Data-Driven Models for Predicting SAGD Production

Ziteng Huang
2022-group2-3

Abstract

This work shows that application of production performance data alone is not sufficient to train a data-driven model while the combination of both SAGD production performance indicators and relevant operational conditions can help a data-driven model better learn the SAGD process to achieve a better predictive ability. The results of the third test suggest that the GRU model can predict the SAGD performance based on real field data better than most of the other data-driven models studied. The key to its success lies with the proper preprocessing of raw information to filter out the noises but preserve the true trend exhibited by the actual measurements. Although the data-driven models can effectively predict the SAGD performance, their ability to predict the future is critically dependent on the data availability, and they are unable to make proper predictions beyond the production stages when the measurements had been taken. This could be a major limitation of the data-driven models.