Efficient steam injection management is vital for successful operation of large-scale steam flood fields, such as the "D" field, which consists of more than 6000 wells located in 13 distinct areas. Maintaining steam injection pressure within the target range of 500-600 psi in both injection areas and Gathering Stations (GS) is a significant challenge. Steam supply originates from multiple sources scattered around the field that need to be managed to anticipate decreasing pressure due to operational problems, for example, equipment maintenance or fuel gas shortage. Steam injection pressure exhibits various behaviors across different areas because of several factors such as pipe configuration and choke sizing in injector wells.

This project aims to develop a Machine Learning (ML) model to simulate pressure behaviors in both steam injector areas and GS during any steam rate changes in each steam generator unit. Applying Machine Learning to modelling pressure in interconnected steam distribution network is a novel approach provided by the project. The utilized model is Nonlinear Autoregressive with eXogenous inputs (NARX), which incorporates historical time series data from both endogenous and exogenous inputs. Steam contribution factor for each source was also determined to augment information to the model about implicit factors that might impact the pressure at each area. To test the model’s robustness against actual data, a moving cross validation dataset was implemented. In addition, alignment with the physical sense was also tested by conducting sensitivity analysis of pressure against controllable parameters such as steam rate and equivalent opening choke area. The model outperformed the naïve forecast as a baseline performance. Furthermore, a dedicated web-based dashboard is constructed, providing engineers and operation teams with a user-friendly graphical interface to facilitate monitoring and decision-making processes. Lastly, the model has been used to improve steam rate management by avoiding oil loss due to decreasing producer well performance that equivalent to $250,000.

You can access this article if you purchase or spend a download.