A brand new Chevron-led work move is permitting the oil firm to marry each natural discipline information with physics-based simulation fashions and machine-learning strategies to reach at a extra correct prediction of properly efficiency and, in the end, a dependable manufacturing forecast for unconventional oil fields.
Standard manufacturing forecast strategies for unconventional asset improvement rely totally on discipline information, which might endure from limitations in each information high quality and amount. Interpreting subsurface dynamics instantly from discipline observations can be a problem. Popular strategies corresponding to decline curve evaluation may be hampered by restricted information samples and too many variables. Reservoir simulation relies upon totally on discovering a superb historical past match for the present discipline, however this methodology is resource-intensive and requires sure experience.
“Everything starts with the data,” stated Kainan Wang, machine-learning scientist at Chevron, talking on the Energy in Data convention earlier this month. “It is the key to successful machine learning.”
Data points can come up usually in unconventional reservoir improvement. Missing information can turn into a problem if there are usually not sufficient wells in a sure space or particular completion design. Limited time on manufacturing additionally could be a issue. Noisy information points often are encountered when the measurements are advanced and error inclined or too costly to accumulate.
The Chevron work move augments the info in-hand with reservoir simulation. The simulation creates “synthetic data” to be able to fill within the blanks in lacking information instances, or tighten the band within the case of noisy information.
“This idea in the machine-learning world is not new,” Wang stated. “You generate synthetic data and put it side by side with the real data and analyze with machine-learning tools the mixture of data sets. In autonomous vehicles, you have simulators for the machine-learning algorithms to learn all different types of scenarios, especially the most dangerous ones. In object detection, you can rotate objects or morph them into different shapes.”
In the oil and fuel world, seismic interpretation has been one self-discipline that has seen some early stage success with artificial information—utilizing artificial seismic volumes with an artificial fault and horizon system and coaching machine-learning algorithms to be plugged into the true information set for fault and horizon detection.
“That’s what we are doing here, but using reservoir simulators instead, and to generate production data,” Wang added.
Combining a physics-based mannequin with machine studying, assuming the machine-learning mannequin is skilled properly, may end up in extraordinarily quick processes on sure purposes, corresponding to predicting volumes of future wells based mostly on present discipline information.
“One of the key components for a successful data-science project is for the model to be interpretable,” Wang stated. “The scientists want to be able to explain the prediction, why the model is saying the well is performing better or worse than the reference wells. One of the biggest advantages of this work flow is, by connecting to the subsurface or Earth models through simulation, we are able to explain why.”