Design and Implementation of a Digital Twin for Live Petroleum Production Optimization
A single simulation generates 63 MB of uncompressed data and 467 data points ( approximately 4TB of data per day ). However , our inverse modeling process requires only 40 of those data points and we use compression to store the portion of the results needed . A single instance of the simulator can process about 15K cases a day ( approximately 10 cases a minute ). We use 4 instances to process 60K cases a day and this process can be scaled up to more instances if needed .
CONCLUSION
The typical processes of the oil and gas industry with respect to data processing , simulation for well modeling and artificial lift set point optimization are time-intensive due to their manual nature . With the limitations of number of engineers per well 12 , and with high decline rates 13 contributing to highly transient behavior , continuous optimization is a challenge . Inability to update set-points along with changes in well behavior may result in sub-optimal production rates . There may be significant economic benefit by optimizing set-points through increase in production , and / or reduction in operational costs 14 .
By harnessing live data feed , on-cloud processing power in combination with simulation and data science tools , it is possible to develop a digital twin for scalable set-point optimization based on physics-based models on fields with hundreds of wells . In the digital twin , there is an interactive system between the field data from the physical world and the virtual data from the simulations . An overall framework for developing such a digital twin has been presented in this paper .
The design and implementation details along with the architecture of the system required to automate continuous field data processing and a massive scale simulation engine that can generate 60,000 + simulations per day has been described . The benefits and challenges in minimizing human involvement for automating key components of the system have been addressed , while also highlighting the components that benefit from a human-in-the-loop . The on-cloud solution also provides an opportunity to scale-up the capacity on an as needed basis by multiplying the computational units .
14
Redden , J . David , Sherman , T . A . Glen , and Jack R . Blann . " Optimizing Gas-Lift Systems ." Paper presented at the Fall Meeting of the Society of Petroleum Engineers of AIME , Houston , Texas , October 1974 . doi :
https :// doi . org / 10.2118 / 5150-MS
- 54 - March 2021