都从数据分析中得到了他们的愿望清单。能否准确满足这些需求是个问题。

他说，例如，完井工程师希望了解阶段的开始和结束位置，并能够预测井口压力和筛分。油藏工程师对递减曲线分析感兴趣。

“油藏工程师数据分析最广泛使用的应用是资产开发优化，特别是针对非常规资源，”他说。

他补充说，生产工程师希望预测液体负载以及设备何时需要维护。

但哈吉加特表示，在石油行业应用数据分析仍然面临挑战。

'数据是最重要的构建块。因此，数据质量、数据可用性、数据一致性和数据管理是一个问题，”他说。

他说，这意味着需要有适当的基础设施来处理数据。

雪佛龙油井数据科学策略师特拉维斯·克拉克 (Travis Clark) 表示，该公司正在努力为实时压裂顾问奠定基础。

“目前正处于起步阶段，”他说。

他说，成功的数据项目首先要正确地界定业务问题并了解成功是什么样子。该计划的支持也至关重要。

“我们创建了这个可以预测某些事情的模型，但我们无法推动采用，因为没有人相信它是可靠的。显然这是一场斗争，”他说。“为了获得这种可信度，你可以走一条路。”

克拉克说，这条道路需要专家了解模型的驱动因素，并诚实地了解模型的优点和缺点。

“数据极其关键，”他说。

数据质量再次成为最重要的部分。并且需要有足够的量。

西方石油公司的分析工程师 Yuxing Ben 表示，对于较小的数据集，质量更加重要。

“我看到了一些利用迁移学习来解决小数据问题的研究工作，”她说，并称这项工作“非常有趣”。

Haghighat 表示，有时，行业会将数据带给数据科学家并要求他们解决问题，有时行业必须对领域专家（例如钻井工程师）进行数据科学方面的教育。他说，问题是哪种方法最有意义。

“两者都有一点，”克拉克说。

他认为培养领域专家的数据科学技能很有价值，因为他们了解业务问题。

“他们可以从头到尾看到这个项目，对吧？他们有能力完成 POC（概念验证）并改进您想做的任何事情，”他说。

他说，模型构建既是一门艺术，也是一个迭代过程，“在这里你尝试一些东西，看看会发生什么，然后你回到起点并重新开始”，因此领域知识至关重要。另一方面，来自非能源背景的数据科学家将提供公正的观点，并以全新的眼光来看待问题。

数据科学、机器学习，这是一个快速变化的领域。克拉克说，因此，拥有一个可能刚刚学习并掌握最新技术的人可能会非常有帮助。“我认为归根结底是将两者结合起来。”

至于工作流程的未来，克拉克表示，雪佛龙的五年愿景是让它们“成为人工智能辅助的东西，无论它是什么样子。”

要实现这一目标，需要深入了解数据在工作流程中的使用方式。

“我们如何在工作流程中使用数据科学和人工智能来更好地规划油井？”他说。“当我们实际钻探时，我们如何使用实时模型来做出更好的决策？”

DENVER—As data floods in from a variety of sources, companies are increasingly embedding data analytics into oil industry workflows.

Data can be used for a plethora of industry operations, including geosteering, but before it can be used, the quality of the data must be confirmed, speakers said during the “Integrating Data Analytics with Petroleum Engineering Workflows: What is Needed?” lunch on June 13 at ​​Unconventional Resources Technology Conference (URTeC). Likewise, artificial intelligence (AI) is also playing a larger role in how things are done. Chevron Corp.’s five-year vision is for workflows to be AI-assisted.

Typically the first thing a drilling engineer wants to do with data analytics is prediction, Alireza Haghighat, director of upstream engineering specialists at S&P Global, said.

“They want to decrease the cycle time. They want to decrease nonproductive time and improve the efficiency,” he said.

Down the line, every part of the team has their wish list from data analytics. Whether those needs can be delivered with accuracy is the question.

Completion engineers, for example, want to know where stages start and end, with the ability to predict wellhead pressure and screen outs, he said. Reservoir engineers are interested in decline curve analysis.

“The most widely used application of data analysis for reservoir engineers is asset development optimization, specifically for unconventional resources,” he said.

Production engineers want predictions about liquid loading and when equipment needs to be maintained, he added.

But challenges remain in applying data analytics in the oil industry, Haghighat said.

“Data is the most important building block. So data quality, data availability, data consistency and data management is an issue,” he said.

That means there needs to be infrastructure in place for dealing with the data, he said.

Travis Clark, a wells data science strategist at Chevron, said the company is working to create the foundation for a real-time fracture adviser.

“We’re in the beginning stage of that,” he said.

Having a successful data project starts with properly framing the business problem and understanding what success looks like, he said. Buy-in for the program is also critical.

“We create this model that's going to predict something, but we can't drive adoption because nobody believes that it's reliable. Obviously that's a struggle,” he said. “To get that credibility, there's kind of a path that you can go down.”

That path involves experts understanding what is driving the model, as well as being honest about the model’s strengths and weaknesses, Clark said.

“The data is extremely critical,” he said.

Data quality, again, is the most important piece. And there needs to be enough.

Yuxing Ben, an analytics engineer for Occidental Petroleum, said quality is even more critical with smaller data sets.

“I've seen some research type work on using transfer learning to solve the small data issues,” she said, calling the work “very interesting."

Sometimes, Haghighat said, the industry will bring data to the data scientist and ask them to solve a problem, and sometimes the industry has to educate domain experts—such as drilling engineers—on data science. The question, he said, is which approach makes the best sense.

“It's a little of both,” Clark said.

He sees value in building the data science skills of domain experts because they understand business problems.

“They can kind of see the project from start to finish, right? They have the skills to go through, really do a POC [proof of concept] and improve whatever you're trying to do,” he said.

Model building, he said, is both an art and an iterative process “where you try something and see what happens and you go back to the beginning and start over,” so domain knowledge can be critical. On the other hand, data scientists coming from a non-energy background will offer an unbiased view and come at the problem with a fresh set of eyes.

“Data science, machine learning, it's a rapidly changing field. So having somebody who's probably just studied and is kept up to date on new techniques” could be really helpful, Clark said. “I think it comes down to combining both.”

As for the future of workflows, Clark said Chevron’s five-year vision is to have them “be kind of AI-assisted, whatever that looks like.”

Getting there requires deeply understanding how data is being used within workflows.

“How are we using data science and AI within the workflows to plan wells better?” he said. “How are we doing, using the real-time models when we're actually drilling to make better decisions?”