Machine Learning and other Artificial Intelligence techniques are driving significant innovations in the geoscience community. The growing use of this technology offers the oil industry a unique way to deal with the massive data deluge that we have been observing over the past 15 years. As opposed to many other industries, the oil and gas E&P market has huge amounts of interpreted data available, and even considering the many data types, scales and formats we handle, it is a rich source of information for Machine Learning.
In all industries, some domain experts are skeptical about Machine Learning and about the risk of relying on a “black box”. Presented this way, it sounds scary. If we dig a little into history, however, we realize that neural networks, in addition to manipulating simple mathematical objects, are the products of hundreds of years of research by the some of the world’s most brilliant scientists. They rely on strong but simple mathematical foundations. As technology is evolving very fast, and as we are faced with the Big Data challenge, it becomes critical for us as geoscientists, to apprehend and understand these simple techniques. The G&G community has a deep understanding of the data required to conduct successful field studies, and machine learning can help extract hidden information from that data. This combination is the ideal solution for conducting successful predictive analytics.
Machine Learning has brought additional insights to the oil industry when deterministic methods cannot easily describe geology from geophysical data. The determination of facies distribution, with well or seismic data, is a great example of the ability of Machine Learning to efficiently transform data into geological information. DNNA (Democratic Neural Network Association) has successfully been used to estimate the probability of rock-type distribution from prestack and poststack data in various geological contexts. The DNNA approach has shaped information in such a way as to make it easier for interpreters and reservoir modelers to make decisions. In such areas as reservoir connectivity stemming from laminated sands, or prediction of heterogeneous diagenetic effects on limestone, the use of DNNA Machine Learning has brought new insights to geoscientists.
In recent years, a new class of Machine Learning techniques which can handle Big Data has emerged: Deep Learning. Deep Learning regroups several families of neural networks, for example, Convolutional Neural Networks. These are very smart, but conceptually simple, evolutions of conventional neural networks. To handle Big Data, they apply the old “divide and rule” precept.
Deep Learning has been very successful in many domains, such as image detection, speech recognition and even trading. Handling speech is signal treatment: this is what we do with seismic data. Image recognition is visual investigation: this is what we do for geological, structural and stratigraphic interpretation. Deep Learning is opening a door, and may create the biggest revolution ever, in our methods for interpreting subsurface data.
Big Data and Deep Learning are also about receiving data in real time, and using the data flow to make better decisions. For example, we will soon be able to use information arriving from rigs and logs acquired while drilling, to update models to adapt quickly to any situation, and analyze fluid distribution in reservoirs in real time. The monitoring and analysis of real time data will enable the detection of abnormal situations at any early stage. Deep Learning, through the analysis of all available data, will enable us to better anticipate changes and adapt our strategy as needed. Downstream, the impact of Deep Learning (for example, the live monitoring of sensors in refineries based on real time analysis of the data), will set a new standard in terms of cost management efficiency.
We have a huge list of use cases for applying Machine Learning in the oil and gas industry, and the list is growing fast.
When Artificial Intelligence is trained, it performs tasks much faster than a human can.
- Many essential but highly repetitive tasks do not need human expertise
- Some processes may need permanent monitoring to limit risk and money loss
These can now be streamlined so that the expert can re-focus on the science and on team collaboration.
At last, the machine will help humans.