There are times we need something more powerful than conventional AI, which is described mainly by the data we feed it. We need to add other things we already know and only ask machine learning to learn what we don’t know. This idea of blending multiple AI components into a hybrid is what we call Evolved AI™ at Lone Star.
But what does that mean? Intelligence algorithms can be classified by how they predict outcomes and prescribe actions. Generally, domain context separates prescriptive algorithms from non-prescriptive.
Evolved AI is better than traditional machine learning.
Conventional (i.e., mainstream, data-driven) AI-like neural networks dates back to the 1960s, inspired by the mid-century understanding of natural intelligence. This works by matching the patterns in observed data. Predictions of future outcomes are limited to data the algorithm has “seen” in the past:
- They are fragile and cannot understand anything novel because they don’t have much context or rules
- They don’t know “why” predictions are made (black box)
- In most cases, this means they don’t know “what” you should do about it
- In IIoT machine learning, it could take multiple lifetimes to learn the failure modes of any given asset
Algorithms guided by knowledge of the domain (such as embedding Physics relationships) can predict an outcome and prescribe why outcomes are likely to occur, even if it has never been seen in prior data. These methods emerged in the 21st century.
Guided AI has shown significant promise in areas where:
- Data collection is limited, costly, and time-consuming to implement
- Some outcomes worth predicting do not happen very often, or have never happened… yet
- At least part of the domain is well understood with governing equations or causal relationships
- Machine learning can be focused on what we don’t know
Traditional Machine Learning has proven to be a challenging prospect for practical applications in oil and gas. Lone Star’s CEO Steve Roemerman ‘s presentation at the Machine Learning in Oil & Gas conference explores Evolved AI™ as a much more effective solution to maximize uptime for critical assets. He explains that for true predictive and prescriptive analytics in O&G assets, a hybrid approach to artificial intelligence using digital twins is the better way to manage and maintain these critical assets effectively.
You can watch the presentation below.
While you’re here, check out our MaxUp Energy – ESP Asset Analytics software and learn how the application helped one producer recover an incremental $684,000 annually (@ $25 / barrel oil) per 1,000 BOE produced per day and extend ESP mean time between failures by months.