Finding more energy-efficient and lower carbon emitting ways to develop heavy oil reservoirs is key for economically unlocking more than 50% of the remaining oil reserves worldwide. Continuous steam injection (CSI) is one of the most widely used yet energy-intensive heavy oil enhanced oil recovery techniques. One of the proposed methods to reduce the carbon footprint of heavy oil developments is using cyclic steam development (CSD) rather than continuous steam injection where applicable.
The study has two objectives. First, we developed a CSD conceptual model which describes the physical principles, represents the full life cycle of a CSD and focuses on key elements. Second, a thermal simulation model (based on the conceptual model) for CSD was calibrated to field data. Using thismodel, we conducted a parametric study to select the most impactful input parameters for machine learning model (MLM) training. We introduced a point-based regression method to train the MLM at every single time step for time series prediction of selected properties such as oil production, reservoir pressure, and so on. To manage large amounts of trained MLMs, we also conducted dimension reduction on prediction matrix using Principal Component Analysis (PCA). Once the MLM was trained with the field and simulated data, a web-based fast predictive tool was developed to provide consistent and robust predictions within seconds. This makes it possible to quickly evaluate various scenarios, assess uncertainties and optimize injection job parameters. Furthermore, the tool generates type-curves to identify the inflection point where increasing injected steam volume (per cycle) will not lead to improved recovery due to longer injection period or down-time for CSD wells. The presented workflow can also be widely applied to different recovery techniques, and asset classes.