InterPore2022

Data-driven production optimization utilizing multi-objective particle swarm algorithm based on ensemble-learning proxy model

1 Jun 2022, 11:30

15m

Oral Presentation (MS15) Machine Learning and Big Data in Porous Media MS15

Speaker

Shuyi Du (University of Science and Technology Beijing)

Description

Production optimization plays an extremely significant role in closed-loop management of reservoirs, affecting the sustainability and profitability of reservoir development directly. Due to the uncertainty of geological structure and the complexity of multiphase flow, traditional physics-based numerical simulator methods tend to suffer from insufficient calculation accuracy and excessive time-consumption. This research establish an ensemble proxy-model-assisted optimization framework in a data-driven approach, combined Random Forest with Bayesian algorithm and multi-objective particle swarm optimization algorithm innovatively. It can optimize production quickly and effectively under the premise of ensuring safety as much as possible. After experimental testing, the ensemble proxy model of the injection-production system based on the BRF algorithm shows better performance in the prediction of dynamic parameters, which can replace the traditional numerical simulator. Compared with deep learning, the proxy model not only has higher prediction accuracy, but the time required for training is only 1/9 of the deep learning. In addition, relying on the ensemble proxy model, the injection mode adjusted by the multi-objective particle swarm optimization algorithm can reduce the gas-oil ratio and increase the oil production by more than 10% for carbonate reservoirs. Meanwhile, Pareto Frontier analysis can provide more options for project decision-makers to balance oil production and gas-oil ratio considering physical and operational constraints.

Presentation materials

Data-driven production optimization utilizing multi-objective particle swarm algorithm based on ensemble-learning proxy model.pptx