When the phrase “Let’s use ML” is uttered, this often immediately raises an element of fear, along with questions such as “How do we do that?” and “Who will do it?" In reply, many assume a “divide and conquer” approach is required, with data scientists handling complex algorithms to deliver models, and subject matter experts (SMEs) then using these models and making changes based on operating conditions.
While seemingly ideal, this approach is inherently flawed because of the gap between the initial model and the domain expertise of the SME. It is further weakened when considering the reality of how people make decisions. Human nature requires confidence in the action suggested by the ML outcome, else little or no action is taken.
If SMEs do not believe a model has the appropriate inputs or is unclear about how the model works, they are unlikely to act. Even when there is confidence in the model, the required action is often unclear, so the model result must be combined with other evidence and information at the SMEs’ disposal to see how it corroborates their intuition.
Proponents of this approach claim it is necessary due to the complexity of models and the expertise required for analysis. Similarly, there was a time when statistical modelling methods were considered new and complex. However, nowadays SMEs do not avoid the use of statistical models, but instead use these and other modelling approaches daily, while leveraging model-building experts as needed to build confidence in their workflows and recommendations.
Since we do not claim that SMEs should not use statistical models simply because there are statisticians around, then why should we assume they cannot drive use of ML as a natural part of their daily workflow? With the right tools and mindset, there is an opportunity to couple human reasoning with the ML functionality found in advanced analytics applications. The key is putting the data and the ML algorithms directly at the fingertips of SMEs who have the required domain knowledge.
This article provides use cases showing how this can be done by highlighting the role of advanced analytics and ML to drive continuous improvement and sustainability in oil and gas applications—whether upstream, midstream, or downstream. But before delving into use cases, it would be useful to examine the connection between analytics and ML.
Analytics and machine learning
The combination of massive data stores and process innovation has unquestionably changed the industrial world. In recent years, the tools applied to descriptive, diagnostic, predictive, and prescriptive analytics have expanded to include ML. In other words, describing what is happening, why is it happening, what will happen, and what should be done about it, respectively.
Perhaps as important when reacting to the idea of using ML is understanding the fundamental question of “What is it?” For starters, ML is a tool – a specific kind of artificial intelligence that includes the scientific study of algorithms and statistical models.
Machine learning has also been described as “the science of getting computers to learn and act like humans do, and improve their learning over time in autonomous fashion, by feeding them data and information in the form of observations and real-world interactions.”1 This idea of providing observations and real-world interactions is exactly why SMEs should be working directly with the data and the models themselves.
Commonalities between the human mind and ML technology include:
- Drawing on compiled patterns.
- Executing responses and generating intuition.
- Taking the opportunity to learn regularities through practice.
- Needing to experience variability while learning.
- Approximating large numbers of items.
- Understanding the reasoning behind the intuition.
- Seeing patterns in randomness.
Successful use of data analytics and ML requires connectivity to data historians and other sources, and feedback loops. Data historians remain an important element in any data analytics strategy, and are changing to provide more agile data storage, including new cloud deployments.
Feedback mechanisms continually improve SME expertise and ML algorithms and are thus critical to address the agile nature of processes.
This is part one of a two-part article. Part two is available to read here: https://www.oilfieldtechnology.com/digital-oilfield/05052020/harnessing-machine-learning--part-two/.
Read the article online at: https://www.oilfieldtechnology.com/digital-oilfield/04052020/harnessing-machine-learning--part-one/