Data-Driven Models for Predicting Production Periods in the SAGD Process

Ziteng Huang


This work shows that the GBDT algorithms are a better choice to predict the Start date while the ANN algorithms have better prediction performances on predicting the End date and Duration values. Utilizing a machine-learning algorithm to directly predict the Start date value is better than utilizing two algorithms to predict the End date and Duration values and then calculating the Start date value. But using an ANN model to predict the End date and the Start date and then calculating the Duration value are better than directly predicting the Duration values. A combination of different machine learning algorithms did not show a higher prediction accuracy than a single machine learning algorithm. Utilizing data-driven models takes less computational time than numerical simulation in predicting SAGD production periods.